IF [l SWP623 II Demography and Poverty Michael Lipton WORLD' BANK STAFF WORKING PAPERS Number 623 WORLD BANK STAFF WORKING PAPERS Number 623 Demography and Poverty Michael Lipton INTERNATIONAL MONEETARY FUND JOINT LIBRERY AR 2 . 1984 INTrAPTITONAUL 3R)iiVE FOR rlEO:4.iT'RCIT1'ON AH DE DELOPM:ENT %1V1iS?:GTC"1. ,, ?. . "@'il The World Bank Washington, D.C., U.S.A. Copyright e 1983 The International Bank for Reconstruction and Development / THE WORLD BANK 1818 H Street, N.W. Washington, D.C. 20433, U.S.A. First printing November 1983 All rights reserved Manufactured in the United States of America This is a working document published informally by the World Bank. To present the results of research with the least possible delay, the typescript has not been prepared in accordance with the procedures appropriate to formal printed texts, and the World Bank accepts no responsibility for errors. The publication is supplied at a token charge to defray part of the cost of manufacture and distribution. The views and interpretations in this document are those of the author(s) and should not be attributed to the World Bank, to its affiliated organizations, or to any individual acting on their behalf. Any maps used have been prepared solely for the convenience of the readers; the denominations used and the boundaries shown do not imply, on the part of the World Bank and its affiliates, any judgment on the legal status of any territory or any endorsement or acceptance of such boundaries. The full range of World Bank publications is described in the Catalog of World Bank Publications; the continuing research program of the Bank is outlined in World Bank Research Program: Abstracts of Current Studies. Both booklets are updated annually; the most recent edition of each is available without charge from the Publications Sales Unit of the Bank in Washington or from the European Office of the Bank, 66, avenue d'Iena, 75116 Paris, France. Michael Lipton is professorial fellow in economics in the Institute of Development Studies, University of Sussex, and a consultant to the Country Programs Department of the World Bank. Library of Congress Cataloging in Publication Data Lipton, Michael. Demography and poverty. (World Bank staff working papers ; no. 623) Bibliography: p. 1. Poor--Developing countries. 2. Developing countries--Population--Economic aspects. 3. Family size--Developing countries. I. Title. II. Series. HC79.P6L548 1983 305.5'69'091724 83-21816 ISBN 0-8213-0286-8 ABSTRACT This paper uses national samples and village studies to examine the demographic characteristics of households at high risk of poverty or ultra-poverty. Before industrialization in now-developed countries, such households were likely to be smaller than others; normally, the opposite is true in low-income countries now. Yet status (job, asset-holding, etc.) continues to be positively linked to household size. In Section II, these facts are traced to differences, over space and time, in marriage-age, fertility, family/household ratio, and family complexity, and to a lesser extent to differential mortality, migration, and life-cycle effects. Section III turns from household size to household structure; poverty risk in today's LICs usually shows strong positive links with high child/adult ratios, weak negative links with age of household head, and very weak links with sex-composition. Scale-economies to household consumption are little researched, but may also affect poverty risk. Many relationships in Secs. II and III weaken, or vanish, among the very poorest. Moreover, inferences to time-series (e.g. about what urbanization does to household size) are more than usually complicated. However, the relationships support strong implications for poverty measurement (Sec. IV), though much more tentative suggestions for policy. Outlay per household is an almost useless poverty measure; outlay per person is almost as good as outlay per consumer unit for broad allocative purposes, though not for "diagnosing" the poverty problems of particular households. Demographic assay, to assess how major policies affect big households with many children, could improve the poverty impact of many types of government action. ACKNOWLEDGEMENTS I am grateful, for discussion and comment at earlier stages, to Hans Binswanger, Nancy Birdsall, Mead Cain, John Caldwell, Tony Churchill, Tim Dyson, Norman Hicks, Dale Hill, William McGreevey, Paul Isenman, Mark Leiserson, Deepak Mazumdar, Graham Pyatt, Alex Shakow and Inderjit Singh; to Paul Schultz and other members of his seminar at the Economic Growth Center, Yale; and to Bevan Waide and other members of an informal World Bank discussion group arranged by him. Responsibility for the contents of this paper remains entirely my own. TABLE OF CONTENTS Page No. Abstract il Acknowledgements ii Table of Contents ii Characteristics of Poor and Poorest: General Introduction to Working Papers 1 (a) Origins 1 (b) Data sources 1 (c) Discontinuities 2 (d) Causality 2 (e) Policy for a non-underclass 3 I. Introduction 5 II. Poverty and Household Size 7 (a) Historically, poor households tended to be small 7 (b) Current and recent evidence: big households tend to be poor 8 (c) High-status groups still tend to have bigger households 11 (d) Poverty, status and household size: possible explanations 14 (e) Mortality, poverty and mean household size (MHS) 15 (f) Fertility and the poor 18 1. Direct impact on household size 18 2. Links between fertility and mortality 22 (g) Age of marriage and duration of unions 24 (h) Complex households: larger, but rare, especially for poor and low-status 27 (i) Migration: the relationship to differences in MHS 33 (j) Family cycles and MHS 35 (k) Non-family members, MHS and poverty 37 (1) Poverty, status and family size: some pointers 39 III. Poverty and household composition 42 (a) Composition, MHS, and poverty definitions 42 (b) Age-composition, dependency, and the risk of poverty 43 (a) Are women poorer? 48 (d) Life-cycles, poverty, and household composition 53 IV. Demography and poverty measurement 58 (a) The choice-of-denominator problem 58 (b) Choice of denominator makes a big difference 58 (a) The case for and against the three measures 61 (d) Economies of scale in consumption 66 V. Demography, poverty and policy 69 (a) Policy for research, measurement and project planning 69 (b) Steering resources towards poverty-prone, but given, household demographies 73 (c) Non-demographic behavior of big, poor households 75 (d) Policies to change the demographies of the poor 77 1. Fertility 77 2. Mortality 79 3. Complexity 80 4. Migration 80 5. Non-family household members 81 (e) Poverty, policy, and the family cycle 82 Abbreviations 84 Tables 85 Footnotes 102 Bibliography 111 CHARACTERISTICS OF POOR AND POOREST: GENERAL INTRODUCTION TO WORKING PAPERS (a) Origins In 1982, a Bank-wide Task Force reported on the impact of Bank activities on poor people. It showed that the proportion of the Bank's lending directed mainly at people in absolute or relative poverty had risen sharply - from about 5% in 1968 to 30% in 1980. Moreover, such activities showed rates of return at least as good as conventional lending, and succeeded, as intended, in benefiting mainly the poor. However, "neither borrowers nor lenders have been very effective in benefiting people who lack productive assets - the poorest 20%" (World Bank, 1983, pp. ii, 3, 5, 6-7). The report stressed the need for increasing the salience of poverty reduction in Bank "policy dialogue" with developing countries. But what can they learn from each other about appropriate policies for the poorest 20%? While working with Alexander Shakow and Norman Hicks on the secretariat of the Task Force, the author of the present paper was examining the characteristics of the poor and the poorest. This search was given special urgency by the Bank's partial success in raising the productivity of the poor, and its relative failure to do the same for the poorest. It may be, of course, that the "power structure" somehow prevents the poorest quintile of households in low-income countries - or the poorest decile in middle-income countries - from sharing the fruits of growth, while allowing moderately poor people to do so. Before accepting such a complicated hypothesis, however, we should look at the alternative: that very poor people (unlike the moderately poor) have characteristics that affect their capacity to benefit from development programs. This is one of four linked Working Papers that aim to identify such characteristics, if any. Three in the series - dealing with nutrition, labor and demography - will appear early in 1983, and the fourth, on assets, later that year. (b) Data sources It had been hoped to draw mainly on data sets for two ecologically oomparable poor regions, one Asian and one African, each with micro-information from good village studies supported by regional data from larger sample surveys. Partly to follow data availability, partly to ensure that climatic fluotuations would permit study of variability as a poverty problem, we selected semi-arid areas in N.W. India and N. Nigeria. In N.W. India, main emphasis was to be placed on Rajasthan, Gujarat and Maharashtra; the first two States permit the use of village studies by the Agro-economic Research Centre at Vallabh Vidyanagar (e.g Brahmbhatt, 1977; R.Patel, 1964, 1964a; V. Patel, 1973), and the last two of Pravin Visaria's disaggregations of National Sample Survey data by household outlay-per-person deciles (Visaria, 1977, 1978, 1980, 1980a). In N. Nigeria, outstanding work at the Ahmadu Bello University (Department of Agricultural Economics) has produced three good surveys, each covering three villages (Simmons, 1976; Norman, 1976; Matlon, oited in Norman et al., 1981). 2 As work progressed, however, it proved essential to support hypotheses from these sources with other enquiries, urban and rural, from a wide range of LDCs. (o) Disoontinuities Not because this was originally expected or planned, but because of the accumulating evidence, the method of these Working Papers came inoreasingly to involve a search, not simply for relationships between poverty and other characteristics (e.g. participation rates, caloric inadequaoy or family size), but for discontinuities in these relationships. It transpired that in LICs such discontinuities usually oocured, not at the "poverty line" (i.e., not between the "poor" and the rest), but at a much lower level of income or outlay, per person or per consumer unit: the level of ultra-poverty. This is consistent with discontinuities observed in experimental work on producers' behavior. Notably, farmers exhibit "threshold" changes in behavior (in respect of risk-aversion, reluctance to innovate, and - given the technology - reversal of the usual inverse relationship between size of operated farm and yield-per-acre) not around the poverty-line, but around the muob lower level of welfare at which subsistence appears to be endangered. The disoontinuities do not normally take the form of sudden, sharp rises or falls - as income or outlay, per person or per consumer unit, increases - in the proportion of persons within a given income or outlay interval. Rather there are reversals or intensifioations - i.e., respectively, turning-points or points of inflexion - in behavior, as welfare changes around levels of great poverty. It is well known that per-person inoome and outlay are usually distributed more or less lognormally. However, as these crude "welfare" indicators fall, adult female workforce participation rates increase until a "welfare" level signifiying extreme poverty is reaohed - and then decrease with further falls in welfare. Ratios of food spending to total outlay, around much the same point on the "welfare" scale, shift from steady rises as poverty increases, to a more or less constant 80-85% level. Unemployment rates, a steadily increasing function of poverty, increase more sharply at very low income.levels, and become more seasonally unstable. (d) Causality On suoh issues, and generally, these Working Papers try to remain agnostio about causality. For example, we find that income-per-person (and other welfare indicators) tend to increase fairly steadily as household size falls, yet paradoxioally also tend to be higher among groups of households with higher wealth or status. In seeking to reconcile these two findings we try not to make our explanations dependent upon whether large family size is cause or effect of (i) higher status or asset ownership, and/or (ii) smaller welfare at a given status or wealth level. Perhaps larger households get poorer, as Malthus posited; perhaps poorer households deoide, or are driven, to get larger, as is averred by economio demographers of both Marxist and neo-classical persuasions. 3 The soarce and soattered data, the extreme rarity of time-series, and our ignoranoe of how very poor people reaoh deoisions all suggest that one should defer attempts to make strong eausal statements. These papers "explore the space" relating poverty to, say, eoonomic and demographic charaeteristics. We must discover the direction and strength of the relationships, the gaps in observations on variables, and the turning-points or points of inflexion. Only then can we make sensible claims about causal directions. The author is too interested in policy inferences (and too incautious) to abjure all causal hypothesizing. However, the preferred form of oonclusion should uaually, at this stage, be neutral between "A causes B"p "B causes A", and "C oauses A and B". These papers are a first shot at outlining the shape, under different circumstances, of functional relations between poverty-variables (A) and characteristics-variables (B). Causal speoification is largely left for others. (e) Policy for a non-underclass What, if any, policy conclusions can be drawn from causally inexplicit relationships? The answer depends on the nature, alterability, and costs of remedying the characteristics associated with moderate and extreme poverty. In particular, is remedying those characteristics likely to enable an affected person significantly to raise his or her outlay or income in a self-sustained way? Or is such improvement, instead, the likeliest way to remedy them? The answer to that question is logically independent of whether the characteristics "caused" the poverty level in the first place, or were caused by it. Hence we may be able to identify important policy implications of the "characteristios-poverty-ultrapoverty" links - implioations independent of the causality of these links. One such implication is central to our whole enquiry. We find that the ultrs-poor have very different behavioral characteristics from other poor (and nonpoor) people. Do these characteristics mean that most of the ultra-poor belong to an "underclass"? It has been argued that the poorest 5-15 percent of people in developed countries have "underclass" characteristics rendering it impossible, or prohibitively costly, to enable them to raise their income and productivity in a self-sustained way. These characteristios may be linked to misfortune (e.g. persistent mental deficiency in persons recently released from hospitals), to earlier choice (e.g. alcoholism), or even to demographic circumstance (e.g. widowed and childless status). Whatever the moral issues, and whether the characteristics cause poverty or are caused by it, the result is the same: the "underclass" cannot at reasonable cost be helped to help itself out of poverty, but must rely on social-security payments or on charity. The evidence of these Working Papers strongly indicates that the great majority of the ultra-poor in LDCs are not, in this sense, an unreachable "underclass". Their extreme poverty is associated with lack of promising human and physioal assets; with weak labor-market positions; with large familiea, high dependenoy ratios, and very high infant mortality; and with significant risks of nutritional damage. Only tiny proportions of the Third World's ultra-poor oould survive as drug addits, aleoholiosp mental defectives, or even single-member families. These ultra-poor are mostly a resource, not a burdensome underclass. This raises a second policy issue, also largely independent of the causal links between characteristics and ultra-poverty. Does an ultra-poor group require different projects and policies, to achieve self-sustained improvements in income and productivity, from those required by the moderately poor? These working papers suggest that "food and health first", especially for children - and policies to improve capacity to contest labor and asset markets - may be necessary preconditions for improved productivity for the ultra-poor. Otherwise, benefits from "poor people's projects" will continue to stop at the second quintile. 5 DEMOGRAPHY AND POVERTY I. INTRODUCTION Within a country, what demographic characteristics tend to be found in households 1/ with a high risk of poverty or ultra-poverty? 2/ To ask this question in a policy context is really to ask four related questions. This paper tackles them separately, to simplify the exposition. However, it may be helpful to say a little about them (and their interconnection) at the beginning. The four questions concern: relationships between incidence of poverty and (1) household sizet (2) household structure; (3) appropriate denominator in measuring adequacy of household "resources"; 3/ and (4) implications for policies against poverty. The first question, treated in Section II, is: what relationships exist between the size of a household (number of persons) and the likelihood that it will be poor or ultra-poor? "Household size" comprises the family's adults (who usually work and earn); its children (who usually consume but do not earn); and non-family servants and lodgers (both usually absent in poor households). Household size is also affected by distinct, but interrelated, demographic events - birth, household splitting through migration, household formation through marriage, death; such effects can be influenced by poverty (as when undernutrition increases infant mortality), or can themselves influence poverty (as when dependency burdens increase). Moreover, households may be unusually large because they are complex, i.e. not nuclear; 4/ because they are at a certain stage in the family cycle; or because they have experienced unusual patterns of demographic events. Each of these three "size-increasing" variables, for given values of the other two, has distinct relationships to the incidence of household poverty: negative in the case of complexity, positive but short-lived for "swollen" phases of the family cycle, positive and long-lasting if a given family has acquired unusually many children who have survived into, say, the age-group 3-8. In view of these intricacies, it is fortunate that a quite clear answer emerges to the first question (pp. 7-13). Section III turns to the question: given the number of persons in a household, how is the household's risk of poverty connected to its demographic structure by age and sex - including the age and sex of head of household, the ratio of dependents to persons of "working age", and the probable position of that structure in the family cycle? The demographic structure of poor households is linked to their size in three ways: static, dynamic and definitional. The static links appear, for example, in cross-sections that, in total populations of an area, show that smaller households are likelier to be female-headed, yet that female-headedness and largeness are each independently linked, positively and significantly, to the incidence of poverty. The dynamic links concern the different potential, of households with different age- and sex-structures, to alter future size and capacity to earn. The definitional links arise from the widespread wish to measure household size, not by number of persons, but in some sense by number of units of ",anfii namant_ 6 Section IV asks: should the poverty or ultra-poverty of a household, and thence the incidence of poverty in a group (e.g. female- headed households), be measured by reference to the inadequacy of "resources" per household, per person, or per unit of requirement? If all persons had the same needs, if household size were statistically independent of both resources and resources-per-person, and if there were neither economies nor diseconomies of scale in consumption, then per-household measures would suffice. Since bigger households clearly tend to have more earners and thus higher total "resources", however, low resources-per-household indicate an indeterminate mixture of household poverty and household smallness. This is only very slightly - and again indeterminately - offset by the tendency of smaller households to have higher needs-per-person (adult/child ratios); by the positive correlation between poverty and household size; 5/ and by (modest) economies of scale in consumption. Hence, as shown empirically in earlier Bank work (such as Datta and Meerman, 1980), income or outlay per household, while widely used to "measure poverty", is demonstrably, and usually completely, worthless for this purpose, because it misclassifies many large poor households as non-poor, and many small non-poor households as poor. Per-person measures are much better; somewhat better still, though costly and controversial, are indicators of resources per consumer unit, whether measured crudely by means of adult-equivalents or subtly by means of equivalence scales (Deaton and Muellbauer, 1983). Finally, Section V asks: what does the present state of knowledge about the links between poverty, household size, and household demographic structure - and the implications of that knowledge for identifying the incidence of poor households and hence persons - mean for policy? We shall seek only limited and modest answers, mainly because the links are not causally clear. For examplet if we do not know whether some households have a high incidence of poverty because they are big, or whether they are big because they are poor, then it is not clear which issue - say, assets for the poor, or incentives to keep families small - is best tackled first. 6/ However, a good deal can be said on the policy issues, without specifying the directions of causation, simply by virtue of being able to identify what sorts of "household demographies" are likely to be associated empirically with a high risk of poverty. Not only can we more easily identify the types of place or time in which a given project is likely to affect poor households, if we can find where or when households have demographic features linked with poverty. At least as important is the prospect of diagnosing the sorts of productive activity through which households, in various comon poverty-linked demographic circumstances, can escape from poverty. For example, consider the oase of expanded asset-ownership associated with extra rural informal-sector non-farm income. The type of assets and activities readily open to female-headed households with small children, to big male-headed households, and to old and single persons will be quite distinct. 7 II. POVERTY AND HOUSEHOLD SIZE (a) Historically, poor households tended to be small This section will argue that poverty is nowadays strongly linked to big household size. Two prior objections are (1) that too few people live in big households to account for much of a phenomenon as pervasive as LDC poverty, and (2) that historically poor households were smaller than better-off ones. How are such objeotions to be met? It used to be widely believed that, in now-developed countries (NDCs) before industrialization and in most developing societies still, households tended to be large, often complex. However, the Cambridge Group for History of Population and Social Structure (see especially Laslett and Wall (eds.), 1972) has shown that the nuclear household of 2 to 6 persons has, in most cultures at most times, predominated (except in some of the serf-based societies of Eastern Europe, where the seigneur applied pressures for peasant complexity: Czap, 1983, esp. p. 145). Between 1574 and 1821, in 100 English communities with relatively good data, mean household size (MHS), excluding institutions, averaged 4.8, and 73% of households were of size 2-6 (Laslett, 1972, pp. 133, 146). Similar data are reported from the pre-industrial history of almost all Western countries; even in exceptional cases such as Serbia, large complex households remain a minority (Laslett and Wall (eds.), 1972, passim; Wall et al. (eds.), 1983, passim). Comparable results are found in recent developing-country censuses, showing in the mid-1960s "no bona fide case of a national average household size larger than six" (Burch, 1972, p. 91). Historical sociology in developing countries, as with Shah's careful analysis of Gujarati census data from 1821 to 1951, suggests if anything even smaller households - and with even less tendency to dwindle over time - than prevailed in the West (A. Shah, 1968). Low mean household size (MHS) cuts the scope for downward variation - in either rich or poor households - and a fortiori for linkage between such variation and wealth or poverty. Moreover, though most NDC series show very gentle downtrends in MHS over several centuries, there is no tendency, as households or communities differentiate themselves upward in early industrialization, for MHS or complexity to fall sharply. If anything, there is a slight increase - around 1750 in England 7/ (Wachter and Laslett, 1978, p. 76; Wall, 1972, pp. 191-2), and in 1920-65 in Japan (Nakane, 1972, p. 531). As if to make more surprising the proposition that big households now tend to be poor, historical evidence suggests that such modest upward variation in household size as did exist in NDCs accompanied greater affluence, permitting a net inflow of servants, early marriage with many children, and/or complex family structures, within a relatively large family house. In sharp contrast to recent evidence from poor countries, child/adult ratios in Tuscany in 1427 appear to have risen sharply with household wealth (Klapisch, p. 274) - pulling up MHS with them, at least among the 10,000-odd Florentine families listed (ibid., p. 277). In 23 Polish villages in 1789, peasant affluence clearly meant higher child/adult ratios and hence bigger households (Kochanowicz, 1983, pp. 158-9). In Hirase, Japan, around 8 1818, again "the wealthier households...were...larger than average for this area" (Nakane, p. 528). In a hundred English communities from 1574 to 1821, laborers and paupers respectively showed MHSs of 4.5 and 4.0, as against the 4.8 averaged by all households (Laslett, 1972, pp. 133,154). Hajnal summarizes the conclusion of pre-industrial demography: "It is often said ... that the rich can maintain larger households than the poor. There is some truth in this proposition" (Hajnal, 1982, p. 454). We need to explain the possibility, and then to demonstrate the fact, that - despite generally small and nuclear households; and despite historical evidence that poorer families used to be smaller than others - there is today sufficient variation in household size, and in a direction sufficiently different from that of historical data, for LDC poverty to crowd heavily into big households. The fact does seem clear- cut in almost all LDCs, despite occasional counter-examples (e.g. Hull and Hull, 1976). However, in reconciling the new fact to the old evidence, we shall bear in mind a curious paradox, which, because it so sharply illuminates the demography of poverty, will be at the center of this paper. It is the paradox that in today's LDCs, although in total populations (in defiance of historical NDC experience) big families tend to be poor families, groups of households with relatively low status - the assetless, the female-headed, the landless, the remote, the low- caste - tend to be relatively small, just as the historical evidence suggests. 8/ (b) Current and recent evidence: big households tend to be poor Almost every recent study, at whatever level of disaggregation, for either a particular group or a total population, shows the incidence of poverty and MHS increasing together. Table 1, calculated from Visaria's presentation of decile-wise relationships for several rural and urban South Asian populations, shows very high rank correlations. For all-India data, three pooled cross-sections of 4118 rural households - for 1968-9, 1970-1 and 1971-2 - confirm that "the probability of a (household's] being poor increases with its size", both for total populations and for cultivating households and day-wage households sampled separately (Gaiha, 1982, pp. 24, 32). Table 2 gives more detail - from large urban and rural samples for 1973-4 - for three States in Western India; once again, MHS is significantly larger for the poorest (MEP below Rs. 34 per month) than for the poor (Rs. 34-43), and much larger for both than for less-poor groups. Table 2 also shows that higher child/adult ratios are responsible for much of the larger MHS among the poor, again in sharp contrast to historical data, but as we see in Section III confirmed by evidence from elsewhere. However, Table 2 confirms - indeed sharpens - the historical hints (p. 7) that urbanization does not help the poor to escape poverty by reducing family size; poor and ultra-poor households are substantially larger in Western Indian towns than in Western Indian villages, though the opposite is true of better-off households. Does this strong size-poverty link - established for several Asian countries, in some detail for India, and for particular States 9 within India - persist within the small comunity, or is it a so-called "ecological fallacy" (Dasgupta, 1977, pp. 90-92), in which better-off places show lower average household size than do poorer places, although within a typical comunity poorer households might be no bigger than the average for that place? This may be so. Seventeen of the village surveys, carried out by an outstanding Indian center for suoh studies between 1961 and 1975, in Gujarat and Rajasthan show average size of household and average income for the village surveyed (Table 3). A village's income per person, deflated by a rough-and-ready price indicator to allow for the differences in survey years, in fact shows a very weak (not significant even at 10%) negative link to the village's mean household size. i/ So we must look at studies of small localities to confirm that in today's LDCs it is poor households, and not just poor places, that show higher MHS. Direct local evidence for the Western Indian places, comparable with the regional evidence of Table 2, is unfortunately scarce, because most of the village surveys took place in the early 1960s, and at that time AERCs normally reported households grouped by total, instead of per-person, income or outlay. The few Western Indian studies giving an income-per-person scaling (V. Patel, 1980; B. Singh, 1981) do confirm, in total populations of a village, that poorer households have larger MHS. Such an association is also shown for individual villages or small groups of villages elsewhere in India (Parthasarathy and Rama Rao, 1973, Table 8). Within each of two villages in Bangladesh, too, "for the two most important food consumption items - food and vegetables - the [regression] evidence is very clear that larger families do worse" (Mahmud and McIntosh, 1980). All these data measure MHS and requirements in numbers of persons. Bigger households in LDCs nowadays tend to have lower income or outlay per person. But big households also tend to have higher child/adult ratios (Table 2 and pp. 43-4). Might this reduce requirements per household sufficiently to render big households no needier than small ones? Table 4 represents the outcome of a crude attempt to see if the size-poverty relationship among households also applies when size is measured in "consumer units", weighted as in Schofield (1979, p. 149) rather than in persons - and, incidentally, to provide further micro- evidence of the size-poverty relationship, and to see if it applies to Africa. A positive, localized link between poverty and household "size in CUs" appears in these Zaria villages; the regressions of Table 4 are supported by differences, significant at 5 percent, in MHS as between (i) the poorest 20 percent of households by income-per-CU (8.5 CUs), and (ii) the other 80 percent (6.1). This is confirmed in the three villages in another area of Northern Nigeria, Sokoto; the respective sizes were (i) 5.1 and (ii) 4.3. 10/ The positive size-poverty relationship in Nigeria also seems to apply to urban data (Rouis, 1980, p. 26), though the methodology is not entirely clear. The NSS rounds also suggest that households with many persons are in India relatively likely to reveal shortage of capacity to buy calories per CU (Rao, 1979, p. 117). This also suggests that the poverty-size relationship applies, even after the linkage of size to 10 high child/adult ratios (and heneo lower requirements) is allowed for. So does the greater vulnerability to illness, undernutrition and death revealed by older siblings and large households (Birdsall, 1977, p. 75; Lipton, 1983). Three cautionary notes should be added. First, some large households may alleviate poverty via scale-economies in consumption, though these are probably small for poor households in LDCs (Sec. IV(d)). Second, larger households tend to have more earners, and hence may fail to report a greater proportion of ineome or outlay than smaller households, especially if only one earner - normally the household head - is interviewed (Cain, pers. comm., 1982); conversely, however, the smaller and on balance better-off households are likelier to conceal affluence from enquirers, for fear of taxation. Third, the positive size-poverty link need not always be smooth. (i) In rural India, a big 1968 sample showed risk of being in the poorest household decile (or the poorest 30%) by income-per-person steadily rising as household size increased from 1 to 9, but it fell for very big households of 10 or more (Gaiha and Kazmi, 1982, Table 9, pp. 24, 30, 32), and the linear positive size-poverty link was accordingly braked by a negative link between household size squared and poverty. (ii) In three villages in Kano, N. Nigeria, AHS fell steadily as income-per-CU rose to the ninth decile, but size increased in the highest income-per-CU decile (Norman, 1981, p. 76, citing Matlon). (iii) Above all, there is strong Indian evidence (Table 12) that 1HS and poverty are unoorrelated for the poorest 5-15 per cent in a State. 11/ Despite occasional doubts and some apparent exceptions, the data at macro- and micro-level are clear enough to be fairly certain that, as a general rule, poverty and large household size within a locality go together. This is consistent with the new consensus that average households are and were, in the great majority of countries, small and nuclear; Table 2 shows even village average household sizes varying from 3.97 to 7.27 across seventeen places in two Indian States alone, and the further scope for intra-village (or intra-city) variation leaves plenty of room for substantial variations, between poor and non- poor households, in MHS. The effects of this are dramatic. One example (of many that could be given) comes from the sample of 4118 households in rural India in 1968. Outlay-per-person fell below a plausible "poverty line" in 31 pereent of 1- and 2-person households, but in 88 percent of households with 7 to 9 members (Gaiha and KAmmi, 1982, p. 18). The direction and statistical significance - though not the size - of such effects of household size upon the risk of falling below an income-per-person ("Actual income II") specification of the poverty-line appear to be stable and consistent, both over time and as between major occupational groups. Thus, for about 900 casual-labor households in rural India, a rise of one person in MHS was associated (at the means) with a .17 rise in the logarithm of the risk of poverty in 1968, and a .46 rise in 1970; both associations were significant at 5%. For about 2350 cultivator households, the respective figures were .0371 (10%) and .1654 (5%) (Gaiha, 1983, Tables 18 and 19) - the lower impact here being probably due to the association of more land with larger households (see fn. 8 above). In explaining this positive sioe-poverty link - a reversal of historioal experience - we need, however, also to confront the linkage Of high-status groups to large household size. Here, apparently inoonsistently, historioal experienoe persists still. (o) High-status groups still tend to have bigger households The historical evidenoe that high status accompanied large household size was first oonvincingly marshaled for the English data by Laslett (1965, 2nd. ed. 1971, pp. 26, 72). In Coodnestone-next-Wingham, Kent, in 1676, three households of "gentry" averaged 9.0 persons; 26 yeomen's households, 5.8; 9 tradesmen's, 3.9; 12 laborers', 3.2; and 12 "poor men's", 2.1. Similar relationships applied to 100 English parishes with relatively good data averaged over 1574-1821: gentry 6.65, yeomen 5.9, laborers 4.5, and so on, with samples sufficient to ensure that inter-group differences are significant at 5 percent. Similar relationships have sinoe been unoovered for many pre- industrial and early-developing sooieties. Among 4808 households of known occupation in Florenoe in 1427, nobles and lawyers (45 households) averaged 5.8 persons per household, major crafts (483) 5.5, minor orafts (841) 4.6, shopkeepers and servioes (1665) 4.3, farmers (24) 4.4, and unskilled laborers and workers (1750) 3.7 (Klapisch, 1972, p. 277). In the town of Romans in S.W. Franoe in 1586, the much higher propensity of households of higher social status to include both resident servants and extended kin (Ladurie, 1979, pp. 4-5) must have rendered them considerably larger than other households, since both groups apparently had similar and large numbers of children per household (ibid., p. 5). In Japan around 1640-1720, status also appears to have affected inter- village differences in household size, with prosperous comeroial and fishing villages showing significantly bigger household averages than poorer farming villages (Nakane, 1972, p. 521). On status-size links, historical evidence for pre-industrial and early developing societies is almost entirely supported by recent evidence. After extensive review of Asian and African micro-data, one author concludes: "In stratified societies, the upper status groups tend to have larger domestic units" (Goody, 1972, p. 122). Since, however, the evidenoe now reverses the finding - which seemed reasonable when Goody wrote - that nowadays "richer ... farmers live in larger groups than the average for Etheir] omunity" (ibid., p. 122), we need to examine the status-size link separately from the poverty-size link, for the two links now seem to pull household sizes in opposite directiona. Poorer people now tend to have bigger households. Yet so do higher-status groups, on several indioators of status. Let us first oonsider landholdilng. For 4118 rural households in India in 1968-70, landlessness affeoted 65 percent of one-person households, 36 percent of 2-6 person households, and 21 percent of larger households; average household size rose from 4.8 among households with below 1 ha. to 9.8 among households owning above 14.5 ha. (Gaiha and Kazmi, 1982, pp.22-3). In big State samples, the simple r between (i) the midpoint of the 12 range of holding size reported for each of seven household groups and (ii) that group's average household size was +.9486 in rural Gujarat and +.9499 in rural Maharashtra. The addition of landless households, to create an eighth pair of observations, hardly affects the results, which remain significant at 1% (calculated from Visaria, 1978, Table 1). Still in Western India (Table 3) but now at village level, laborers' households - i.e. those deriving most income from employee work and thus normally operating little land - were, in almost every village surveyed in 1961-75, a good deal smaller than "owner-operator" households, deriving most income from owned farmland. This is confirmed for ten Central Gujarat villages in 1969-70 (Singh, 1980, p. 278), and for the 1970s for eight villages in coastal Andhra (Parthasarathy and Rama Rao, 1973, p. A-119); and in three Kerala districts marital fertility, both in 1965-70 and in 1975-80, rose sharply and significantly as land owned rose (Zachariah and Kurup, 1982, Table 3). 12/ Landholding and household size were linked positively outside India - e.g. in a large rural sample (Peach and Januszi, 1979, Table D-1) and in two villages (Mahmud and McIntosh, 1980, p. 506), in Bangladesh; and in three villages in Sokoto, N. Nigeria (Norman et al., 1976, pp.32, 37; 47, 58; 71, 83). Suggestive evidence extends this positive link of household size to assets, beyond farmland. Larger households have been associated with a greater incidence of plough ownership (Katsina, N. Nigeria, 1967: Anthony and Johnston, 1968, p. 48a); of livestock (Corsica 1769-71: Dupaquier and Jadin, 1972, p. 295); and of total wealth (Florence 1427 - though completely assetless households were somewhat larger than those with few assets: Klapisch, 1972, p. 277). These relationships may be partly life-cyclical. In rural Botswana, increasing age of male household heads, and presumably average household size, accompany "rapid asset growth" (Kossoudji and Mueller, 1980, p. 22). Both in Karnataka and in parts of Nigeria, parental house space - itself a major asset by value - appears to affect the age at which adult children are pressed to form separate households (J. Caldwell, pers. comm., Dec. 1982). Occupational structure, like asset-holding, strongly links high status with large family size, in both historical and recent studies. The data from Florence in 1427 (p. 11) are echoed in recent work; for example, in the Central Provinces of Kenya in 1971-2, the size of rural laborer households averaged 4.88, and of cultivator households 6.07 (Collier and Lal, 1980, p. 40). Table 3 provides telling evidence that, in specific Western Indian villages around 1962-75, households depending for most of their income upon owner-cultivation, trade, transport, and the professions were significantly bigger than laborer households. Laborers and cultivators, respectively, averaged over 13 villages in Gujarat (at various dates between 1961 and 1973), numbered 4.8 and 6.6 persons per household; over 7 villages in Rajasthan the respective averages were 5.1 and 6.0 (Table 3). Both differences were significant at 1%, and were found in a large majority of individual villages. Singh (1981, p. 278) reports similar findings in ten Central Gujarat villages in 1969-70. The difference, like the ambivalent position of craft households, appears also to apply in Southern Karnataka (J. Caldwellp pers. comm., Dec. 1982). 13/ Generally we know too little about the relationship of household size and structure to oooupations. It is an important issue, affecting the viability for poor 13 households of alternative forms of informal-sector family enterprise. However, the correlation between high-status work and large households is clear. Western Indian caste structure, for twelve villages in the 1960s, provides a parallel link between bigger and high-caste households (Table 5). "Intermediate caste" households comprise mostly the dominant farmer castes, such as the Jats. They are substantially bigger than "other" (i.e. lower-caste), scheduled-caste, and non-Hindu households in the Gujarat villages, and somewhat bigger in Rajasthan. These groups comprise the overwhelming majority of Indians classified by religion or caste, and confirm the general "high status - big households" rule; interestingly, in these villages, a few of the highest-status groups (Brahmans, Banias, Jains) 14/ appear to be exceptions, showing relatively small households. This may be a typical phenomenon, though the exception for Brahmans is also claimed by Kolenda (A. Shah, 1973, p. 224); Oscar Lewis's classic North Indian study, for "Rampur" in 1953, shows the 15 Brahman households averaging 7.3 members, not far below the 78 Jat households (8.3) and well above the 38 scheduled-caste households (5.2) (cited in Goody, 1972, p. 113). A Tibetan village also shows a clear link between big households and high status (Carrasco, 1959, pp. 67-9). Female-headedness is also associated with low social and political status in most underdeveloped communities. For obvious reasons, female-headedness is also associated with small household size. This was true historically, e.g. in early-industrializing Rheims in 1802 (Fauve-Chamoux, 1983, p. 480). In today's LICs, Visaria (1980, p. 59). shows dramatic declines in the proportion of female heads as household size increased for ten large South Asian samples in 1968-74: for example, in rural Maharashtra (1972-3), females comprised 61 per cent of one-person, 6 per cent of five-person, and 2 percent of nine-person household heads; in Malaysia (1973) the corresponding figures were 35 percent, 15 percent, and 12 percent. In Botswana, the 1974-5 Rural Income Distribution Survey (1974-5) also showed that female-headed households were smaller, but only if no adult male was present: the 47% of households with a resident male head averaged 7.3 members; the 14% that were female-headed but contained a resident adult male, 7.6; and the 29% female-headed without a resident adult male, only 5.1 (Kossoudji and Mueller, 1980, p. 11). We can now state, quite sharply, a demographic paradox of poverty in today's LDCs. In total populations, larger households tend to be poorer (the opposite of most historical situations). Yet many status-scalings - by land owned or operated, by wealth, by type of work, by caste or other social status, by sex of household head - link low status, now as in the historical past, to smaller households. In analyzing this apparent contradiction, we shall come to understand much about the demographic characteristics of the poor. 14 (d) Poverty, status and household size: possible explanations It is useful to distinguish seven demographic variables, related to MRS, and eaoh possibly related also to poverty. Each variable may help or hinder our task of explaining why poverty, in LICs now but not in NDCs in the past, goes with large MIS; or why low asset and job status goes with small MS; or both. The first six household variables affect principally the size of the family, i.e. the numbers of coresident kin. Only the seventh variable explicitly deals with household members who are not family members. The first three variables are standard demographic indicators. First, mortality risks over a lifetime may vary with the level of poverty; usually this applies much more to infant, and to a lesser extent child, mortality than at later ages. Seoond, live births per year per cohabiting couple, given the oouple's ages, oan differ between poor and other persons or groups. Third, the duration of cohabitation 15/ can vary - can start or end at a different average age, oan be interrupted with different frequency, or can show different (usually small) probabilities of not happening at all. Fourth - if we turn to variables affecting demography indirectly, i.e. not by changing birth or death rates instantaneously - rich and poor persons or groups may show different propensities to form separate households when the ohildren marry, or (if they do not at once separate) different periods of complex family living. (Complex families, of this "multiple" nature or otherwise, 16/ are likely to imply larger households in three ways, as compared with nuclear families: directly, by retaining the child and adding a spouse; by increasing the proportion of fertile couples; and by changing the mortality risk of the household). Fifth, sets of households may differ in respect of in-migration or out-migration. Sixth, their household size may reflect, not lifelong associations with them of their members, but temporary passage through parts of a life-cycle. For example, laborers may tend to average smaller MES because they are "waiting" to inherit parents' land, to enlarge their households by having children, and to move from the status-group of "laborer" to that of "owner-cultivator". Our final variable allows for possible inequality between household and co-resident family. One set, A, of households may be more likely to include non-family members - espeoially resident servants or lodgers - than another set, B. This normally has a "double-geared" effect on MHS differences between the sets, for set B is then normally more likely to out household size, e.g. as adolesoents leave B to go into domestic service with set A. There is no reason why just one of these seven variables should help to explain either or both parts of our paradox. Nor need every variable even point in the right direotion (i.e. associate large ISHS with poverty in LICs now, affluence in NDC history, or high status). Indeed, both these things are very unlikely. Reality is more usually messy than elegant. Moreover, this partioular (sevenfold) classification of correlates of household size has no logioal primacy; it just happens to follow one pattern of frequently available recent data. Logioally, one could as well separate faotors causing household formation, household separation, and household size between formation and separation. Relevant, too, are the classifioations of successful historical "predictors" of mean household size for communities, 15 i.e. urban or rural plaoes: for 382 places in Suwa county, Japant in 1671-1870 (Hayami and Uohida, 1972, pp. 488-92), and 100 in England in 1574-1821 CLaslett, 1972, p. 155). The only comon successful "prediotors" were the proportions of households headed by married oouples (r 2 .600 on MRS for Japan, .296 for England) and with servants (.444 and .599). Vhiohever way one outs the oake, it is of major analytical and polioy importance which potential correlates turn out in fact to be associated with different household size. (e) Mortality, poverty and mean household size (IMS) Historical data, permitting inferenoes about group differenoes in age-specific mortality within NDCs during the pre-industrial or early-industrializing periods, are scarce and controversial. Howevert there are two prima facie reasons to believe that the ratios of poor or low-status households' death-rates to those of other households, for most age-groups, were substantially lower for NDCs, especially in the pre-industrial period, than in today's LDCs. First, many killing infeotions and other tranmissible diseases, from bubonic plague to malaria, are not so very different, in incidence or in virulence, as between a country's gentry and laborera, or rich and poor. It is above all such diseases whose incidence, and hence contribution to death-rates, has plummeted worldwide since the late nineteenth century. j1/ Second, knowledge about how to avoid other diseases, notably puerperal fever, dysentery and TB, has advanoed enormouslyt as between NDC pre-industrial history and today's LDCs. Thus better-off or better-educated people in poor countries have become able to greatly reduce their risks - e.g. by boiling drinking-water, by purohasing hygienic ohildbirth, or by ohoosing to live in a suburb with better sewerage. Certainly, at the crudest level of comparison, country cross-sections today reveal strong links between mortality and severe potential poverty as suggested by low real GNP per person. As a country's GNP per person, in oonstant 1963 prioes, "rose" to about $400-500, but not for further rises, there were big mortality "declines" both in the 19305 and - from a lower baset due to an improved health environment - in the 1960s (Preston, oited in Schultz, 1981, p. 117). Let us assume (until p. 18) that groups with death-rates above the national average do not offset this by also featuring higher fortility, greater household complexity, or other faotors tending to pull MRS up. In that case, high-mortality groups will also feature low MRS. Indeed, the younger the ages at which a group suffera abnormally high rates of death, the greater is the proportionate out in the group's MRS. In LDCs today, inter-group variation in mortality is muoh greater in early life. The relatively reliable Sample Registration Survey (SRS) showed that inter-State male infant mortality rates (CIIs) in 1972 ranged from 75 in Kerala to 210 in Uttar Pradesh (the respective female IMRs were 62 and 259); at age 5, male life expectanoy was 61.8 in Kerala and 55.9 in UP (female, 60.7 and 49.9) (Dyson, 1979, cited in Ruzicka, 1982, Table 8). A 1979 Indian survey covering some 750,000 SRS households found proportionate gaps, by caste, maternal education, and looal infrastructure, in IMRs well above the gaps in death-rates at ages 1-4 (Ruzicka, 1982, Table 9) and even these gaps were well above any that are plausible at later ages. African mortality 16 estimates are generally much less reliable, but again suggest that - with the important exception of West Africa - differences in infant and child mortality account for the great mass of differences in life expectation at birth, and that inter-country and probably inter-group differences in adult mortality are relatively very small (A. Hill, 1981, pp. 32-4, 79-84). To the extent that inter-household mortality differences are not compensated - or are less than fully compensated - by other differences affecting household size, the earliness of high and differential death risks in LDCs must mean that they greatly affect inter-group variation in MHS. But how do groups of households in LDCs vary in respect of death risks? First, one should refer to an unusual yet important effect: unusual in that it tends to cut MHS in poorer and lower-status households by raising their adult, as well as infant and child, mortality. African research "has universally found that mortality at all ages is substantially lower in urban areas than in rural areas" (A. Hill, 1981, p. 35). In India, there is "indisputably higher [age-specific] mortality of rural [persons) at every age-group compared to urban" (Mitra, 1978, p. 223); indeed, for under-fives, and less certainly for adult men, the gap substantially widened in the 1970s (Ruzicka, 1982, Tables 5-6). Poor households, and poor people, are in most regions heavily over-represented in rural areas, so that higher rural death-risks - assuming no, or incomplete, compensation via fertility or otherwise - would mean smaller rural households. This is a powerful effect, because (while also affecting adults) excess rural mortality still strikes infants hardest - IMRs, in two good Indian surveys in early the 1970s, were 136 and 137 in rural areas, but 90 and 92 in towns (Natrajan, n.d., p. 7). The effect of death-rate differences, in making rural households smaller than urban, should apply especially to poor people, who have much greater difficulty than the better-off in escaping from such risks by using modern public-health, medical, or nutritional assets or knowledge. The recency of many African conurbations, however, and the consequent large proportion of recent immigrants with incomplete or single households, means that African data are unsuitable to check this proposition. Indian data, though, do support it. It is not only that Indian urban households show MHS significantly, and probably increasingly, above rural levels; this does probably mean, among other things, that urban status raises MHS by reducing depletion by death, especially child death. More tellingly, as poverty worsens, Indian urban MHS rises much more steeply than rural - so that in some States well-off urban households are actually smaller than rural households with similar MEP, although poor urban households are much bigger than their equally poor rural counterparts (Table 2). This is largely due to the high child/adult ratio of the urban poor. Their children - and to a lesser extent their adults - are far likelier to survive in cities than in villages (the exact reverse of the position in industrialising England: McKeown, 1979, p. 76). In today's LICs, once towns do not comprise mainly recent immigrants, urban status tends to confer (i) relatively greater real MEP, given household size; (ii) for this and other reasons, lower death-risk, given real MDP; (iii) for poorer people, a large MHS (as more children, especially, survive), relative both to less-poor townspeople and to equally poor villagers. Poorer households are 17 bigger, in both town and country; but urban residence confers on the poor the capacity to avoid many child deaths that they would suffer in the villages, and thus makes their households relatively bigger still. A second possible "group link" of status to household size is via differential female mortality. Especially among girls, female excess mortality in some parts of India and Bangladesh (Miller, 1982; Chen et al., 1981) is undoubted. However, its causes are disputed, and "there is no trace in African data of ... marked systematic female disadvantage in mortality (Hill, 1981, p. 35). Even in the few parts of South Asia where households have been shown to discriminate against the survival chances of girls, it is unclear whether this discrimination especially cuts MHS in poorer or lower-status households (Chen et al., 1981; Levinson, cited in Carloni, 1981; Lipton, 1983) or in higher-status ones (Miller, 1982, and the evidence there cited). Hence the effect on the size-poverty relationship of India's higher rural female mortality in all age-groups from birth up to 34 years (Mitra, 1978, p. 223), accompanying a higher propensity among Indian rural females to be in smaller households, is not clear. Moreover, evidence from Western India, Sri Lanka, Nepal, Taiwan and Malaysia clearly shows that women are not over-represented in the lower deciles of households by MEP (Visaria, 1980, p. 60). So urban "status" lowers death-risk; usually raises settled MHS; and (at least in India) especially raises MHS among the poor. Female status (while seldom affecting death-risk much) sometimes raises it; is associated with membership of lower-MHS households with high proportions of females; but appears unrelated to household poverty, though certainly not to individual female disadvantage (Lipton, 1983a). What of gaps between groups more overtly linked with poverty or status? In 1978, scheduled castes showed IMRs of 159 in rural India, as against 136 for Hindus (but 108 for the generally poorer Muslims: Ruzicka, 1982, Table 9). In rural Karnataka (Mysore) in 1961, laborers and "temporary tenants" had IMRs over 67% higher than owner-cultivators farming over 3 acres or "permanent tenants" with over 3 acres. A recent analysis of child mortality in two States, based on the 1951 Census, shows that the proportion of children dying before the age of 5 was then 10% higher among farm laborers than among farm operators in Madhya Pradesh, and 42% higher (on a much lower base - of about half the rate) in Kerala (Mitra, 1978, p. 21; Natrajan, n.d., p. 12). This should be set against the evidence that only the nutritional status associated with extreme poverty - such as afflicts 10-15% of Indians (and at most 20% of children) - increases infant and child mortality (Chen, 1980, pp. 1836-45; Lipton, 1983, pp. 21-3). Thus these differences in rates of infant mortality must be associated with a very great overlap between the high-mortality groups and the extremely poor. In urban and rural Karnataka, moreover, housing-related indicators of economic levels - such as are likely to discriminate between poor, poorest and other - were linked strongly, in the expected direotion, to both infant and overall death-rates (Mitra, 1978, p. 210), and this also applied to a 1978 all-India survey for IMRs: in rural areas, living in a village with electricity, water-supply or medical faoilities was associated with IMRs lower by about 30 percent (Ruzicka, 1982, Table 9). An Indonesian study implicitly relates the large urban-rural gap in 18 survivorship ratios in 1972-3 to higher urban levels of maternal eduoation - itself largely a surrogate for income; holding the mother's education and income constant, we find that the urban-rural mortality gap vanishes (calculated from Hull and Hull, 1976, pp. 8, 15). In Afrioan studies, "mortality is also invariably lower among children of women with education" (A. Hill, 1981, P. 35). In India in 1978, rural IDR was 132 for children born to illiterate mothers, but 64 where mothers had completed primary oducation; the comparable urban rates were 81 and 49 (Rusicka, 1982, Table 9). Do these relationships help to explain the links between poverty and MRS? Higher infant and child mortality - and slightly higher subsequent mortality - among poor groups, and poor households in total populations, obviously outs MS1, relative to other households, direotly. Thus the relatively high IKR among laborers (and probably among the associated castes), rural women, villagers, the ill-housed, ete., does directly relate to their smaller MHS. It renders smaller households for low-status groups easier to explain - and for less-poor persons, harder. If indeed receding tides of general infection and ignorance have left, even more important than in the past, those life-threatening illnesses best avoided by adequate education or income (p. 15), then the growing direct effect of differential death-risk in reducing poor people's MHS poverty to low 141 (though it could explain why the very poorest households (fn. 11), where alone IIIR shoots up, are still no larger than the other poor). However, the direct effect is not the whole story. Higher mortality in poor households might "trigger" indirect responses that actually raised HBS relative to other households. This is very unlikely to operate via a positive effect of mortality upon household complexity. Indeed, since lower life expectancy means that parents of resident married children die sooner, it implies - under most norms of household formation - fewer complex (and thus normally even smaller) households for poor and/or low-status persons. However, simulations suggest that such effects, even from big differences in mortality, are very small (Wachter and Hammell, 1978, p. 48). More important, however, is a much more plausible way in which the higher mortality of low-status households can actually raise their MRS: by triggering sufficient higher fertility, via either longer cohabitation or higher fecundity (sections (f)-(g)). The apparently higher mortality, within each job and status group, at low income and asset levels may be outweighed by higher fertility - replacement fertility on insuranaee principles (Cassen, 1978, pp. 60-61), or due to earlier cessation of post-partum amenorrhoea as infants die. Thus higher mortality in poor households, paradoxically, could help explain why their MHS tends to be larger in total populations, and therefore a fortiori within most income or status-groups. (f) Fertility and the poor 1. Direct impaot on household size Households with higher fertility, unless it is outweighed by other factors, are bigger households. Japanese historial evicence suggests that, in the long run, it is not outweighed, though short-run localized links of fertility to family size are rather weak (Haysmi and Uohida, 1972, p. 496). In today's LDCs, where infant mortality has fallen, greater or lesser 19 responsive deolines in fertility could well be a min explanation of why household sizes differ. The number of ohildren born to a woman in her life is the outcome of her age of menarohe, her age of menopause, frequenoy of lnteroourse between those ages, proportion leading to conoeptions, and proportion of oonoeptions followed by oampleted pregnancy. Unfortunately, very few data sets analyse these variables separately. More usual is clasaifioation of rural or urban women, sometimes by age-groups, by (i) numbers of ohildren ever-born, dealt with in this section; and (ii) age of marriage (Section (g)). Both these olassifications are sometimes further broken down by income or outlay per person in the woman's household, by the household's apparent main oocupation or social group, or by the level of motherts or father's education (of course closely linked to inome-per-person). Another approach seeks to disoover whether the poorer, lower-statua, or otherwise "different" woan also differs in respeot of "intervening variables" affecting either children ever-born to her per year of sexually aotive union, or else proportion of her life spent in such union; auch variables are physical fecundity, marital disruption, and post-partum abstinence (Hull and Hull, 1976, p. 19, fn.; Davis and Blake, 1956). Finally, income or status-group might be related to such aspects of fertility through physioal determination, or through optimising choloes aa stressed by the "Yale school" (Schultz, 1981). The balance of evidence (Birdsall, 1974, pp. 5-79 and 1980, pp. 53-6; Hull and Hull, 1976, p. 9) suggests an "inverted-U" overall income-fertility relationship. This almost certainly holds across LDCa. National fertility probably rises as annual average income-per-person rises up to a low threshold (perhaps $125 in East Asia and $300 in Latin Amerloap in 1965 prioes), mainly because, below that threshold, some women's physioal fecundity is reduced by severe undernutrition (sometimes by famine) and by disease. However, once that threshold of poverty is surpassed, increasing inome-per-person - being associated with higher levels of female education, lower infant and child mortality, and (as schools replace ohild-labor) a rising opportunity-oost of ohildren - reduces fertility, though at a falling rate as a family approaohes levels of reasonably adequate living. Broadly, often strongly, supportive evidence for part or all of this relationship is available from cross-sections within LDCa, viz. Iran, India, Puerto Rico, Turkey, the Philippines and Taiwan (Birdsall, 1974, 1980). Karnataka in 1961 appeared exactly to oonfirm the "inverted U'. Bangalore, presumably the best-off city, showed a clear positive link between apparent fertility and likely poverty - the 7% of women in the worst housing showed a 25% higher orude birth-rate than the remaining 93%; in other town, no clear relationship prevailed; and in the relatively poor rural areas, very low socio-economic status acaompanied lower lifetime fertility. There, women over 15 in households deriving income mainly as "agricultural laborers and temporary tenants' averaged 6% fewer ohildren than "owner-oultivators with less than 3 acres and permanent tenants with less than 5 acres" and 12.5% fewer than larger cultivators - highly significant figures for such big amples - and sharedt respectivelyt 16% and 25% lower birth-rates per woman aged 15-44 (Mitrat 1978, pp. 209-10). A similar pattern prevailed in Colombia. Age-speolfi fertility is negatively assoolated with both husband's and wifets wage-rate in towns - but 20 positively (except for younger wives) in the (much poorer) rural areas (Schultz, 1981, p. 177). 18/ One tentative interpretation is this. Laborers - and presumably the associated social (e.g. caste) groups - suffer higher infant and child mortality than cultivators. While they sometimes compensate via higher fertility - marital fertility differentials by easte in Kerala (Zachariah and Kurup, 1982, Table 2) being an example - the very poorest often, far from over-insuring (p. 23), show lower fertility. Hence, in part, the inter-group link between low status and low household size. However, poorer people (though not the very poorest) within each job and status group not only show rather higher infant and child mortality rates, but "overcompensate" with higher replacement fertility, both because of shorter post-partum amenorrhoea due to higher IMRs, and to insure against them. Hence, in part, the intra-group (and aggregate) link between poverty and high household size. The weakness of this link for the very poorest (fn. 11) could be explained by their concentration in low-status groups. Work in the Punjab (Wyon and Gordon, 1971; Kielmann et al., 1978) appears to support this. Moreover, the data from Matar Taluka in rural Gujarat (1965-75) suggest that the intra-group relationship is strongest at the low end of the income scale - among the poorest, and some of the poor - and that their higher replacement fertility tends to inhibit asset formation and keep these poorest poor (Repetto and Deolalikar, n.d., p. 57). This higher fertility reflects (i) (perfectly rational) over-insurance against what are, for the very poorest who alone are at poverty-induced nutritional threat to infant life, genuinely higher risks, and (ii) the low net costs of small children in these groups (Cassen, 1978). Extreme poverty, at least, however, does also have links to factors that reduce fertility, and hence, other things being equal, family size. Severe undernutrition probably reduces fecundity, retards menarche (first menstruation), reduces fecundity between menarche and menopause, and accelerates menopause (Friach, 1978, 1980, 1982). Fertility may also increase directly with hygiene in ways especially discriminating against the poorest. Several fertility indicators in Kerala in 1965-80 were positively linked to house quality variables, especially to toilet facilities (Zachariah and Kurup, 1982, Table 3). Apart from such extreme physical fertility effects, choice by poorer couples occasionally also reduces fertility. They are likelier to be temporarily separated by the migration of just one partner. They also - as in a village in Central Java in 1972-3 - feature much greater voluntary post-partum abstinence, and more marital disruption. Nevertheless, in the same village, poorer couples show shorter interpregnancy intervals (Hull and Hull, pp. 26-9). If we judge by worldwide KAP survey evidence, the poor normally make less use of the more modern and effective methods of contraception. There is no doubt that the poor, but not always the poorest, show higher marital fertility in the great majority of LIC situations. One interpretation of the inverse-U relationship between fertility (y) and income per person (x) is suggested by the strong positive link between child wage-rates and couple fertility (Schultz, 1981, pp. 50-1). This link could well associate fertility with poverty, by leading poor couples (for whom 21 alone child labor is an important income source - Lipton, 1983a, Sec. II(d) - since others are likelier to prefer to postpone and enlarge income benefits by educating their children) to choose higher fertility. However, the correlation could also associate fertility among the poor with escape from extreme poverty, as children's earnings - net of child costs - raised household income-per-CU for the more fertile poor. Like the extreme physical effects, this may help to account for the inverse-U relationship. In the NW Indian villages, the very poorest are likeliest to be "isolated" from the rest of the population in a group with distinct job, status, and assets: the low-caste landless laborers. If this is generally the case, then the inverse-U shaped fertility-income function could be a "linkerw, helping to explain both legs of our central puzzle (p. 13). Over most of the range, fertility tends to fall as income-per-CU rises; but for, say, the poorest 10-20 percent, who tend to be in a separate status-group (especially at the very poorest "end"), the reverse relationship may apply. For the whole population, a positive fertility-poverty correlation would be found (as in the Kerala sample: Zachariah and Kurup, 1982, Table 2), because of the dominance of the "less-poor 80%" in the regressions; but, as between groups, the overlap of low-status and very poor households could connect very low status to somewhat lower fertility. This is consistent with the weakening of MHS-poverty relationships among the ultra-poor (fn. 11). How might parental choice link larger household size, via higher fertility, to (A) poorer families in LDCs now, (B) better-off families in the pre-industrial past of now-rich countries (Klapisch, 1972, p. 274; Laslett, 1972, pp. 153-4), (C) higher-status families, usually now, almost always in the past? The "Yale school" seeks to interpret data as if couples choose, at the same time, probable family size - partly by weighting expected couple income from children's work against costs of parental time spent in child care - and couple income and leisure. It is also argued that the effect of reduced poverty "on fertility depends on the source of the increment to income, and hence the 'price-of-time effect'", i.e. the opportunity-cost of extra child care, "embodied in that source". It is argued that, unless extra couple income is achieved in ways that would be inhibited by extra child care, it will not affect fertility and thus household size: "Would we expect fertility in ... Libya to fall because per capita incomes are among the highest in the world?" (Schultz, 1981, p. 5). 19/ This explanation of (A) above, by reference mainly to substitution of (high) earnings for child-care in better-off households, but not in worse-off ones - this virtual denial of income-effect - could acount for (A) alone. It could also handle (B) above, albeit tortuously, e.g. if it were argued that the greater prevalence of servants in the pre-industrial West drastically cut costs of marginal child-care for the rich, but not for the poor, as compared with LDCs today; better-off children's higher potential earnings could perhaps have weighed more heavily then (making richer couples relatively fertile), although their higher potential opportunity-costs do so now (making richer couples choose lower fertility). However, it is hard to expel income-effect from the voluntarist account of (A) above. Partly this is because of three sorts of evidence that income-effect induces women to divert time away from work - evidence that should apply just as well to the decision to have more children. (i) As income-per-CU rises, women withdraw from the workforce, though potential 22 earning presumably rise. (li) This Is also true as women transfer from village to town (Lipton, 1983a, Seo. 11(d)). (iii) And female education - which plainly raises potential earnings - does not significantly out age-specific maternal fertility if the education ends before primary oompletion, or at all for older rural women (Stycos et al., 1978, for Costa Rica; Schultz, 1981, pp. 120-1, for Colombia; Zaohariah and Kurup, Table 2, for marital fertility aged 20-40 in Kerala; Hull and Hull, p. 10, for Indonesia). 20/ Partly, to explain (A) above via parental response to the cost of child care (higher in riher households if, but only if, mothers could earn more than in poorer ones) - via pure substitution-effeot without income-effect - is implausible because of (C) above. Higher status, work-group, or access to earned assets must often be associated more convincingly with potential earnings (foregone in the event of child oare) than is higher income-per-CU. Yet such sooial differentiation is associated with bigger households, more fertility, and more child-care in the higher-status groups: the reverse of what substitution-effeot alone would induce. We do know that normal negative substitution-effect relates female wage-rates to female partioipation rates (e.g. Smith (ed.), 1980; Quizon and Evenson, 1982); hence, both in (C) above and in regard to the three types of evidence listed, we can be oonfident that effects other than substitution-effect are at work in the opposite direotion. Moreover, we may need to explain fertility differences among status, job, and asset olasses by sooial variables, not just by income and substitution effects. It Is quite plausible to reconoile (A), (B) and (C) above by hypothesizing that (A) is due to income and substitution effects together; that the price of child-eare, relative to the discounted potential earnings of ohildren, was much lower for the rich, as compared to the poor, in the pro-industrial West than in today's LDCs, explaining (B); but that status-groups generate habits, expectations and norms regarding fertility, explaining (C). All this is tentative, almost speculative, partly because data are soaroe. However, differential fertility of unions - and its links, biological or sooio-economio, to poverty, nutrition and health, and job-asset-status group - should prove oruoial in reoonoiling the apparent paradox on p. 13. 2. Links between fertilitr and mortality An with mortality, so with fertility: we have concentrated on its direct links with household size, as they might render poorer households bigger (or households with poverty-related status, job or asset positions smaller) than others. Indireot links might assooiate mortality or fertility with the size of such households either via the prooess of household formation and division, or via reciprocal effeots between fertility and mortality. It is hard to see how different household formation rules or timings oould greatly affeot mortality or fertility; and, though it is quite conceivable that ohanges in mortality or fertility could alter family size by changing the time during which a household is simple or complex (e.g. if high death-rates swiftly eliminate the patriarch ln a multiple family 21/), simulation shows that this does not make much difference for plausible values (Wachter and Hamrell, 1978). Hence indireot links", if important, must be between fertility and mortallty. For example, If a child's death oaused Its parents to inorease 23 planned births by more than one - e.g. beoause it improved the peroeived oase for insurancep in order to achieve a given completed family size norm - that death might inorease family size. If so, higher ohild mortality among poor households would not out their sizoe as our oeteris paribus presentation on pp. 18-22 has implied. Many links between mortality and fertility - biologioal or economio-optimising, in both causal directions, and with either positive or negative sign - have been hypothesised (an exoellent sumary is Nugent and Walther, 1981, pp. 29-33). Such links can be collapsed into two sets. Via replacement effeots of mortality on fertility, extra children are produoed to compensate for past child deaths (partly, wholly, or, as above, more than wholly). Via crowdina effects of fertility on mortality, more ohildren in a family alter - normally, inorease - the death-risks to the siblings and, exoept possibly for the seoond pregnanoy, to the mother in childbirth. Replacement effects arise partly as deliberately-chosen compensation, and partly because pregnancy prospeots increase when suckling ceases. In the Kerala study in 1981-82 "the average birth interval was 22.7 months if the initial birth ended in a death within the first month ... 27.6 months if tnot, but within] the first year; 33.4 months" otherwise (Zachariah and Kurup, 1982, p. 18). Replacement may well be more than oomplete, as a form of rational insurance to reaoh a given family size norm, because "child losses are more frequent among women who experienoed such a loss previously" (Ruzicka, 1982, p. 26). However, the importance of the opposite, crowding, effect - habitually neglected in disoussions of "over-insurance" against child deaths, via replaoement effects above unity - is exemplified by Indian evidence from the early 19708 that, as the birth interval from the previous child rose from "less than 1.5 years" to "more than 4.5", the IDR fell by over 75 percent (Ruzicka, 1982, p. 24). To test the net outcome of the two effeots, 95 post-war populations were sampled, to discover whether "the number of living offspring of women of a specific age is roughly constant in a particular year, regardless of child mortality". It proved impossible to rejeot the hypothesis that over a woman's lifetime "cross-sectional variation in cumulative fertility is exaotly compensating for variation in child mortality", i.e. that positive replaoement effects and (normally) negative crowding effects were equal. A Brazilian time-series has similar implioations. However, in several LDCJ, rural-urban fertility gaps appear to imply overoorrection for child mortality gaps - i.e. if the national replacement rate is unity, it is usually more in rural and less in urban areas (Sohultz, 1981, pp. 137-40). That, in turn, seems almost wholly traceable to higher levels of urban education, especially female (and hence later marriage and lower couple fertility), rather than to higher rural replacement fertility at given levels of eduoation (Birdeall, 1980, p. 52). Indeed, micro-studies suggest a substantially greater "derivative of births with respect to ohild deaths" for urban than for rural areas (Schultz, 1981, pp. 141-3). Do these links between fertility and mortality help explain why big households now tend to be poor; and/or why poor groups tend to contain persons who live in smaller households? It is tempting to argue that child replaoement effect must be highest relative to crowding effeot for the poorest, and least for the well-off, because family size and child/adult ratios do inorease with poverty (since, as we see on pp. 33-8, family structures and net emigration rates if anything offset this relationship; it 24 has to be births and deaths that explain it). That tempting argument is unsound. Replacement effect is only one of many determinants of birth-rates; and crowding effect, of death-rates. Indeed, the "derivative of births with respect to child deaths" appears to be higher for the better-off than for the poor (and for Rio de Janeiro than for India); but it is not clear whether this finding (reported in Schultz, 1981, pp. 141-3) nets out crowding effects, viz. the derivative of child deaths with respect to births. If so, the finding is surprising, because it would operate against the known positive link of poverty to household size; but it would not be inconsistent with that link. We know too little about (environmentally variable?) relationships of poverty levels to (i) fertility, (ii) mortality - even if they are treated separately as if there were no crowding and replacement effects - and a fortiori (iii) the modifications due to (non-linear?) variation with income-per-CU in exposure to crowding effect, and in propensities to insure via replacement effect. Since undernutrition significantly increases death-risk only if severe (Chen et al., 1981; Lipton, 1983), commonsense suggests that crowding effect is greater in the poorest families than elsewhere, and that - if the old-age-security motive is predominant, and the net cost of extra children to the parents small - insurance via replacement effect would also matter most to such families. (g) Age of marriage and duration of unions We are fairly certain that - holding education constant - poverty and low status normally characterize people who marry later, cohabit for shorter proportions of their marriage, and are fertile for shorter portions of the periods of cohabitation; but that female education has similar effects, although it normally goes with wealth and status. Inferences from these facts to MHS are fraught with difficulties, however. Why? For each variable that might be related to mean household size (MHS) - e.g. for mortality, marital fertility, age and duration of marriage - we have to ask three questions. What is the usual sign (+ or -) of the direct relationship, assuming other variables that might affect MHS do not alter? How does the importance of the direct relationship, given its usual sign, vary among sets of households at different levels of poverty (e.g. MEP) and of status (e.g. landholding, job, caste)? Are there indirect relationships, e.g. via the effect of mortality upon fertility, and if so how do these differ among the household sets? In the case of the two variables so far considered, the third question, that of indirect relationships, could be, if not ignored, reasonably treated as subsidiary. (Thus replacement effect and sib crowding effect have plainly offsetting effects on MHS). Therefore, the answer to the first question, about sign of relationship, was fairly straightforward. So we could concentrate on the second question, differential levels of the variable, and hence directly of MHS, among poverty- and status-groups. For example, marital fertility obviously raises MHS; within the normal range of variation, it is not greatly altered by other demographic variables; so we could concentrate on the inter-group differences in the variable's effeot on MHS. Duration of cohabitation - which is overwhelmingly dependent on age 25 of marriage - does not permit this simplified reduction of three questions to the second only. It interacts with two variables, household struoture and overall (though not greatly age-specific marital) fertility, and therefore has major indirect relationships with household size, raising the third question, and making the second (direct relationships, and the differences in their effect on MHS as between groups of households) hard to sort out. Indeed, by concentrating on one or other indirect effect, some analysts have appeared to turn the answer to even the first question - the sign of the relationship between marriage-age and MHS - into a near-tautology. Unfortunately, the "tautology", for different experts, asserts relationships of opposite signl Some argue that later marriage, via longer retention of unmarried children in the average household, must increase MHS: "AssumEing] that (i) all household heads (HH) are men; (ii) every man becomes a HH if he survives long enough; and (iii) once a man is a HH he remains an HH, [then, because MHS equals total population divided by number of heads of household, we can assert that:] if men become HHs later in life, there will, at any one time, be fewer household heads [and] thus fewer households and MHS will be greater" (HaJnal, 1982, p. 463-4). Others argue that later marriage, by decreasing average lifetime fertility, must make households smaller: "If the age at marriage falls, [then because over the fertile period the per-woman] birth-rate rises ... MHS will [rise]" (Hayami and Uchida, 1972, p. 499). Plainly, each of these conflicting propositions can be stated, on distinct assumptions, as a tautology. If births-per-woman are unaffected by age of marriage, and if - to take one set of household formation rules consistent with Hajnal's three assumptions - all children leave the parental home to set up new households if and only if they marry, then later marriage must increase MHS. If children's period of residence with parents is unaffected by their age of marriage, but births-per-person vary inversely with it, then later marriage must reduce MHS. Unfortunately for simplicity, it is fairly obvious that later marriage reduces births-per-woman (for a summary of the strong effects in LDCs, see Birdsall, 1974, pp. 26-7, and 1980, pp. 49 (fn. 4), 57-8) and tends to prolong each partner's period of residence at a parental home. The effects are further complicated by two factors. First, multiple households (fn. 16) retain married sons or daughters for varying periods, and are liable to become smaller - through death of the parent(s) - as the children's marriage age is delayed. This is an important effect where mortality is high; simulations suggest that, under one plausible set of household formation rules, a rise in bride's age of marriage from 19 to 25 reduces the proportion of nuclear households from 62 percent to 54 percent (Wachter and Hammell, 1978, p. 49). (This is especially suggestive in explaining links of poverty to higher MHS, because nuclearity, tending on its own to make later marriage increase family size where mortality is high, is - like both mortality and delayed marriage - associated with poverty: see below, pp. 27-32). The second oomplication is that household splitting (whether multiple households are common or rare) often happens for reasons other than marriage, espeoially migration. Solitary male migration is a very oamon "life-cyole" phenomenon, especially for the rural poor but not ultra-poor, (Seo. (i)). To the extent that later marriage delays the oonversion of these one-person, 26 mainly male, migrant households into married couples later, it reduoes MkS, even assuming that the couple's lifetime fertility is unaffeoted. To the extent that later marriage affects mainly oomplete families, where the partners had, before their marriage, resided with their parents, the "Hajnal effeot" predominates, and MRS inoreases. It is my strong hunch - presented as such, in this Working Paper, for oritical review - that in the fairly long term HaJnal effect is overwhelmed by the effeot of early marriage in ramiing femle lifetime fertility; so that early marriage indeed goes with lar8er family size. This hunoh arises because there are so many ways to escape the assumptions required for the Hajnal effect. Children can split from the parental household before marriage, e.g. by migration; and they need not split from it after marriage. The number of CUs that can be supported by a household's main occupations - and possibly the size of the house, to judge by evidenoe from both Nigeria and Karnataka (Caldwell, 1982, pers. comm.) - substantially affeots, for the poor perhaps largely determines, the number of adults it supports over and above the parental couple. Later marriage would, therefore, by adding temporarily to the size of the parental home, induce offsetting faotors, viz. a greater propensity by the parents to encourage migration by unmarried children, and a lesser propensity to retain the married couple in the parental home for any signifioant period. It is, however, not nearly so easy to offset the increase in size due to higher fertility and more small children, especially in a nuclear family. So in what follows it is assumed, plausibly but of course pending proper testing, that in the fairly long run klS-increasing effects (via more births-per-woman) of early marriage in a group strongly outweigh MHS-reducing offects (via briefer retention in the parental home of potential new heads of mall households). Historical, and to a lesser extent recent, work suggests that later marriage accompanies bad harvests, recessions, and - given the level of education - oross-seotional poverty (Cassen, 1978, pp. 22-30; Schultz, p. 13; Laslett, 1971, p. 86, and 1972, p. 154; Birdsall, 1974, pp. 27-8). Historioally, prolonged resident service was a major factor enlarging MHS in richer, and reducing it in poorer, households (see below, See. (k)) - a factor closely conneoted to delayed marriage by the poor (Hajnal, 1982, pp. 482-3). Resident service is normally much less important in today's LDCs than historically in the NDCs; this weakens one major factor associating poverty with later marriage (and hence, we hypothesize, with smaller MkS). Moreover, education is spreading; it is strongly linked (i) to higher income or MEP (and less strongly, to higher status, job or asset position), and (ii) to delayed marriage; both the links are strongest for female post-primary, but are not confined to that form of education. For example, for a large sample of wives in Kerala in 1980, the regression ooeffioient upon "years of schooling" (holding desired family size and KAP oonstant) of "age at marriage" among women aged over 25 was 0.69 (F e 5.5, sig. at 5%); "educational achievement is the principal determinant of age at marriage Euhich] showed an inorease of 3.7 years over a 25-year period". Since "for a boy, his chance of receiving a good dowry is much brighter ... when he completes his educations, and sinoe this "chance" is anyway suah likelier to be important for persons of higher income and status, the spread of post-primary edueation associates delayed marriage (and hence, we hypothesize, lower kiHS) with better-off people - reversing the historioal association - even where suoh education is 27 overwhelmingly male; this oan be inferred from the strong association, in the Kerala data, between higher oaste, higher MEP, and age of bride at marriage even holding her schooling constant (Zacharaiah and Kurup, 1982, Tables 4-5 and pp. 20-22). While age of marriage is the main channel through whioh duration of cohabitation influenoes inter-household variations in MHS - the inoidenoe of illegitimaoy, historically in NDCs (Laslett, 1971, p. 142) and currently, we suspeot, in most Asian and Afrioan coounities being rather small, well below 5 peroent - also important are duration and interruption of marriage. In most of Africa, migrant male mineworking - and in muoh of Latin Amerioa, migrant female domestic service - often leaves single, widowed, divoroed or abandoned women. These tend to be poor by assooiation with three thingst their status and its socially ascribed disadvantages, their own or their husbands' type of unskilled labor, and their households' high dependenoy ratio. In seotion III(c) below, we show that much of the exoess poverty of female-headed households - which are normally amaller than male-headed households - is associated with their heads' greater propensity to be widowed, divoroed or "abandoned", a factor obvioualy also responsible in part for the household's amallness, both by definition and because the male's absenoe probably reduces incentives to ohildren to prolong their own stay (e.g. inheritance). In a Central Javanese village in 1972-3, among the poorest women aged 40-44, one-quarter of potential reproductive time had been lost through marital disruption, as against 9% for middle-income and 2% for better-off women (Hull and Hull, 1976, p. 23, ftn.) This curtailment of unions may be as important as the numerical effect of an absent spouse in "linking" the relative smallness of female-headed households to their relative poverty. In general, marital fertility seems to outweigh duration of unions - at least as represented by the main oomponent of the latter, age at marriage - as an influence on how income-groups or status-groups differ in respeot of the number of ohildren ever born to a oouple. In rural Karnataka in 1961, in households of "laborers and temporary tenants" a 9.2% higher proportion of women aged 15-44 were married than was the oase in the households of less-poor (top 70% by land-holding) owner-oultivators and permanent tenants. However, per 1,000 ourrently-married women aged 15-44, the former group had 12.6% fewer births (Mitra, 1978, p. 213). Income-specific evidenoe, within groups or otherwise, for India is not availablet but education, literacy and urban residence - all independently associated with later marriage (Cassen, 1978, p. 51) - are all linked also to higher income, both as between job and status groups and within them. But the link of mother's education and literacy to fertility deoline is stronger than their link to age of marriage. Through both routes, the low level of female eduoation among the poorest may alone go far to explain their large families. (h) Complex householdas: larger, but rare, espeoially for poor and low-status Household complexity in most LICs probably takes two main forms: lateral extension of several married siblings, and multiplicity of couples from auccesaive generations (normally because one or more married children stay in a parental household). Complexity normally increases a group's MHS. Complexity is held by some to be bad for "development" and to be aasociated 28 to share income with related couples who are less energetic. Others claim that complexity is good for "development", because each couple provides insurance to others in the same household, and thus encourages risk-taking (presumably pooled reserves could permit other scale-economies also). Our concern is not to adjudicate these disputes, but to ask whether complexity can help in explaining the correlations of high MHS with household poverty (now), household affluence (historically), and/or high status, job and asset positions (now and historically). However, the disputes are relevant to these questions, for three reasons. First, if complexity were clearly "developmental" (or anti-developmental), then - assuming at least some of the benefits to be internalized - that would create long-term links of complexity to affluence (or poverty), at least in growing economies. Second, settlement of the disputes would have major policy implications, for the sort of family structure (and implicitly MHS) which investment and incentive policies should seek to reward or penalize. Third, the disputants often each imply a teleology, in which "development" would lead to major falls in complexity (and probably MHS), especially in more-successful and less-poor household groups, if complexity impeded development incentive - or to rises in complexity and MHS, again especially among the non-poor, if complexity assisted development via insurance. Both the pro-complexity and the anti-complexity view are unhistorical, and rather unempirical. Both are unhistorical, because they see complexity or nuclearity as a given condition, tending simply to change the pattern of growth, poverty and development - whereas most sociologists increasingly emphasize how such societal norms are flexible, and adaptive to economic circumstance, even if supported by quite elaborate structures of norms or theologies (see, for instance, Rudolph and Rudolph, 1967). And pro-complexity and anti-complexity are unempirical, because they look neither at forms of activity - small farming, rural carpentry, urban hawking? - in order to identify the sectors and places in which complexity (and other determinants of higher MHS) increase or worsen the prospect of escape from poverty, nor at the evidence about the levels and trends, in various circumstances, of complexity itself. Such historical evidence as we have suggests that multiple or extended households in pre-industrial Europe, even Eastern Europe - except Russia and Serbia - seldom accounted for more than 10-20 percent of homes and 15-25 percent of persons; that, cross-sectionally, such households were, and are, especially rare among the poor; and that there is no systematic relationship, in time-series, between complexity (in its various forms) and the various indicators of household poverty-reduction or "development". The process of "Sanskritization" (Srinivas, 1962) - of imitation by poor households, as poverty recedes, of the styles of life common in higher-status examplars - extends far beyond Hinduism or India, and may today push the urbanising or self-enriching household towards complexity, at least as strongly as the olear links between MHS and poverty push it the other way. In any case, it is olear that the cross-seotion association of poverty and nuelearity has not (in any low-income sooiety for which we have evidence) yet been overborne by any time-series dissociation between them that might exist. 29 Complex households are likely to be bigger than othera. In Aidan-ka-Was, Rajasthan, in 1961-2 - a village of 47 households and 303 persons - all eleven households with nine or more members, oontaining 37% of the village's population, were complex, as against 13 of the 24 households with 5-8 members (49%), and only one of the twelve smaller households (14%) (Choudhary, 1964, pp. 137-40). In the early 1970s, "complex" MHS was 14.9 persons in Mali villages, and "nuclear" MHS was 5.1 (about half the households being nuclear). Comparable figures in three Northern Nigerian areas, each for three villages, were 7.1 and 4.9 in Sokoto (72% nuclear); 10.9 and 6.2 in Zaria (51% nuclear); and 7.5% and 5.1% in Bauchi (64% nuolear) (Norman et al., 1981, pp. 21-2). In two Bangladesh villages the gaps were smaller: 7.42 (5.86) in Bhatpara, 7.26 (5.99) in Bhabanipur (Khan et al., 1981, pp. 8, 10). Does the larger average size of complex households mean that groups - by location, status, MEP, eto. - with a bigger proportion of oomplex households will usually show higher MHS? It need not follow, for familiar reasons of "ecological fallacy"; for example, a community, initially comprising entirely nuclear households, could - while maintaining the same total number of households and persons - transfer members from some nuclear households to other related nuclear households, making the latter complex (and bigger in MHS than the former), but obviously without ohanging community MHS. A cross-seotion version of such a situation, indeed, must underlie the claims for England (1574-1821) that "mean household size in a oommunity does not appear to be correlated at all highly with" the proportion of complex households (Laslett, 1972, p. 126; a similar lack of relationship among Indian groups is posited by Shah, 1968). Yet in reality, for 64 English communities in 1574-1821, the simple r between MHS and complexity, averaging Laslett's two indicators of the latter, is *.3412 (Laslett et al., 1978, pp. 70-73). The three villages in each of three Nigerian areas (Norman et al., 1981, pp. 21-2) show a close relationship between MHS and oomplexity, though N = 3 is too few to oonclude muohl Anyway, the link of complexity to MHS is stronger aoross groups than aoross places. How strong is the inference from a group's high degree of oomplexity to its high MHS? This must partly depend on the extent of endogamy within that household group. Pressures within a caste, status-group, or asset-owning group tend fairly strongly towards endogamy; much more weakly so, for an income, MEP or occupational group; and for many localities, such as Indian villages, the pressures are towards exogamy. The "more endogamous" a group, the likelier it is that high MHS among its subgroup of complex households is "purchased" at the cost of lower MHS among its (oomplement) subgroup of nuclear, including sole, households; though if this "purchase" goes far enough it eliminates some of the smaller nuclear households, e.g. by retaining married offspring inside the parental home. In general, however, for both households and (less certainly) household groups, greater complexity acoompanies greater size. Before we show that complexity also goes with lesser poverty and higher status - in cross-sections, but not clearly in time-series - we need to examine whether complexity is at all *important". The incidenoes cited from Africa and India on p. 29 suggest that significant proportions of households - and even more, given the larger typioal size of complex households, of persons - are affected 30 by oomplexity at any moment of sampling; and since the couples heading many nuclear households, at any such moment, have previously been embedded in an in-law's home (and have then turned two nuclear households into one complex household), a substantially larger proportion of households has had some experience of complexity during its family cycle. However, there is some reason to believe (A. Shah, 1968) that our scanty Indian village data somewhat overestimate the average rural incidence of complex households. Epstein obtained 8-10% for two Karnataka villages in 1953, which "conforms to the all-India picture" (A. Shah, 1973, p. 199), but a rather higher proportion, say 15% nationally, may be applicable: of. the 20-29% estimated from a small sample of studies by Kolenda (1968, P. 390) and the 32% in Shah's Gujarat village (ibid., p. 27). Certainly, in the history of most NDCs, Laslett's teams have demonstrated that complexity is far less important than was once believed. Of 64 relatively well-documented English settlements (sampled data range from 1574 to 1821), about 8.6% of households were "solitary"; 3.2% "no family", including a few unmarried or widowed siblings; 71.2% nuclear; 11.9% extended; 3.7% multiple; and about 1% indeterminate (Laslett et al., 1978, p. 74). One in six households, and perhaps one in four or five persons, lived in complex households. Other pre-industrial NDCs, except in Serbia and parts of Russia, appear to show a similarly small incidence of complexity, very seldom reaching 25% of households (Duptquier and Jadin, 1972, pp. 290-1, for Corsica; van der Woude, 1972, pp. 306-7, for the Netherlands; Andorka and Farago', 1983, pp. 288, 293, for Hungary; Schmidtbauer, 1983, pp. 364-6, for Austria; Danhieux, 1983, p. 414, for Flanders). Simulations of the effect of quite large changes in mortality (Vachter and Hamell, 1978, esp. p. 48) show that this low complexity cannot be attributed, to any significant degree, to high death-rates among families "wanting" to remain complex - e.g. by a high risk that heads of multiple households, who are mostly fathers sharing a house with their married children living at home, would anyway be old enough to die soon. A low incidence of complex households is due mainly, not to demographic variables (though marriage-age is of some importance), but to a preference for nuolearity. There is, however, evidence associating such preference with high status, asset, and job positions; for India, perhaps surprisingly but quite clearly and understandably, 22/ with absenoe of industrialization and urbanization; and to a lesser extent with poverty (lower MEP). Shah demonstrated the linkage of high caste status to complexity, both for his Gujarati village in both 1955 and in 1821 (A. Shah, 1973, pp. 93-101; see also Kolenda, 1968). He also indicates that complexity is linked with urbanization, and perhaps industrialization also (see especially Lambert, 1963, for Poona), from a wide range of almost unanimous quantitative studies, for post-Independence India as a whole (ibid., pp. 148-51). Other historical and recent data confirm this. In England from 1570 to 1820, great houses were much likelier than the poor to have complex households (Laslett, 1971, pp. 95, 181; 1978, pp. 93-4) Data for Romans, S. France, in 1586 confirm this (Ladurie, 1979, p. 4). In Florence in 1427, 22.7 percent of the 427 richest households contained two or more nuclear family units, as against 6-9 percent of 9374 other households (Klapiseh, 1972, pp. 277, 279). The data in the 1983 papers reported above (p. 30) provide 31 unanimous support for the link between complexity, status and affluenoe from pre-industrial NDCa. Similarly, a review of post-war African and Asian data concludes that "households ... consisting of kin and non-kin generally oour among rioher individuals" (Goody, 1972, p. 122). In kidan-ka-Was, Rajasthang in 1961-2, the scheduled castes (the likeliest "poor group") were mainly nuoleated; but 19 out of 32 intermediate (mainly farming) oastes were joint (Choudhary, 1964, pp. 137-40). In Patadia villaget Gujarat, in 1963, the 71 scheduled-tribe families (plus the non-Hindu family) were split 50-50 between oomplex and nuclear; but ten of the twelve non-soheduled Hindu families were oomplex (Shetty, 1963, p. 40). In urban Malaysia, poorer households also tended to be simpler (Mazumdar, 1981, pp. 4-5). In rural Bangladesh, 665 of nuolear (but "only" 47% of extended) families in Bhatpara village oultivated below one acre; the respective proportions were 81% (75%) in Shabanipur village (Khan et al., 1981, p. 8). 23/ What are the mechanisms? First, extendedness - where feasible - may well decrease risk of poverty given family size. For example, in three N. Nigerian villages in 1974-59 all six large nuclear households (over 9 residents) were among the poorest 30% of all households by inoome-per-CU, as against 7 of the 19 large - and 1 of the 5 largest - extended households (Matlon, 1979). This may be linked with eoonomies of scale in oonsumption (Sec. IV (d)). Second, poverty aocompanies nuolearity partly because both go with low workforoe/dependent ratios. In a 1975 urban sample in Malaysia, "Joint households help families esoape from poverty, beoause (they] add more to earning strength than to dependenoy" (Mazumdar, 1981, p. 359). In nine N. Nigerian villages, extended families averaged about 25% fewer dependents per male adult than nuolear families (Norman et al., 1981, p. 22). This could mean that (i) extended families are less prone to ohild mortality, or less fertile, than nuolear families; or that (ii) extended families are easier to form and keep together when the various nuclear couples have few small children. If (i) occurs, it is an effect (rather than a oause) of a low incidenoe of poverty; extended families, because less likely to suffer from very low income-per-CU, would indeed be expected to show lower ohild mortality than nuclear families. As for (ii), the evidence does not show any systematic relationship - given the level of income per person or per CU - between nuolearity, or extendednesst and low (or high) fertility (Nugent and Walther, 1981, p. 14; Mitra, 1978, pp. 224-5). Third, the complex household oan be a sign of status, to which people aspire as their status increases. This is associated with "Sanskritization" in India, and helps to account not only for the close relationships of complexity to higher oastes, but also for its persistence, even inarease - though not from a high base level - with growth, industrializationt and urbanization (A. Shah, 1973). Finally, there are the three major "eoonomic" explanations of "multiple downward extension" (Laslett and Wall, 1972, pp. 30-31) - married ohildren living in a household of whioh one of their parents is the head - which, even if itself short-lived, ia probably the starting-point of most household complexity. Two of these explanations explicitly suggest that partioular groups of not-very-poor people would be likeliest to be involved. 32 One theory suggests that extended families are most comon where the experience of the elderly is rendered valuable by environmental uncertainty or fluctuation. Aoross Indian districts, high farm profit variability is significantly linked to the incidence of extended families. Farming families, especially in risky areas, would then be more prone to "extendedness" than (normally poorer) labor households, and so it turns out (Rosenzweig and Wolpin, 1979, p. 4 and Table 4). Another explanation of why better-off people are less unlikely to have complex households is wealth transfer - the "will-shaking" theory of household complexity (Butler, 1903). Less-poor fathers can keep their working children on the family farm (or in the firm) by the lure of inheritance; the poorest, in particular, have little to pass on. "Poor men's gandaye [extended family farm systems in Northern Nigeria] effectively collapse because their sons migrate, concentrate on outside work, or fail to obey them" (P. Hill, 1972, p. 147). Inheritance rules, written or implicit, in many communities stipulate "that children live with and care for their parents", and types of joint household have been empirically linked to types of inheritance system that promise advantage to the younger members (Nugent and Walther, 1981, p. 18). These two explanations - experience and will-shaking - address the question of why a son or daughter might prefer not to nuoleate upon marriage; both favor non-nucleation much more in households with heritable assets. (So does the obvious fact that a multi-couple household is more tolerable if it can afford to allot each couple a separate bedroom.) The other common explanation, search for security in old age, addresses the question of why the parents might seek a non-nuclear solution. Since absolute risk aversion increases as income-per-CU falls, and since the poorest have fewest prospects of a secure old age unless supported by their family, parental demand for extendedness is presumably highest among the poorest; yet we know that extendedness is least prevalent there. Therefore, the correlation, with poverty, of parents' "demand" for extendedness must be outweighed by the correlation, with affluence, of married children's "supply" (given the "income-elasticities" of the two functions). Extended families tend to be bigger and better-off than nuclear families. They therefore could help to explain why higher-status groups tend to have larger families. The evidence from Patadia and Aidan-ka-Was, locating high propensities to extendedness as a characteristic of such groups (rather than of individuals), supports this; so does Shah's evidence, and so do the frequent links between social group membership, type and scale of asset holding, inheritance system, relevance of experience of elders, and family type. But why do the poor in total populations have bigger families? Here, extendedness seems to point the wrong way, for it is less frequent among the poor. If extendedness is a norm for groups - the high-caste, the landed, etc. - and is only weakly linked to poverty if status is held constant, that "wrong pointer" is removed, but the data do not cast any light on this issue. 33 (1) Migration: the relationship to differenoes in MRS If an individual is likelier to migrate from poorer - or lower-status - households than other households (even without thereby changing them from oomplex to nuolear in form), then suoh emigration would tend to make these households of origin smaller than others. Sinoe individual migration in MICs is ooncentrated among single young persons, their households of destination would also tend to initially (and for the first few years afterwards, perhaps) to show both relative-status and HEP oharaoteristics similar to those of the households of origin. Thus migration - at least of some important types - tends to reduce MRS in the status and MEP groups that are more affeoted by it, or among which it is becoming more important relative to other groups. Before looking at the evidence, and the impaot on poverty-status-MRS relationships, we should stress that they are almost unaffected by the three main sorts of migration - all largely intra-rural - in LICs Marriage migration (usually for a few miles only, to meet the requirements of village exogamy) remains much the likeliest explanation of any observation that a person in India (Bose, 1967), and probably in most other LICs, lives away from his or her birthplace; although one can build models where marriage migration is associated with systematic differenoes in MRS by poverty or status, such models are not plausible. Total household migration appears to be less important, as a share of long-term migration, among the better-off than among the poor (Connell et al., 1976, pp. 12-13), but it, too, would not seem to alter the migrant households' MRS. Temporary migration for a few weeks, usually to seasonal farmwork, is similarly "poor-selective" (lbid., pp. 79-80, 122-3) but by definition leaves inter-group variations in "normal" MHS unaffected - though it could help explain low MRS among laborers in peak-aeason surveysi Therefore, in asking whether migration affects MRS differently among poor and others, or among low-status and others, we are enquiring mainly into long-run, individual (as against household), non-marriage, and mainly rural-to-urban migration - hereafter "LINIM". Prior to this enquiry, three faots should be stressed. First, the proportions of LIC - as opposed to MIC - populations who undertake LINUM, and hence the rate of urbanization and the urban shares in most LICs, are quite small (Lipton, 1982). For example, in 1961 only one Indian in 33 was for any reason a rural-born town-dweller (Bose, 1967), so these persons are unlikely to account for muoh of the very strong size-poverty or size-status links. Second, LICs show a quite new link between LINUM and poverty-status-MRS relationships. In most pre- and early-industrial NDCo - and even now In Latin Amerioa - LINUX tended to produce female-headed, and henoe (p. 13) relatively small, urban households. In today's LICs in Asia and Afrioa, LINUM is predominantly young and male (Connell et al., 1976, p. 39). Since migrating male-headed households tend to be larger, and for other reasons also, such migration does not appear to counter the tendenoy for urban households, especially among poorer persons (Table 2), to be at least as big as rural households in LICs. 24/ Third, the effects on the poverty-status-MRS links from LINUM are not only small and historically novel, but also very oomplex and hard to isolate 34 from other ractors. Thus (i) an MEP-declle or status group, exposed to an unusually high rate of suoh migration, would show below-average MHS in originating households, if other things were equal. However, they are unequal, because multi-sibling households and persons of higher birth-order are more prone to emigration (Connell et al., 1976, p. 46). LINUM therefore tends to be higher from households with high MHS (Harbison, 1982, p. 232). Indeed, this even shows up as somewhat above-average MHS, even after LINUM, in the originating households (Lucas, 1982, p. 19, for Botswana). As for the effect of an above-average LINUM upon a group's MRS in destination households, it is presumably positive if migrants join kin, negative if they set up new one-person homes. Moreover (ii) LINUM selects the fairly-poor and fairly-affluent - principally young men from the second-highest and aecond-lowest quintiles of households by MEP - from unequal villages. Further, (iii) these migrants tend, respectively, to be students (or persons following relatives in a "ohain" to assured jobs), and work-seeking "step" migrants (Connell et al., 1976, pp. 198-200) - so even LINUM is not uniform in type. Finally, (iv) suocessful LINUM migrants - and they are many, since LINUM selects risk-takers, the better-educated, and people with better information - move up both MEP and status scales, while unsuccessful ones often return (daVanzo, 1982, pp. 115-7); both groups change MRS in the process. As far as urban households are concerned, recent ones, formed by immigrants or otherwise, tend to have lower MRS. Are they also poorer? It appears to have become almost a new conventional wisdom that the urban poor, recent immigrants, underemployed, and informal workers are practically non-overlapping groups. At least, it seems to be suggested, membership of any one group renders membership of any other less likely (Bromley, ed., 1978; Lloyd, 1979; Mohan and Hartline, 1980). If this were so, then the normally smaller families of recent migrants would be linked to lower risk of poverty. However, the evidence for the new view is primarily Latin American, and even there it is thin. In Africa and South Asia - in the LICs especially - detailed survey work confirms that recent urban immigrants are likelier than other urban persons to be young men with no, or small, households; to be unemployed, informal, or casual in workforce status; and to have low income-per-CU (Papanek, 1975; Sabot, 1977; Lipton, 1983a, Sec. III. d.4). At the urban end, in countries with substantial poverty, destination households for LINUX therefore tend to "link" poverty and low status - relative to setttled urban households - with lower MRS. (Clearly this is overwhelmed by other, non-migration linkers, for urban MHS is strongly and positively linked to poverty: see Table 2). At the rural end, we have crudely characterized above the households most liable to LINUX. The poor, rather than the poorest, are pushed out into a wandering, probably increasingly rural, quest for work; the modestly better-off, rather than those tied to substantial rural assets, are able to support temporary eosts of their sons' "pull" migration towards urban schools and Jobs. What is the impact on rural households' size-poverty nexus from such migration patterns? For Rajasthan and Gujarat, medium-term migration of individuals from households is signifioantly contributing to reduced MRS in some villages. For 11 villages, average household size and the proportion of out-migrants to the 35 population of reporting households are inversely related (r = -.6154, significant at 5%); the same relationship holds for student and working migrants separately (Connell et al., 1976, p. 193; household size from Tables 1-3 of this chapter). 25/ Thus - although at household level it is the bigger households, with more siblings, which tend to become smaller via emigration - villages with substantial emigration, at least in this Indian data set, average lower MHS. Which rural households can get smailer, and so perhaps escape poverty, by this process? Agricultural laborers, at least in North and East India, appear to migrate (i) as total households rather than as indivlduals, and/or (ii) for temporary rural, rather than medium-term urban, settlement. It is in the intermediate castes - including the main "dominant" farming castes - that fairly high emigration rates reflect the chances for better-off households to reduce size and increase income (ibid., pp. 187-8). In rural areas, therefore, migration helps to link poverty to large MHS w'.thin the intermediate castes, and therefore within total populations. Small MRS among laborers, as an occupation or as a caste-linked group, is not as such associated with total migration; but laborers' relatively greater poverty (and hence shortage of information, education, and risk-taking capacity) is associated with types of migration - seasonal, step, or total-household - offering relatively unpromising prospects. Thus, as between rural groups, emigration does not help to link smallness (among labor househ-olds) to poverty, because it is the less-pocr non-laboring castes that are likelier to reduce MHS via emigration. Thus, in towns, migration helps explain the link of low-status groups to low MHS, but makes it harder to explain the link of poverty to high MHS. In rural areas, the opposite applies. But in either case, since only LINUM has obviously major effects upon MHS and is itself rather small (p. 33), the contribution of migration to poverty-status-MHS links is smallish also. Finally, individual migration (tending to decrease MHS) and joint or extended families (tending to reduce it) orten go together. Hence, among any particular set of households, they offset each other's impact on family size as a cause of poverty. Extended and joint families provide structures favorable to - and empirically are oorrelated with (Connell et al., 1976, p. 48) - the trying-out, comunioating, initially tisk-bearing and -sharing, ultimately portfolio-diversifying process of chain migration (Stark, 1975). Conversely, in Northern Nigeria, nucleation, male gandu, and female purdah combine to prevent any but short-term, seasonal migration (Connell et al., 1976, p. 48). (j) Family cycles and MHS The history of most couples and their offspring - from marriage, through the period when there are small non-earning children, to the time when children increasingly augment family income, and ultimately to the departure of the children from one or both surviving parents - may link poverty or status to MRS in two ways. First, family history may affect poverty (or status) and MHS together; for eam-ple, it is often claimed that the completed but young family and the single-member, especially widow, family are 36 partioularly likely to be poor. Second, the different household formation patterns of groups, in a society, with distinot status or affluenoe - for example through later marriage, or greater household complexity, among the better-o"f - may give such groups distinct family histories, and hence HHSs. Both effects vary among societies and over time, and can be drastically altered by "the requirements of domestic production" (Sieder and Mitterauer, 1983, pp. 339-45) or by an unbalanced sex-ratio (Wall, 1983, p. 473). Ideally a long-term panel survey is needed to investigate these two effects. Lacking this, we can examine age of household head, child/adult ratios, and reported marriage age in various group, though we then risk misreading genuine societal change as "merely" a difference between older and younger household heads, etc. The overall discussion of family cycles, as they interact with poverty and status, is left to Sec. III(d), but two conclusions from that discussion are relevant to our topic here - the effect of the family cyole on poverty-status-MHS links. First, the cycle seems to have muoh less relationship to poverty in most of today's LICs than in pre-industrial history in most NDCs (p. 54; of. HaJnal, 1982). Second, in several Indian villages, poverty is attributable to the family-cycle to the extent that jobs and land are not automatioally ascribed to persons born into a particular lifelong status-group, viz. a caste (pp. 54-5). The first finding reduces somewhat the likely power of family-cycles - unless they are very different among groups in an LDC community - to explain poverty-status-MlS relationships there. The second finding, however, could help explain why groups rigidly "condemned' to low status and poor earnings prospects keep small MHSs. Parents in suoh groups see little prospect that maturing children can overcome the barriers against training for, or entry to, lucrative trades or professions. They may thus see little gain from having many children. The "old" end of the life-cycle in LICs is clearer: it has very little to do with poverty-status-MHS links. Contrary to the experienoe of rich countries today - and with one major exoeption, that of widows in some Islamic societies - the old, unsupported, single-person household of low status is not a major source even of relative poverty, though micro-level evidence seems to be limited. In two village surveys, from the 1960s, in each of two Indian States (Rajasthan and Gujarat), single-member status is quite a good proxy for "widowed or deserted" and "old and alone" phases of some life-cycles. Some 7.7% of the 983 households in these four villages are single-membered, but these contain only 1.5% of the 5,200 persons. Among scheduled castes and tribes, 9.2% of the 130 households, with 1.9% of the 621 persons, are single-membered (R. Patel, 1964, p. 71; and 1964a, p. 421; M. Desai, 1966, p. 37; anon., 1971, p. 30). These differences (non-significant at 5%) are less than one would expeot from the smaller average size of scheduled-easte households. These phases in the family oycle, at least in these villages, contribute very little to the extreme poverty of the soheduled castes and tribes. In Kerala (Caldwell et al., 1982, pp. 28) this may be because isolated widowhood is so parlous for the poor that they go to great lengths to avoid it. As for occupation-linked poverty, only one village survey each in Rajasthan and Gujarat shows single-member families by main income source. While 11.0% of all 372 households (and 2.1% of 1,939 persons) in the two 37 villages had single-member status, laborers actually featured lower inoidences - 5.5% of 110 labor households, 1.1% of 541 persons (Patel, 1964a, p. 43; anon., 1971, p. 31). These are only a few villages; but it does not look as if the family-cycle phase that creates single-member households, in rural India at least, is a major reason why either laborers or low-status social groups are very poor despite small household size. The reasons for believing that this is more generally true are the negative empirical link in several LICs between age and risk of poverty (See. III(b)); and the need for old single non-workers, where there is no sooial security, to subsist from savings or family support, in either case breaking the link with relative poverty that is assumed in Western societies. The absence of a clear, stable or consistent link between poverty-risk and age of household head, both among laboring and among cultivating households, is confirmed by Gaiha's rural Indian sample for 1968-70 (Gaiha, 1973, Tables 18, 19). However, some evidence that the prime-age phases of the family oycle contribute to both the within-group and the total-population relationships of poverty to MHS is considered in Seo. III(d). In a sense, indeed, the argument that - since most families in most groups are nuclear - the phases when a household contains many small children "pulls together" poverty and high MHS does not require elaborate empirical support. We can also be confident that family cycles mean that point surveys overestimate (i) disutility due to poverty, and (1i) association of poverty with inequality. Both are weakened to the extent that the members of the "poorest decile" - for example - of households by income-per-CU at survey time are not always in that decile. Good and bad luck alone make this likely; systematic family cycle variations strengthen the likelihood. However, great caution is needed in attributing poverty or affluence, or big or small households, to "the" family cycle. It varies across cultures - though the small nuclear family norm, especially for poorer households, is widespread - and within societies. Also, as the evidence on age of marriage (pp. 25-6) shows, family cycles can be much altered, even in a given group, by economic incentive and stress. The direction of family-eycle effects on poverty and MHS in mid-life is fairly clear, but their size may be small, as we shall see on p. 54. (k) Non-family members, MRS and poverty In NDC pre-industrial history, recent researoh shows that an astonishingly large proportion of persons appears to have been resident as servants. This residence substantially, but not totally, aocounted for the positive MEP-MRS relationship then prevailing; the much smaller role of such resident servants in today's LICs partly explains the negative relationship found there today. Furthermore (Hajnal, 1982) the frequenoy of service by young unmarried persons in much of pre-industrial Europe was associated with later marriage among the poor; because the poor enter much more rarely into such service in today's LICs than in NDC pre-industrial history (while the well-off more ceomonly enter into higher education), it is now the better-off who tend to marry later (pp. 24-7). So the whole question of family size - and family cycles - as it affects status and poverty, is tied up with the pattern of non-family membership of households. 38 Among the hundred English ounuities in 1574-1821 with the most reliable data, servants oomprised 13.4% of persons - presumably 20-25% of workers; and 28.5% of households had servants (Laslett, 1972, p. 152). In 1748-9, In two Dutoh provinoes, the proportions were 14% and 12% (van der Woude, 1972, p. 308). Evidenoe of large numbers of servants in pre-industrial NDCa has been oompiled for plaoes with mostly nuclear families (HaJnal, 1982, pp. 456-7, 470-4), but even in Belgrade in 1733-4, where oomplex households were quite ooon, 10.3% of persons were resident servants (Laslett and Clarke, 1972, p. 379). Servant proportions seem to have fallen sharply with or before early industrialization, and as between country and town (Wall, 1983, p. 497; Hayami and Uohida, 1972, p. 504). Data from today's LICs very seldom show anything approaohing this inoidence of resident servants. In the Census of 1951, all unrelated persons formed only 1.2 percent of India's population; in rural China in 1929-31 the proportion was even smaller (Hajnal, 1982). Resident servants appear similarly rare in the intensive mioro-studies that I have surveyed, espeoially from India and Northern Nigeria. (This rarity is most unlikely to be due to any great extent to the alleged concealment of bonded labor from enquirers (P. Hill, 1982)). Several reasons seem plausible. The higher person/land ratios of today's LICs make it more costly to acoommodate a servant (relative to hiring a laborer) than was the oase in pre-industrial NICs. Today, too, the better-off have more access to labor-saving domestio devioes. Alao, higher rates of child survival may now help more of the better-off to meet family labor needs without servants; in a village in West Flanders in 1814 (Danhieux, 1983, p. 418), and in an Austrian village even holding the employer's eoonomic status roughly oonstant (S9hmidtbauer, 1983, p. 355), more servants normally went with fewer children aged over ten or so. 26/ Servant status itself, too, was sharply life-oyclioal in most pre-industrial NDCs: for instance, in seventeenth-oentury Austria, 35% of all servants of known age were under 19, and 46% were aged 20-29 (Schmidtbauer, 1983, p. 358). The great majority of resident servants were of one sex - sometimes female, e.g. if servant work overlapped with textile artisanship as in Brugea, Belgium, in 1814 (Wall, 1983, p. 461), perhaps more usually male; and servant families in residence appear to have been unusual (Sohmidtbauer, 1983, pp. 359, 362). Therefore, servant status historioally delayed marriage for the poor. Sooietioe where 10-15 peroent or more of persons were resident servants, therefore, reduced MRS among the poor in two ways. First, adoloscent sons went into service and transferred residence to richer households. Sooond, they thereby delayed marriage and reduced oompleted family size. Demonstrably, this was a major reason why in NDC pre-industrial history poor households were amaller than others; and aimilarly, the deoline of resident service - acoompanied by education that selectively defers leas-ooor ohildron's marriages (pp. 26-7) - pays a big part in the reversal in today's LICs of the traditional NDC link between poverty and small MKS. To a man extent, migration in today's LICs may have replaced service as a write of passage3, enabling big poor households to shed, for a while, unmarried sons. But migration is far less powerful than service in reduoing poor households' MRS relative to other househods. First, ohildren from om rich households migrate too. Seoond, while migration is like "going lnto servioew in reducing the size of the poor household of origin, only the 39 latter also inoreases MS4 in the (wealthier) destination households. The power of this second, double, effect, oan be interred from a single village in Kent, England, in April 1676 (Laslett, 1971, p. 66). Of the 178 persons in the 29 homes of "gentry and yeomen", 49 were servants. Not one servant lived among the 53 persons (24 hoseholds) of "laborers and poor men". Probably, these households were so small in part because many of their ohildren were in servioe with better-off households. (1) Poverty, status and family size: some Dointers Poor households, more than others, need the "degreea of freedom" supplied by ability to make overt, explicit choices, with predictable outcomes, in regard to variables influencing MHS and household composition. Yet poor households are less likely to Let such freedom. For example, fewer poor households are oomplex. Complex households oan provide insuranoe to eaoh couple (support by the others in sickness or other mishap), soale-eoonomies in consumption, and reserve labor (espeoially if such households feature relatively low fertility or child/adult ratios: Laslett, 1972, p. 56; Andorka and Farago", 1983, p. 306); or oan split into simple households when these advantages of complexity are outweighed by drawbaoks. Voluntary and "pull-induced" migration; controlled spacing of births; choices of the age of marriage or of the time when post-primary education (or of service or apprenticeship) takes an adolesoent or young adult out of the parental home: all these choices affecting MHS are least accessible to poor households. Yet poor households moat need these choicest to help them plan to avoid dangerous periods of stress, in whioh 1S might overstretch supportive working capaoity (given the household members' ages, the demands and rewards for their economic activity, and their mutual domestio requirements). On the other hand, poor households, more than others, suffer if affeoted by unoontrolled variables ohanging 4S1, and altering the planned balance between needs and production oapaoities. The "Yale school" (Schultz, 1981) persuasively argues that most households, rioh or poor, aot as if the parents planned the demographio variables, inoludingt by implioation, the time-path of MH3. However, the variables listed above - oomplexity, pull-migration, later marriage, out-of-household training - are in today's LICs usually muoh less open to poor couples than to rioh couples. In deoiding how many children they want, and perhaps sometimes even the oare or food that eaoh sex of child will receive, poor couples remain able implicitly to aot as if they "ohose" other variables - the central demographic ones of fertility and mortality - determining the time-path of M34. But these latter choioes are much more uncertain than the "rich oouplea' ohooseables" in the previous paragraph. For example, by remaining oomplex, a household almost enaurest for some years, bigger size and different structure than if it had split into simple nuclear households; but by oontrolling fortilityt especially with pro-modern means of birth-oontrol, a sexually aotive couple implicitly "chooses" the expected value of 31S (along its time-path) only with a very high coefficient of variation. The greater oapaoity of richer households explioitly to ahoose, with low uncertainty, variables adapting M1S time-trends to those of available Oonsumables is strengthened by two faotors. First, rioher households are better able to buy information, e.g. about birth-oontrol methods or migratory 40 prospects. Second, richer households are less pressed than poor households to adapt ohoices affeoting MHS to short-run requirements of production and earning - e.g. to the need for young teenagers to contribute family income (in excess of their oonsumption) by performing particular tasks - and can thus emphasize oonsumption-MHS balances more, when planning the time-path of MHS. The evidenoe suggests some worsening, as between pre-industrial NDCs and today's LICs, in the disadvantages of the poor in planning, with reasonable certainty, MHS time-paths that reduce the risk of periods of consumption stress. In some respects, little has changed. Of the MHS-influencing variables with fairly clear outcomes, complexity remains an option much more readily open to better-off people (though a somewhat less rare option, perhaps, in today's LICs than in most NDC history). However, "going into service", as an option to out parental MHS (and usually to delay marriage), was chosen by very large proportions of the poor in NDC history, but is available to much smaller proportions of the LIC poor today. Conversely, delayed marriage - historically the response of poor young adults either to poverty or to emergency - has in today's LICs become the response of the better-off to the prospect of further education. The upshot is not merely that the historical correlation between big households and rich households has been reversed, but that this has happened in a way especially damaging to the relative capacity of poor households so to plan MHS as to reduce consumption stress. Traditionally the NDC rich - despite having more servants, etc., in the household - had markedly higher ratios of children (under 10-12) to adults than did the poor. In most LICs today, the poor - despite a rather low rate of transfer, out of their households, of adolescents and young adults "going into service" - have much higher child/adult ratios than the non-poor. This rise in poor couples' relative dependency burdens (i) reduces their chances of escaping poverty, except to the (rather small?) extent that poverty is a life-cycle phenomenon, aotually overlapping only with the period of high dependency; (ii) means that ohildren, being heavily ooncentrated into poor households, tend as a group (and especially in suoh households) to be with adults who must give priority to immediate earning over child-oare and -socialization; and (iii) in the case of extreme poverty (Lipton, 1983), concentrates on under-fives where the parental couple oannot avoid nutritional risk to their healthy development, or even survival. To over-simplify, the better-off in LICs are able to respond to reduced mortality with a faster, more explicitly volitional, less bumpy "demographic transition" than are the poor. This has major policy importance - not, of course, because polioymakers should, or perhaps oan, intervene in (for example) the extent to which poor households send adolescents into resident domestic servioel But the importance of the issue is not matched by our information or analysis. (It will have been observed that much of my information on variables affecting MHS is from a handful of NDC-historioal sources, plus a handful of LIC village studies and national samples.) Hence the summary in Table 6 is extremely tentative. Table 6 takes the seven key household variables affecting MHS: mortality, couple fertility, duration of fertile unions, complexity, migration, family-cycles, and non-family membership. For eaoh variable, we 41 ask two sets of questions. First, does it help "link" poverty to high MHSs in today's LICJ, or to small ones in pre-industrial NDC., or to explain the reversal of the "MH3A-NH" relationship? Second, does the variable help "link" high status to high MHS, in pre-industrial NDC8 and in oontemporary LICs alike - and to explain the paradox in the latter, with high MRS linked to high status, jobs, assets, etc. and yet to poverty also? This summary sub-section has said little about the status-MRS link, though it was half of the paradox with whioh we began (p. 13). Mainly, that is because Table 6 summarizes most of the (highly tentative) findings. However, two other points are perhaps worth drawing out. First, the positive status-size link may represent a time-lag: people's income, and the assooiated MHS-related and other behaviour, ohanges more readily than the status, assets, jobs eto. that, wholly or partly, they inherit. Seoond, the fate of children may be worst in a high-status but low-income household, whioh is pushed by both income and status into a child/adult ratio and MHS exceeding what can be safely supported. 42 III. POVERTY AND HOUSEHOLD COMPOSITION (a) Composition. MRS. and Doverty definitions How are the household's risks of poverty related to the age of its members (inoluding the household head, HH), and hence to its dependency burden? How does the sex-composition interact with poverty risk? How do these effects wary over the family's development, from marriage to household separation, widowhood, etc.? These three issues of houwehold composition are dealt with in sub-aections (b), (c) and (d) below. The composition of a household, however, is not rigorously separable from the issues of poverty-MRS relationships treated in Sec. II, nor from those of demography and poverty-measurement in Sec. IV. The interaction of poverty risk and household composition with MHS is exemplified by the effect of HR's sex. On pp. 51-3 we suggest that, on most evidence, female-headed households are as suoh little, if any, more likely to be poor in LICs than are other households. If we ignored MHS, this faot could wrongly be taken to imply that women's "disadvantages" as household heads were small. However, we have seen (p. 13) that female-headedness, in NDC history and in today's LICs, typically accompanies much smaller MHS - saller by 25-45% on most data sets. Smallness normally, in today's LICs, accompanies higher MEP and lower poverty-risk (Sec. II(b)). Female-headed households, being typically muoh smaller, "should be" less poor. They are not; and this itself suggests female disadvantage. (Of course, to show a positive partial oorrelation - holding MKS oonstant - between poverty-risk, or low HEP, and female-headedness is to restate the problem, not to explain it.) Our findings about household composition as a correlate of poverty are bound to depend, too, on how we measure poverty. Also, a good indicator of "poverty" will, explicitly or implicitly, divide real household income or expenditure by an appropriately weighted indicator of the requirements of the household's men, women, and different age-groups of children. For a potentially poor household in LICs, explicit division of monthly expenditure by the household's "size" in Lusk CUs is not a bad indicator of satisfaction of requirements; indeed, even monthly expenditure per person is tolerable, since it ranks households not too differently from expenditure per Lusk CU (Seo. I(o)). However, findings about composition-poverty linkages depend on, and affeot, ohoices of indicator. For example, the use of "food/outlay ratio above 75-80%" as an indicator of severe risk of poverty - while it has some attractions as a short out (Lipton, 1983, Sec. II(b)) - probably overstates that risk among households with many small ohildren and understates it among other households. 27/ A similar, but more rigorous, critique of the Engel procedure is given by Deaton and Muellbauer (1983, pp.23-4), who suggest alternative methods of "equivalenoe soaling" - iM]lioit divison of MEP by an index of the absorption of welfare likeliest to be caused by the household's size and age-aex composition. In what follows, we are usually foroed by data shortages to use MBE, rather than per CU, as a poverty indicator. This is probably not a very serious distortion (Seo. IV(c)), but may invalidate some of the "oloser" finding and hunches about relationships of poverty to household oamposition. 43 (b) Age-composition. dependency, and the risk of poverty Two issues arise. The first, allowing the more alear-out conolusions, is how MEP, risk of poverty, eto. vary with the age-composition and hence dependency-ratio (DR) of the household. 28/ While poor households in today's LICs (especially, on the Indian evidence, in towns) have muoh higher proportions of very young people than do the better-off households, the curves may well flatten as we descend into extreme poverty, so that the ultra-poor and the poor feature rather similar (high) child/adult ratios. The second issue is the relationship between poverty and the age of the HH. Here, we find complex and conflicting results in different areas. The results differ not only in the linearity (and even sign) of the relationship, but also in its differential impact on men and women. This suggests that we must either fall back on the evasion, "culture-specificity" (which amounts to a denial that we can ever, in general, know what characteristics poor or ultra-poor people are likely to have), or else find intervening variables affecting the shape of the curve linking poverty to age of HH. Part of the problem is that the effect on risk of poverty if HHs are older due to later marriage (or remarriage) is different from the effect if the HH is older because further along the family cycle following a marriage at the national average age. A useful useful analysis of the relationship between age-structure and poverty is Visaria (1977), based on the 1972-3 NSS. To ensure that these results are not special to the two Indian States reviewed (Maharashtra and Gujarat) or the year, we have also looked at 1973-4 data; at all-India and Rajasthan data; and briefly at other States. These series confirm: - that the proportion of persons aged 30-44 seems to vary little, and unsystematically, around 16-18% with changing MEP; - that the proportion of persons over 60 rises only gently (though persistently) with increasing household MEP, e.g. from 3.4% in the poorest decile in urban Gujarat to 5.9% in the richest quintile (Table 7); - that the (much larger) proportion of persons under 14, except among the poorest 5-10% of households, decreases sharply as household MEP declines, especially in urban areas, e.g. from 49.6% in the poorest deoile to 20.5% in the richest quintile in urban Maharashtra (Table 7); - that, therefore (and espeoially in towns), not only child/adult ratios, but also DRs, rise sharply as poverty presses harder, with some alleviation of the increasing trend in the poorest decile. Table 7 also shows most of these features, more sharply in towns than in villages; and all this is confirmed by 1973-4 data. 29/ Moreover, the clear poverty-DR link is not just Indian. In 1969-70 there were 1319 under-fourteens and over-sixties per 1000 persons aged 15-59 in the poorest household deolle in Sri Lanka, 1138 in the seoond-poorest, 798 in the middle quintile, and 506 for the richest doclle. Comparable figures for urban Nepal (1973-4) were 1279, 1135, 783 and 417; and for Maharashtra (1972-3), 1203, 1037, 733 and 331 urban, and 1304, 1203, 960, and 776 rural. For nine South Asian data sets that rank household (MEP deoiles) by DRa, the simple r's between MEP decile-rank and DR range from -.973 to -.997 (Visaria, 1980, Table 4). 44 Table 8 shows the effect, for three Indian States, among "riner" MCP-groups. Most of the cellwise sub-samples are quite large. They suggest a child/adult ratio that clearly falls as HEP rises - but only arter some (very low) threshold MEP is reached. No clear link exists for the poorest 8-14% of households in rural Gujarat (4-11% in urban areas), and for the poorest 6-11% (8-14% urban) in Maharashtra. Other States JO/ show similar thresholds - for DRa as for MHS (ft. 11) - as do the all-India urban data (the all-India rural aeries in Table 8 seems to show a smooth rise in child/adult ratios even at extreme poverty levelst but the Statewise data show that this is an "ecological fallacy"). Although some of this flattening-out of the rise of child/adult ratios with increasing poverty is caught even in the "rougher" analysis of Table 7 (of. the data for rural older children), a flattening-out of the ratio, if it applied to no more households than the poorest 10-14% or so, would not be fully captured by the deoile-wise data. Table 8 shows it olearly. This flattening-out is especially important if, as Tables 8 and 12 and fn. 11 suggest, it oorresponds to similar flattening among the ultra-poor of the (otherwise positive) MHS-poverty relationship. Subsets of each of the poorest 2-5 MEP groups in Table 8 - subsets making up increasing proportions of groups as poverty deepens - are in households unable to afford enough calories to safeguard under-fives from periodio risk of serious growth faltering, and/or of Grade III undernutrition; unlike the mild-to-moderate undernutrition assooiated with less extreme poverty, this signifioantly raises infant and child mortality (Chen, 1980; Lipton, 1983), and probably depresses fertility (Frisoh, 1978, 1982). The very poor, like the moderately poor, are subject to economic-demographic pressures that tend to reduce average household age, increase child/adult ratios, and thus raise MHS as poverty deepens; but for the very poor, unlike the moderately poor, these pressures are offset by the physical and health impacts of undernutrition on births, IMR, and child deaths. Hence, for the very poor - unlike the moderately poor - a clear relationship of MEP to child/adult ratios or MHS may not emerge. These interpretations are tentative. A few sub-samples in Table 8 are rather small (though consistent, and cumulatively suggestive). Less-poor LDCs - perhaps lacking suffiioent physioal pressures, even among the poorest, to oonstrain fertility (Frisoh, 1978, 1982) - may show no "flattening" of ohild-adult ratio increases as MEP falls; there is a steady relationship for urban Colombla through all MEP deoiles (Birdsall, 1979, p. 130). Nor are plausible data from localized village or urban surveys available. But the smaller families, and lower ohild/adult ratios, among rural laborers (p. 12) are suggestive; that group overlaps strongly with the ultra-poor. Moreover, a turning-point in M1S and child/adult ratioa, around ultra-poverty, would oonform to a general pattern of turning-points in nutritional and labor-market behavior (Lipton, 1983 and 1983a). Among other variables, age- and sex-speoific participation rates, food/outlay ratios and now perhaps child/adult ratios and even 1411 - while they rise, up to a point, as poverty Dresses harder - seem to reach a oeiling, perhaps associated with physical ill-effects from extreme poverty, above whioh further rises with ultra-poverty are very small, absent, or negative. Some of these variables help oause or maintain poverty; some of them (not necessarily different ones), poverty helps to oause or maintain; and some are comon causes or consequenoes, alongside poverty, of different variables. That is one reason why we seek characteristics, not causes, of poverty and ultra-poverty. But DRa do conform 45 to a pattern of observations that the ultra-poor in LICs, while plainly not an underclass, are "different from us" and also from the moderately poor. Only for Gujarat is there substantial evidenoe on whether these dramatic differences in age-structure, as between poor and poorest and others, are sex-speoific. In rural areas, this seems not to be signfioantly the oase. Boys (age 0-14) comprised 50.2S of males in the poorest deoile of rural GuJarat's homes in 1972-3, and girls 51.0% of females; in the top decile the corresponding proportions were 32.2% and 32.3% At the other end of the rural lifespan, too, the proportions of males and females aged over 60 were similar in any given MEP decile (though as usual old women greatly outnumbered old men in each deoile); sex-selectivity did not aocount for the much greater presenoe of old persons in better-off rural households (Visaria, 1977, Tables 1-2). In urban Oujarat, espeoially among the poorest MEP decile of households, males clearly contributed more than females to the very high conoentration of the "young-end" dependency burden upon poorer households: 51.8% of males in the poorest deoile of households were under 14 (37.1% being aged 5-14) as against 48.3% of females (33.5% aged 5-14); in the richest deoile, only 19.8% of males, but 25.6% of females, were aged under 14 (ibid., Table 2). This is an interesting pendant to the finding (pp. 48-50) that, in muoh of India and perhaps in other LICs, the huge exoess of men over women in the towns deoreases as MEP falls, and is reversed in the poorest deoile. The Gujarat data do not suggest that the poorest seleotively deprive small girls of care or food. If suoh deprivation killa, it is almost certainly at ages 0-4 (where alone it is evident in Bangladesh: Chen et al., 1981). Yet, despite the (world-wide) excess of male births, in both rural and urban areas the proportion of lowest-decile females aged 0-4 is greater than the proportion of lowest-deoile males. It would be valuable to see data for other States (e.g. Maharashtra, where rural-urban migration and return migration, and severely unbalanoed sex-ratios, especially in specific age-groups, are more important than in Gujarat), and for LICs other than India. Visaria's Asian data (1980, Table 4) suggest that this DR gap between poorest, poor and others will, unless attaoked by specific policies, worsen with development. (This is not to say that the poor will get absolutely poorer; only that a worsening dependency ratio will provide an inoreasingly severe obstacle to their advance relative to the less-poor). First, the gap is larger in towns, both because urban sex-ratios beoome more balanood as poverty increases (p. 48) sO that more children can be prooreated per adult, and because the return of older persons to the villages for retirement reduces the DR of the rich relative to the poor (who have a much smaller proportion of older persons). So, as urbanisation prooeeds, people move to places where (a) the rich-poor gap in the DR is bigger, and (b) the impaot of growth, in widening the gap, is greater. Second, less-poor areas of Asia as a whole - Taiwan, Malaysia, even urban India and Nepal - tend to feature muoh greater inter-deoile differences in DRs than some poorer places such as rural Gujarat and Maharashtra, and Sri Lanka (Visaria, 1980, p.65); development may well be widening the DR gap. 46 While age-composition and DRs of households are quite systematically related to poverty in LICs today, the curve relating age of HH (x) to risk of poverty (y) is very obscure. If the only factors at work were decisions on age of marriage (household formation) and death-rates of partners, a simple curve - monotonic decreasing - could be expected, because poorer people clearly tend to marry younger in LICs now, 31/ while adult life expectancy is somewhat (not very much) lower among poorer groups, leaving very old HHs accounting for rather smaller proportions of the poor than of the rich. If we could trace the marriage of an LIC couple, "average" in respect of level of poverty and the partners' ages upon marriage, through the risk of poverty, a more complicated but still clear-cut curve --_______ - shaped - can be inferred from family-cycle considerations. On marriage, both partners often can work, and sometimes there are wedding-gifts to enhance living-standards. Soon, poverty-risk rises as small children arrive, cutting per-person income by sharing what the household receives, and sometimes by requiring child care and stopping the mother from earning, 32/ cutting total household income too. Later, poverty-risk is cut, both by inheritance and as children reach earning age-groups. Later still, poverty-risk rises: the oouple is likelier to get ill, and earning children leave home (and maybe fail to remit money). Almost exactly this curve is found in a big 1974 Malaysian sample (Datta and Meerman, 1980, p. 13). Is the HH-age to poverty-risk function a sum of -- and effects? Unfortunately it is more complicated. The latter curve related to an "average" LIC couple in respect of marriage-age and poverty at marriage. A poorer couple at marriage - which tends to marry earlier - follows the latter curve, but (apart from, obviously, showing a higher probability of poverty than a richer couple) has fluctuations of a different amplitude. Poor households' poverty risk as a proportion of their lifetime average (as against less poor households' proportion) is lowered immediately after marriage by relatively high female participation rates (poorer women are much likelier to seek work: Lipton, 1983a) and correspondingly is raised as small children arrive - because (i) more participation is desired, and hence frustrated, than for better-off women; (ii) extended-family help in child-care is less available (pp. 30-1); (iii) for the poorest, high death-rates for under-fives (pp. 15-16) raise the ratio of younger children needing maternal care to older children contributing to work. The comonly explored linear relationships between poverty-risk (or MEP) and age of HH - sometimes with an added element for the square of age of HH (e.g. Gaiha, 1983) - therefore, not surprisingly, produce no clear-cut results. A logit analysis of 4105 rural Indian households in 1968-69, using a (somewhat high) poverty-line of Rs.355 income-per-head per year, suggested that the logarithm of risk of falling below it rose by .067 (significant at 2.5%) for each rise of one year in HH age. 3U/ However, the corresponding coefficients for the 2350 cultivating households showed inconsistent results in different years, neither significant at 2.5% (.0697 in 1968, sig. at 5%; -.0106 in 1970, n.s.). This also applied to the 900-odd oasual-labor households (Gaiha, 1983, Tables 13, 18, 19). We have seen robust relationships between poverty risk and age-structures, MRS and child/adult ratios (e.g. Tables 8, 12). However, Gaiha's large, disaggregable, three-year Indian rural sample suggests that 47 orude linear regressions of poverty risk on HH age - with or without ageg - are unstable and weak. Gaiha's data set is more useful in exploring poverty risk by five-year HH-age groups (Gaiha and Kaymi, 1982, pp.6, 57). In 1968-69, 61.01 of the 4105 households were "poor" (income-per-head below Rs.29.6). Only seven such households (and eight others) had HH under 20, but for older five-year age-groups the cell size was substantial; the proportions with income-per-head below Rs.29.6 were: HH aged 20-24, 66.9%; 25-29, 69.6%; 30-34, 65.6%; 35-39, 68.0%; 40-44, 69.0%; 45-49, 70.1%; 50-54, 66.9%; 55-59, 62.5%; 60-64, 57.5%; and 65-69, 54.8%. The pattern that 'the bottom deciles include a higher-than-average proportion of middle-aged heads of households' is confirmed for several South Asian samples - probably (Visaria, 1980, p.53) because 4HS is usually highest for iiddle-aged HHs. Two sidelights on these data are available.. From Malaysia - one of the several countries where Visaria confirmed that risk of poverty (y) was related to age of HH (x) in a ----N-E -curve - another very oareful survey analysis instead revealed a simpler e-~- -curve, but it excluded one-person households and did not disaggregate the over-fifties (Hazumdar, 1981, pp.3, 36-9); this suggests that the downturn in poverty-risk among older HHs owes much to the over-60s, and/or to single persons who can live alone as HHs (instead of with younger, earning HHs) only if better-off. From rural Botswana, where male migration to the S. African mines leaves many female HHs, another careful survey (for 1974-75) suggests that the downturn in poverty-risk is closely linked to faster asset accretion by male HHs. "Households headed by women under 30 earn as much as households headed by men under 30. However, beyond that age incomes in male-headed households rise substantially, while those in female-headed households decline" (Kossoudji and Mueller, 1980, p.12). The latter finding has an anafogue from an otherwise totally different "developing" environment, that of Belo Horizonte, Brazil; poverty in female-headed households, as compared with male-headed, is muoh more heavily ooncentrated where the HH is of prime age (15-59), although poor prime-age households tend to have lower DRs if female-headed (Merrick and Sohmink, 1982, pp.303-4). Probably - in urban Brazil as in rural Botswana - it is the barriers and diffioulties faced by women in aecess - to post-primary education, and later to eapital and jobs - that explain growing female disadvantage as mHs after age 30. The relationship of age of HH to household poverty is complex, but of enormous polioy importance. Peak incidence of poverty among HHs in mid-span, say 35-45, probably means most conoentration of damage on small children, where health and nutrition are likeliest to suffer irreversibly. (Indeed, this is also implied in the strong positive link of poverty to DRs; see pp. 43-5.) Second, it also makes for relatively good chances of remedy: households at greatest risk are headed by persons of prime working age (although their oapacity to work may be limited by ohild-oare and high DNs). Third, muoh of the age-poverty pattern Is not life-eyolioal, but is due to poor people's earlier marriage and lesser access to post-primary education (from whioh benefits aocrue mostly in later life), and to the need for desperately poor old persons to cease being HHs, and to join younger Hms to survive; enemies of poverty cannot, therefore, happily assume that - because poverty riak falls after HH age 45-55 - partioular households "grow out of poverty". Finally, the sex distribution among HHs of asset aocretion is uneven, so that male-headed and female-headed households may show quite 48 different links between poverty and HH age. (c) Are women poorer? Much of the reoent literature of development (for instance Buvinic et al., 1981) gives a fairly uniform picture of female disadvantage. Elsewhere in these papers, a much more mixed position has been suggested. Selective, dangerous undernutrition, relative to males in the same households, seems to be confined to Bangladesh and Northern India (Lipton, 1983). Labor-market discrimination against women (i) reduces their access to (and training for) work with better earnings prospects, (ii) is expressed in a wider sense through the societal pressures on women to work a "double day" at home and at work, but (ill) does not, in most LIC conditions, take the form of much less reward for identical work (Lipton, 1983a). In respect of the two main issues affeoting female demography and poverty - the incidence of women in poorer groups, and the problem faced by female-headed households - the data also suggest something other than generalised disadvantage. Women outnumber men significantly not in poor households as a whole, but only in the very poorest 1-5%, whioh are so poor owing to high child/adult ratios (atypical of female-headed households, even poor ones) or a high incidence of unsupported widowhood; and to some extent in poorer urban households. The respective remedies are to provide women with means and power to limit births, and to ease their path to effective migration decisions. Visaria's analysis of decile-wise data for Maharashtra and Gujarat in 1972-3 showed that the ratio of women to men aged 15-59 "showed no clear tendenoy to rise with tMEP] decile in rural areas", but did so in towns (Visaria, 1977, p.14). However, in the poorest, second, and third deciles of urban households, this ratio was respectively 0.993, 1.022 and 1.029 in Gujarat; and 0.995, 1.033 and 1.070 in Maharashtra (ibid., pp.8A-8B). Very slightlv more urban men than women aged 15-59 were ultra-Door or poor. The real problem was that women's chances of escape - either from rural poverty via townward migration, or from urban poverty via upward mobility - were far less than men's. Thus, by the time the top urban decile was reaohed, the urban male-female ratio was 1.171 in Gujarat, and 1.437 in Maharashtra (ibid., pp. 8A-8B). The effects of this on the poor (a) damage women through migratory and child-related processes rather than through discrimination against localized women with given child/women ratios, (b) are political, rather than overtly reflected in economic outcomes. Before dealing with them, I look at four other issues. First, is India's apparent "equality in poverty" of the sexes - their near-balance in the poorest deciles - paralleled in other areas of South Asia, and in other LICs? Soeond, does the appearance result from looking only at persons aged 15-59? Related to this issue is the question of whether single, divorced or widow/widower status - or household size - affects poverty-risk differently for men and for women. Finally, if these considerations confirm that any over-representation of women among the poorest MEP deciles appears to be relatively small in most LICs (Visaria, 1980, p.62, for Asian evidence), "examination of data for smaller groups of households ... might show a different picture" (ibid., p. 63): does it? 49 The first two questions are addressed Jointly by Visaria (1980, pp. 58, 61). His nine South Asian samples for 1969-75 all show the proportion of females of all ages 34/ in eaoh household MNP deoile.35/ This proportion falls signifioantly 36/ with rising affluence only in urban Maharashtra (from 50% in the poorest decile to 41% in the riohest, in 1972-3) and urban Nepal (51% to 42% in the 11-town sample for 1973-4, and 52% to 36% in the 7-town sample for 1974-5). The really important point, however, is that once more there is no signfioant over-representation of females in the poorer groups - the peroentage even in the bottom docile of households is below 53% in all nine samples. Females of all ages In these South Asian oases - like adult women in Western India (p. 48) - do not comprise unduly high proportions of the urban or rural poor, but are much less likely than the male poor to beomoe urban and affluent. We have little evidence from outside South Asia, but nutritional (Schofield, 1979) and other reports confirm that women are less likely to be over-represented among the poor in non-Asian LICs. There are some caveats in respeot of the statement that poor households in most LICs do not show female/male ratios muoh above unity. Reverting to the third of our "issues" (p. 48), divoroed or widowed status appears to worsen women's poverty prospeots more than men's, espeoially for Hsa (p. 52). This may well apply with special force to cultures where husbands can, with relative ease, divorce or abandon their wives; in two villages in Bangladesh, females oomprised 57% and 52% of landless households, but 47% and 45% respectively of the "self-sufficient" (better-off landed) households (Mahmud and McIntosh, 1980, p. 504), and the role of widowhood in suoh oases is well documented (e.g. Cain, 1981). Households without adult couPles, however, may be linked to male as well as female poverty: there ia aome evidence from rural India that households with adult male/female ratios very far from unity in either direction are exoeptionally prone to poverty (Oaiha and Kazmi, 1982, p. 23). As for the final issue raised on p. 48, the data confirm the hunch (Visaria, 1980, p. 63) that "smaller groups of households than deoiles might show a different picture". Table 9 shows quite substantial "surpluses" of women among the poorest 3-5 peroent of both urban and rural Indiana. This is oonsistent with the (much smaller) State samples. Women's hard life, and Its link to low-KR householda, cannot however be assessed by that NBP alone. Poorer groups of households (Table 9) have a much higher ratio of children to adult women. Female levels of well-being are thus lowered among the poor, in ways not revealed by NLP (and increasingly as poverty bites harder), but the need to ohild-mind as well as earnlng - a need made more oneroua by higher direot economic participation among the poor (Lipton, 1983a) - mean more frequent "double days" of housework and earnings. Women'a well-being is also more likely to be lowered, as poverty deepona, by numerous pregnanoies (see Harrington, 1982, on Nigeria), all too often ending in stillbirths or infant deaths. Both these facts are hinted at by Table 9. In the poorest fifth of Indian households, there were almost two ohildren per adult women - far above the average. This is confirmed not only at State level (Table 9) but also by four village samples in Gujarat in 1970-71. The poorest 18 households (100 persons) did not "overrepresent women" (there were, indeed, 1.08 adult males 50 per adult female), but imposed on each woman 1.92 children. The best-ott 19 households (101 persons) had a larger share in the local "surplus" of adult males - 1.40 per adult female - but only 0.96 children per adult woman (V. Patel, 1973, p. ix). In the poorer households, then, a high ohild/female ratio, rather than (except in the very poorest 5% or so) female over-representation, is the main female-specific "burden of poverty". At national and political level, the data once more suggest that it is not female surplus among the poor, but male surplus among the decision-making and educated urban rich, that constitutes women's "poverty problem". In 1972-3, in urban Maharashtra, there were 127 men aged 15-59 per 100 women in the second-richest household quintile, 138 in the second-richest decile, and 144 in the richest - and presumably most influential - deoile (Visaria, 1977, p. 8B). 3/ This has several important effects. First, men dominate politics, especially in LICs; probably, the more localized and "economic" the issues, the greater the male dominance. Thus, though there is not a big female surplus among the poor, the concentration of urban men among the better-orff and of urban women among the poor, increases the relative weakness of the poor and of the women alike. Second, the dominance of the urban sector in political decision-making is strengthened by the overlap of maleness with higher incomes in the cities. Third, the higher male/female ratios among the top deciles, especially but not only in the town, increase the dominance of the rich. Fourth, access to education is, notoriously, less among women, especially in remote and rural places (see, for example, Deijomeah and Anusionwu, 1979, p. 52; Visaria, 1980, p. 75). Insofar as women concentrate in rural and poorer urban groups, there is mutual reinforcement among the dominance of the educated, of the male, of the urban, and of the rich. All four effects are probably understated by Table 9, because the sex-ratio in a particular area among small children is usually near unity in all income-groups and places, so that the dominance of males in adult urban elites is even greater than the figures of Table 9 suggest (see fn. 37). Tables 2 and 9 together show that in India the child/adult ratio rises much more rapidly, with deepening poverty, than the female/male ratio - which rises at all only in urban areas, and even there only a little, if at all, above unity. This implies a sharp rise in child/female ratios with poverty. We have discussed above the implications for adult women. But the importance of this "characteristic of the poor" extends beyond one generation, and to both sexes: to the social inheritance of disadvantage in "cycles of poverty", via inevitably inadequate child stimulation and care. Even more directly, income earned outside the home by rural women makes much less contribution to children's nutritional status than does own-farm income, given total MEP (Kumar, 1977). Similarly for urban areas: in 1969 Berg implicitly showed direct damage to child nutrition, given NEP, from female faotory-work in urban Calcutta (Reutlinger and Selowsky, 1976). Yet it is poorer women who have higher participation rates; who, even at any given rate, are likelier (because they own few productive assets) to be pushed into work outside the home (Lipton, 1983a); and who neither can afford domestio employees, nor are likely to have access to complex household structures (pp. 29-31), to help with children. It is such poor women who, as we now see, typioally each have more children; these are likely to "inherit" the losses 51 consequent on their mother's more probably uncovered absencea. The orucial importance, to ohild welfare in poor households, of appropriate income-earning tasks, rewards and conditions for working mothers in the home - tasks often glibly and uniformly dismissed as exploitative - is obvious in.the light of such data as those of Table 9, and will be reoonsidered in a brief polioy disoussion in Section V. We have seen that poorer urban households usually have a near-balance sex-ratio. However (Connell et al., 1976), the huge excesa of young men in Asian and African rural-urban migrant streams (and hence in urban family-farming age-groups) is also familiar. The oonsistenoy of these two apparently contradictory facts - via great male predominance in rioher urban groups; and oonsistently also with relatively high child/female ratios in the poorer groups, whose women and espeoially children are most damaged by a high ratio - effeotively oonstitutes a new demography of urban poverty. Poor people in LIC towns are ooming to show adult male/female ratios around, or even below, unity; the high overall urban ratios are due to very big male surpluses among the better-off. Partly, as work in Bombay (Joshi, 1976, esp. p. 1303) suggests, this is because poor women have oome to form a rising proportion of urbanising migrants. Partly it is because the urban "ladders to success" favor men (in Botawana, women's disadvantage as migrants rests entirely in their worse urban prospeots of employment and wage-levelsa and if these are held oonstant women's propensity to move exoeeds men's: Lucas, 1982, p. 17). Partly, it is because failed migrants tend to return to the village of origin (Connell et al., 1976, pp. 126-8). Whatever the reason, the effeots of the new demography of urban poverty upon women are not, for the most part, direot via MEP. They operate at national level, by oonoentrating male and urban Dolitioal advantage; at household level, by placing most ohildren in poor families where women (and men) oan least afford to stay at home and care for them; and, in both ways, by transmitting female disadvantage (which is not crude absolute MBP disadvantage, for sex-ratios among the poor, exoept in the very poorest 5% of households, are usually not very unbalanoed) from the poorer 20-25% of mothers to both daughters and sons. These effects are all felt in rural areas too, but they are perhaps strongest in cities. 3a8/ a * S Are women in LICs - for social, eoonomic or other reasons - likelier to be poor because they are in some sense demographic victims? The idea oannot be simply supported by the available NEP data: sex-ratios are around unity in most poor groups. However, the idea receives strong support from subtler demographic evidence linked with those same data (elite concentration of urban males; high child/female ratios for poor groups). Surprisingly similar conclusions apply to the question of whether female HHs suffer special disadvantages. Contrary to some interpretations, such households seem to suffer little, if any, greater risk of very low HEP, when appropriate "other things" are allowed for or held oonatant. But female HHs, like females in general, do appear to suffer poverty-linked disadvantages not simply associated with lower statio MIP itselft less chance of raiing MEP with time, less adaptable household sizes and structures, more drawbaoka from (and likelihood of) "poverty-fixing" form of oivil status. 52 Visaria notes oa small over-representation of females among HHls in the bottom one or two deoiles in urban [Western India], Sri Tanka, Taiwan in 1968 [but not 1974] and Peninsular Malaysia; but ... not ... suoh as would justify widespread oonoern". This, if anything, overstates the importanoe of female-headedness as a oorrelate of poverty. The relationship of sex of HH to poverty is extremely weak, even in these oases. In urban Haharashtra, for example, 10.4% of HEH in the poorest deoile were female, but so were 10.7% in the rioheat, and an average of 8.0% overall. In Malaysia (1973), 20.4% of poorest-deoile households were female-headed, but 16.8% of seoond-poorest, 19.0% of seoond-riohest, and 18.1% overall (Visaria, 1980, pp. 54-5), and another large survey found that average inoomes per person "by sex of household head are identioal' as between male and female HHs (Datta and Meerman, 1980, p. 8) so that any greater inoidenoe of poverty among the latter would imply worse distribution among female HHs, whioh there is no obvious reason to expect. Suoh sex differentials in poverty as do exist between HHs 39/ are dynamio, not statio. In partioular, they are due to the risk of widowed status. 'Only 25% of all Indian males over 60 ... are without wives ... over 70% of all Indian women over 60 are widows" (Visaria, 1980); this is attributed (Nugent and Walther, 1981, p. 34) to "higher remarriage rates for husbands ... and ... sizeable age differentials ... at first marriage', but the universally greater life expeotanoy of women at age 60 must play a major role. Women thus run muoh greater risks of widowed status - and hence, especially in the poorer household with its lower inoidence of oomplexity, of depending on their sons' support - than do men; also, because widowers are likelier than widows to own property, sons have less self-interest in supporting a mother than a father after bereavement. Cain points to the sharp oontrast between Islamio Bangladesh and India, in respect of the very severe risk of desoent into destitution associated with widowhood in the former (Cain, 1981, pp. 458-9). However, in a largely Hindu study area in Karnataka, "the situation of a widow with no surviving sons oan be so bad that one somtimes wonders that any woman would restriot her fertility and oourt even a sm1ll ohanoe of suoh a fate ... Twelve widows live on their own in the study area (5 per oent of all widows), some in miserable oonditions" (Caldwell et al.9 1982, p. 28). It may well be, then, (i) that the differential risk and effeot of widowed status explaina most of suoh disadvantage of female HHs as oeists; (ii) that high fertility among the poor, partly to reduce the risks inherent for them in suoh status, is a major indiroot effeot of Dotential femle-headednesst an effoot by whioh poverty is deepened, not for widow-headed households as suoh, but for all poor households - mostly male-headed. Data from Botana again stress the dynamios of female-headedness - its association over time with other poverty-related phenomena. 'Households headed by womn under 30 earn as muoh [per CU] as households headed by males under 30.' However, beyond that age incomes per CU in the latter 'rise substantially while those in femle-headed households deoline' (Kossoudji and Mueller, 1980, pp. 12, 14). Several faotors are probably at work heres differential inoidenoe and impaot of widowhood (see above) and also of abandoment and divorce; capacity to deoumulate savings out of past incomes from (seleotively male) migration; and, perhaps above all, the effect of se-bias In inheritance, education, and in other patterns of asset accretion. Animal husbandry contributed 47% of income for working male HHs, but only 32% 53 for female HHs with no working-age male in the household. Cattle value per CU, respectively, was 261 Rand and 138 Rand (ibid., pp. 16, 24). In general, female HHo as such may not have much greater poverty risk than male HHs. But they "ought" to have much lower risk - for they tend to have smaller households, lower child/adult ratios, and greater HH age. Indeed, the demographio evidence (like that of other papers in this series) does not deny that women, and female HHs, suffer special disadvantages associated with poverty - notably the large part of life that poor women spend in pregnancy and childbirth, often for children who die in infancy (Harrington, 1982). However, women's disadvantages are iomobility, laok of access, and hence more severe and less easily remediable effeots of poverty - in job search, in migration, in food behaviour, and much else. To interpret these disadvantages as a much greater, specifically female, risk of poverty at each (static) moment is to neglect both the politics and the dynamios of the problem. It is also usually incorrect. (d) Life-cycles. poverty. and household eomDosition In Section II(j) we asked whether domestic cycles might help to explain the links between MHS, poverty, and status. Here, we lookc at these cycles as possible "linkers" of poverty to age-structure and sex-structure. Do the same households (for example) at one time show high child/female ratios associated with high poverty risk, and at another time the low ratios associated with low risk? A given percentage shortfall, behind some poverty line, is less "inequitable" if it is experienced by (say) 80 percent of all households for 10 percent of the time when a particular person is HH - normally the 10% when under-fives are the largest proportion of household members - rather than by 10 percent of households during 80 percent of a particular headship. 40/ Unfortunately, any approach to such questions is hampered by shortage of long-term panel data. Cross-seotion data are almost always ambiguous. For instance, household income-per-person falls in Malaysia as the HH0s age rises from 25 to 37.5 and again from 50 to 62.5 (Datta and Meerman, 1980, p.13). This may be because the poor marry later (or separate sooner) within these intervals; for life-cycle reasons; or because of changing economio oonditions, e.g. permitting different savings rates over time. Moreover, most survey data, one-shot or panel, are oollected at the level of households or of individuals, but not at both levels. Yet life-cycles that alter a household's age and sex composition eomonly detach individuals from, or attaoh them to, pre-existing householda. Bspeoially with one-shot surveys, but even to some extent with panel data, such processes are unlikely to be "caught" unless surveys are conducted at both individual and household levels. To some extent, however, surveyed poverty - while understated by the timing, looation and prooedures of moat surveys 41/ - is overstated beoause life-cycle factors, of both MHS and composition, are ignored. Survey estimates of one-period "proportions in poverty" (and Son indicess Son, 1981, pp. 35-8) are maximum estimates of the impact of lifetime'poverty, and normally overstate that impact. Suppose a population contains 50,000 persons, 54 of constant age- and sex-structure in eaoh MEP group. Suppose that ten suocessive yearly surveys all place 10,000 persons below a (constant) absolute poverty line; that, in each survey, the MEP gaps between the poverty line and the lowest, second-lowest ... 10,000th-lowest MEP are all identical; and that the impact of a person's age and sex, upon his risk either of being in poverty at all or of being above a particular MEP distance beneath the poverty line, is identical at each of the ten surveys. Even then, the "head-count" proportion of persons in absolute poverty (in fact 20%) - and the Sen index (Sen, 1981, pp. 35-8) of poverty - calculated from each of the ten surveys would, though both were unchanged in ten surveys, both be maximum estimates. Unless every person, if and only if ever among the poorest 20% of people, were alwaya among the poorest 20%, the head-count overestimates poverty incidence. Unless the ranking of the poorest 20% never varies - not easy even to define rigorously in a growing, or otherwise changing, population - the Sen. index overestimates welfare impact on adults. The main thing leading to changes in the composition of persons below the poverty line, and to the ranking of those who remain below it, between surveys is the life-cycle. This fact, indeed, points up the seriousness of even transitory poverty: to say that a one-shot survey (because it counts households as poor even if they are below some critical MEP level only for the reference month or year before survey) maximally estimates the welfare impact on adults is not to play down poverty - particularly if life-cycle poverty homes in on families just when they include two or three nutritionally vulnerable under-fives. Policy-makers need to know how much poverty is a transitional and life-cyclic phenomenon, not because such poverty in any sense does not count (though standard welfare economics correctly insists that misery shared - if survived - is revealed preferred to "the same" misery concentrated 42/), but because different policies are likely to be cost-effective against (i) high life-long poverty-risk for few and (ii) poverty risk that rises sharply, for many, in particular phases of the life-cyole. The policies for (i) and for (Ui) differ even more sharply, if the selection, among individuals at risk, of those in fact struck by poverty depends on contingenoies other than the life-cycle itself, such as illnesses or bad harvests. We revert to these policy issues in Seo. V. What evidence do we have about the importance of life-cycles in total poverty in LICs? Most disoussions seem to assume that cross-section measures of age of HH, as a function of MIP (or, better, income per CU), allow us to infer the pattern of poverty as people get older. Even leaving aside the facts that most people are never HHs (especially in LICs, where 1 in 4 to 7 persons dies before adulthood), and that even HHs do not all enter or leave hardship at the same age, we have seen that such inferences are not feasible (p. 53). A much more careful piece of inference suggests that for Colombia - a middle-income country, for which wages were muoh the main income source - the oontribution to inequality (and henoe, given average inoome-per-person, to poverty) made by age-wage funotions was proportionately muoh less than in the USA, and that made by wage differenoes within a given age-group correspondingly much more (Schultz, 1981a, esp. pp. 12, 15). * * I The hypothesis suggested by Indian village data is that lite-cycles 55 are imDortant contributors to poverty to the extent that aooess to better-Daid work or to assets is aohieved rather than asoribed. The evidenoe is that - in all, and only, those survey villages where people with low inoones (or in groups, e.g. laborer oastes, likely to have low incomes) in the reference period would maintain MEP well above monthly income-per-person - caste "assignment" of menial work and landlessneas was relatively lax. Now why should anybody be prepared to lend to a very poor laborer? Presumably - sinoe "bonded labor" is seldom prevalent or enforoeable (P. Hill, 1982) - repayment usually depends on either the borrower's oollateral or his prospects. Collateral implies assets inherited, or accumulated (probably by non-menial work), in the past; prospects imply a chance of more assets, or better-paid work, in the future. Hence we should expeot lifelong poverty, refleoted in lack of creditworthiness in bad times, to aocompany oaste (or class) rigidity in acoess to the sources of wealth - good jobs, land. If correct, this also implies that life-cycle poverty will tend to be less extreme even per unit of time than lifelong poverty, beeause life-cyole poverty, being more strongly associated with a better past or future oondition, oan be partly alleviated by borrowing, whereas a norm that low income-per-person is lifelong undermines the chances of that "person" to borrow, and hence to show HEP ahead of his income. To test this, we should juxtapose inoome and consumption expenditure per person, in various "groups" according to inoome per person, occupation and sooial group. Our expeotation is that the better-off will show income above consumption expenditure; that the moderately poor (easpecially if landowners) will show the reverse; and that the worst off seek to run down savings, or to borrow against presumably acoumulated collateral - or future prospects to earn and repay. To the extent that these attempts succeed, life-oycle faotors are associated with - and modify the impaot of - their extreme ourrent poverty: they saved out of past income, or can borrow against future expeotations. Unfortunately there are data problems. II/ Only one - apparently reliable - study shows how deficita, of income below oonaumption expenditure, are ooncentrated within the various inoome-groups. Here, 94 households (488 persons) in four Gujarat villages are placed in five groups, by (1970-1) inoome-per-person. The poorest 18 households (100 persons), with Rs. 149 per person, were reoorded as spending Rs. 348 per person on consumption - 2.3 times inoome: still a low enough outlay to leave them somewhat below the then ourrent poverty line, i.e. for us to oonclude that they were at some risk of being undernourished 44/ (V. Patel, 1973, App., p. ix, Table 4). Only slightly less extreme is the position of the "poor but not poorest": 210 persons (in 44 households) with average income of Rs. 283, yet average outlay of Rs. 417. Suoh gaps, with this careful study, oannot be explained by the usual (small) overstatements of expenditure and/or understatements of income. Almost certainly, both groups of households are either deoumulating past savings or stooks, or borrowing against future labor or assets. In either case a life-cycle pattern is indioated. Especially does this have to be the oase for the subsets within the poverty groups upon whom, as this study shows, the deficits were heavily concentrated. Eight of the poorest 18 households (by income-per-peroon) somehow managed to consume about Rs. 2500 per year eaoh, over and above recorded household income averaging about Rs. 8001 Those 8 (of 18), and 12 of 56 the next-poorest 44, accounted for the total defioits of their respective household inoome groups (V. Patel, 1973, App. Tables 4-5, pp. x-ix). Another Gujarat study (of the 268 households, and 1533 persons, in Ankodia village in 1960-1) shows similar large deficits. The 23 households (81 persons) with income below Rs. 300 averaged Rs. 61.4 of income per person, but spent on consumption Rs. 153.7 per person. The next 68 households (289 persons) averaged Rs. 112.6 of income per person yet Rs. 176.6 of outlay per person (R. Patel, 1964, pp. 178-9). We cannot acourately identify the per-household income ranking with a genuine poverty-ranking (Seo. IV(c)). However, these figures are suggestive - especially as the 23 lowest-inoome households, presumably including many of the poorest even in MEP terms, were on average actually increasing assets-per-household significantly during the reference year (ibid., p. 203). Life-cycle behavior, in respect of income-expenditure gaps, may well be closely linked to land inheritance. In the four-village study in Gujarat, the 22 households (103 persons) subsisting mainly from "agricultural labor without land" - though earning only Rs. 260 per person, just 60% or the village average - were spending, on consumption, 2.4 times their annual income (as against 1.2 times for the village as whole). Landless farm-workers, receiving 12.7% of village income, were incurring 29.4% of village net deficit, although they were among the poorest villagers (V. Patel, 1973, App. Tables 1-2). Apart from the probable rarity of bonded labor (P. Hill, 1982), unless the survey year was far below average not much of this deficit can be explained by increases in bonded labor. Either young farm laborers - i.e. persons with farm labor as the main source of income in 1970-1 - were borrowing, perhaps from parents, in expectation of land inheritance; or old ones, having passed on land to children, were living partly off past savings. It is not plausible that "lifetime assetless" persons, whether or not in a particular and usually landless caste, could overspend like this. But can we find direct evidence to link life-cycle poverty, borrowing, and less ascriptive land and job assignments? Further light is cast on the "occupational life-cycle" and the deficit by two further studies for Gujarat - Afawa in 1961-2 (R. Patel, 1966, pp. 150, 157), Oon in 1963 (N. Shah, 1968, p. 164); and by two for Rajasthan - Hasteda in 1964-5 (M. Desai, 1966, Tables IX-18 and X-2) and Dingri in 1963-4 (Saxena'and Charan, 1973, Tables VIII-18 and IX-2). One village reported a slight overall surplus of income over consumption (Oon, 1.1% of income) and two reported slight overall deficits (1.1% of income for Dingri and 1.9% for Afawa) - both probably reflecting the usual income understatement and outlay overstatement; Hasteda's overall deficit, 8.3%, probably also indicates a bad year. In Atawa, as expected, laborers (156 households, 649 persons) as a whole - averaging only 40% of village outlay-per-person, and all belonging to "ex-untouchable" oastes - could not incur deficits, and indeed reported oonsumption about 20% below their miserable incomes (B. Patel, 1966, p. 97 and Table I-2). In Dingri, though with less inequality, the (proportionately fewer) laborers also showed a slight surplus, i.e. appear to have been unable to borrow; once again, all the 20 laboring households (and only one of the 66 village households engaged in other occupations) belonged to the scheduled imas oaste (Saxena and Charan, 1973, p. 30). Oon performed similarly: of the village's 189 households, the 65 laboring families (336 out of 1139 persons) consumed 4-5% below income, and were heavily ooncentrated among scheduled 57 oastes, soheduled tribos, and non-Hindus. In Hasteda, however, the situation in the tour grouped villages (p. 55) recurs: the laborer households are oonsuming 57% above their inoomes. Consistently with the hypothesis (P. 53-4) the overlap between caste and oocupation is muoh weaker: of the 54 laborer households (268 persons) only 15 belong to the sohoduled oastes and 2 to the soheduled tribes, and 9 are Muslims (M. Desai, 1966, p. 19). Even with tentative aupport from other data, these observations from four Gujarat villages - observations relating to caste averaae income and expenditure per household only - oannot be more than hints. However, they do suggest that oollateral acquired from past income, or the epeootation of future inoome from assets (usually land, or better jobs), allows very poor workers - even if ourrently landless or near-landless - to borrow or dissave to the extent that asset ownership (and/or type of work) is linked to age-cohort and/or life period of a person and/or his or her parents - but not to membership of an "unalterable" group (such as a sex or rao or caste) to which poor persons do not belong. In effect, rigorous asoription of social status, if it severely oonstrains asset acquisition and mobility towards better-rewarded work, prevents the "lower" groups from following the trajeotory correctly predloted by the "life-oycle hypothesis" on the determinants of personal savings/inoome ratios in developed sooieties (Modigliani and Brumberg, 1954). People who are ascribed a poverty-linked status tend to be poor life-long. This is partly because they are less able to adapt to "good" phases of the life-cyole by repaying or saving or to "bad" phases by borrowing or dissaving. Hence the life-oycle element in poverty tends to explain a greater portion of poverty in developed than in developing sooietieo, to the extent that the latter are more orientated to asoription and the former to achievement (Parsons and Shils, 1951; Hoselitz, 1968, p. 425). This explains, in part, the US-Colombian oontrast found (p. 54) by Schultz. It also means that the lifetime incidence and impaot of poverty and inequality - while everywhere "less" than the maximal estimates in one-off surveys (fi. 41) - are closer to those estimates in more rigid and immobile sooieties, where the poor are likelier to stay poor. 58 IV. DEMOGRAPHY AND POVERTY HMASUREIENT (a) The ohoice-of-denominator problem To decide whether a household is poor, and how poor it is, we ask whether, and how far, there is a shortfall below a critical level. But in what? Outlay (or income) 4L5/ per person, per oonsumer unit, or per household? So far, we have assumed that the CU is the best denominator; that per-person indicators are less good, but adequate (which is crucial, as they are much more readily available); and that per-household indicators are almost useless. On this assumption, we have established that several demographic indicators - e.g. large household size (in persons and to a slightly lesser extent in CUs), high child/adult ratios, and to a very small extent high rural female/male ratios - are correlated with poverty. We now have to look at this assumption about the denominator. All the above findings could well be invalid if per-household measures best indioated the risk of poverty. Indeed, findings about any "charaoteristics of poverty" always depend for their foree, and often for their validity, on the demographio denominator (outlay per what?) used to measure poverty. With "demographio characteristios of poverty" there is a ohicken-and-egg difficulty: we enquire, for example, whether size of household, i.e. person/household ratio, is linked to "poverty", but per-person (or per-household) denominators in defining poverty will push the answer in positive (negative) directions. An approach via equivalenoe scaling (pp. 64-6) would, if successful, avoid the problem of choosing among denominators; however, it cannot properly allow for the needs of the "extra" new-born child, and places demands on data that most LIC household surveys cannot meet. Again, if we knew which of the potential denominators was most closely correlated with some nutritional "touchstone" of ultra-poverty - e.g. being unable, with patterns of spending typical of households of a particular size and structure and total outlay, to meet even 80% of the average dietary energy requirements of the ages, sexes and activity-groups represented in the household - we could choose that denominator. However, once again, many data sets will not provide this information. We therefore need to ask how the ohoice of denominators should be made. In Seotion (b) we ask if the choice really matters. Are different persons classiired as poor on per-household, per-CU, and per-person measures? In Section (a) we ask: under what assumptions is each denominator most appropriate in deoiding who is poor and by how much? We then examine the main "problem assumption": absenoe of major economies of scale in oonsumption, whioh is assumed in rejecting per-household in favor of per-person indicators (Seo. (d)). (b) Choice of denominator makes a big difrerence Outlay per person ranks persons and households very differently from outlay per household. In ten South Asian data sets from 1968 to 1975 (Visaria, 1980, text table 2), only 14 to 28 percent of households - containing even smaller proportions of persons, 13 to 23 percent - were 59 assigned to the same deciles when ranked by per-household and per-person oriteria. The surveys showed that over two in three households - and persona - normally belonged to different auintiles by NEP and by total outlayl This great disparity between per-household and per-person rankings is systematio, not aocidental. Big households tend to have low outlay per person (pp. 8-11). But they also tend to have high total outlay. Hembership of high deciles by household size, therefore, is strongly oorrelated with membership of high MEP deoiles (Table 1) and there remains a signifioant, though muoh weaker, oorrelation at individual household level (Botswana, Government of, 1976, pp. 99-103). In household oross-seotiona, as HEP rises, household size rises more slowly, so that a weak positive relationship remains between a household's deoile rank in output per person and per household (Datta and Meerman, 1980, p. 4, fn. 2; Visaria, 1980a, p. 5, fn. 1). However, the negative size-MEP and positive size-outlay links mean that a low-rank deoile by household outlay oomprises a mixture of small households (some with quite large MEP) and low-MEP households (some with many household members). Tables 10 and 11 provide examples from Kerala and N. Nigeria. Apart from Visaria's South Asian work, several other studies show this tendency of smaller households to have lower total household income. The link is notable in Kenya (Collier and Lal, 1980, pp. 1, 39), Botawana (Botswana, Govt. of, 1975, pp. 99-103), and Malaya. In Malaya the result is that 38 percent of persons in the lowest quintile of households by income-per-household were not in the lowest quintile of households by inoome-per-person (Datta and Meerman, 1980, p. 5). Therefore, per-household and per-person denominators seldom identify anything remotely like the "same" people or households as poor. In defining poverty-lines, too, we would classify very different Droportions of households and persons as poor, with very different average intensities (and Sen indexes) of poverty, acoording to whether per-household or per-person outlay were used. a * 0 We are trying to answer the question: in what sort of households, plaeos, occupations, eto. is poverty most severe? That is, where are the proportions of persons below a poverty (or ultra-poverty) line greatest, and where do they fall furthest behind such a level? Many studies present only per-household outlay (or income), and can at best assess "adequaoy of MEP" by estimating whether suoh outlays would suffioe to maintain a given MEP for a household of average size and oomposition; this will be a bad assessment, since households with low MEP, low calories-per-CU, high ohild/adult ration, high IMR, etc., tend to show MHS and child/adult ratios. well above national or local averages. As we saw, rankings of households by outlay-per-household overlap very badly with rankings of households by outlay-per-CU. Fortunately, a big minority of surveys does report households, or at least ranked groups of households, by Der-Person outlay or inoome. Only rarely, however, do they report per-CU data, or give household composition sufficiently disaggregated to permit households ranked by per-person outlay to be re-ranked by per-CU outlays. It therefore becomes important to estimate how muoh better or worse at indioating the risk of poverty, under various circumstances, per-CU indicators are than per-person indioators (Seo.(o)). First, however, we can report that the count of households that are poor (or their ranking in order of poverty-risk) by MEP is not an intolerably bad guide 60 to the count (or ranking) by outlay-per-CU. It is bad enough to lead to some risk of error, but - unlike outlay-per-household as a guide to outlay-per-person - not so bad as to invalidate most inferences for groups from "poverty" on one criterion to "poverty" on the other. In each of Visaria's ten South Asian surveys, for example, household size could be calculated (i) in persons and (ii) in Indian-weighted 46/ adult-equivalents. Among surveys, the range of rils is small - 0.96 to 0.98 (Visaria, 1980, p. 201) - but all are very persuasive. In India, even among quite fine MEP groups, the ratio of CUs to persons in the average household (in a 1971-2 sample of 11,468 rural and 19,459 urban Indian households) - while tending to rise with MEP - varied extremely little: it averaged 0.788 among the poorest 3.9% of rural households, 0.794 for the next-poorest 10.5%, 0.793 (7.1%), 0.791 (10.2%), 0.797 (15.2%) ... 0.807 (best-off 4.2%); and in urban areas 0.790 (poorest 0.9%), 0.787 (next-poorest 3.7%), 0.793 (3.6%), 0.796 (6.0%), 0.800 (10.2%) ... 0.840 (best-off 17.0%) (Rao, 1979, p. 117). So it is not surprising that - in South Asian large-sample surveys in 1968-75 - in sharp contrast to the mere 26-40% of households (25-38% of persons) assigned to the same quintile of households when these were ranked (i) by MEP and (ii) by household-outlay rankings - 77-86% of households (78-87% of persons) were in the same quintile of households, when these were ranked (i) by MEP and (ii) by outlay-per-CU on Indian weights (Visaria, 1980, pp. 39, 46.) It is tempting to conclude, as similar Nigerian and other micro-data suggest, that "to give all persons equal weight regardless of age and sex does not produce too much distortion" (Simmons, 1976, pp. 26-7). And certainly distribution "using ... per capita income probably will be very similar to the distribution when using ... income per [CUJ" Datta and Meerman, 1980, pi 7). However - even if, across individual households, 47/ we find (i) high r between size in persons and size in CUs, (ii) high ri between outlay per person and outlay per CU, (iii) similar summary statistics (e.g. Gini coefficients) of distribution whether we measure outlay per person or per CU - we cannot be confident that, when two sets of households (e.g. in two areas) are compared, the estimate of numbers in poverty will be almost the same whether per-CU or per-person indioators are used. After all, if "58-74% of the households (including 61-76% of the population) fall into the identical decile" on both indioators in a group of surveys (Visaria, 1980a, p. 12), then 26-42% of households fall into different deciles. We know little about the distribution of the latter households between rich and poor, big and small, or country and city. They might well be concentrated among the poor. 48/ If outlay, in a household, is below a level producing poverty for a household of "typical" size, that provides almost no evidence that outlay-per-person in that household is below the poverty line. Outlay-per-person in a household (of unknown CU/person ratio) is a muoh better guide to its outlay-per-CU - though still far from perfect - than is total outlay, in a household of unknown size, to outlay-per-person. Finally, total household outlay probably ranks households (of unknown size and CUs) somewhat less badly by outlay-per-CU than by outlay-per-person. This is because households with big total outlay, while clearly larger, tend 61 also to have slightly higher child/adult ratios. The latter tendency, however, is quite insufficient to "rescue" household outlay as a poverty measure, if we believe that per-CU measures are the best but are not available in a particular case. Outlay-per-household ranks households slightly less badly in order of outlay-per-CU than in order of outlay-per-person; but outlay-per-person remains likely to rank them muoh more closely to their outlay-per-CU order than total household outlay would do. Per-person and per-CU are not too dissimilar; per-household is totally different. (c) The case for and against the three measures So far, we have said little about whether inadequacy of per-person or per-household resources best indicates "poverty". All that has been shown is that the two indioators produce completely different results. It is a defensible rule of thumb, in LICs with more or less typical income distribution, that at most the poorest 20 percent or so of households are ultra-poor - likely in normal years to be unable, at some time, to afford the minimum dietary energy intake required for good health (for a summary of the evidence see Lipton, 1983). Between four and seven in ten households assigned to the "poorest quintile" by income (or outlay) per household are not so assigned on a per-person basis in LICs; and between four and seven in ten households in poverty by income (or outlay) per person are not so on a per-household basis. It may seem self-evident that per-person indicators are better measures of poverty, or economio welfare, than per-household indicators. However, African "poverty datum lines" are often specified as the wage required to support an average household. Also, clearly competent research by individuals (e.g. Ahmed, 1981; Nugent and Tarawneh, 1982), governments (e.g. Government of Botawana, 1975) and international organisations (e.g. Mathew and Scott, 1980) persistently uses per-household resouroe indicators to establish whether a household is poor. Sometimes the reasons are administrative, or are based on laok of information about household size in different inoome-groups. Often, however, researchers and their organisations staunchly defend the use of per-household indicators of poverty. It is correctly asserted, for example, that resouroe allooations are usually to the household (or fungible inside it); Al/ that it allooates internally in ways little amenable to policy; and that these internal allocations, notably of food, are not necessarily optimal. None of these assertions, however, justifies using outlay-per-household as an indioator of poverty in preference to per-person indicators. The case for preferring these (e.g. Datta and Meerman, 1980) seems overwhelmingly strong, and the attempt to isolate "characteristios of poverty" linked to per-household indicators of the latter (e.g. Collier and Lal, 1980, pp. 38-9, 46; A4-6) appears rather risky and sometimes misleading. Only under two circumstances might per-household indicators be best. First, due to the importance in household incomes of the yield from jointly consumable owned assets (car, furniture, eto.) - assets having aome of the characteristics of "public goods" with all members of just one household as the public - or for other reasons, eoonomies of soale in oonsumDtion could be very large. If five oan live as cheaply as three, and can enjoy flows of services that generate the same levels of welfare, then it is wrong to assess 62 welfare by dividing household outlay by five for the family of five and by three for the family of three. These effeots are examined in Seo. (d), but there is some a Driori ground for doubt that, for a poor household in a LIC, they are very important. First, such a household spends 70-80% of outlay upon simple foods, for which scale-economies (while extant, in buying and cooking) are small. Second, if they were important, poor people - being likelier than wealthy ones to prefer income to privacy - would presumably reveal suoh preferences, and thus internalize the potential scale-economieat through a much higher propensity to live in more complex households than is shown by wealthy people. We know (Seo. II(h)) that is not the case. The second circumstance, under which outlay-per-household is a better guide to poverty-risk than outlay-per-person, is more plausible. We have seen that households with low NEP tend to be bigger than households with high pEP, but also to have higher child/adult ratios (Tables 2, 8). Suppose that low-NEP households - e.g. because they tended espeoially to have very small children - were hardly bigger in CUs than high-MNS households. Suppose also that outlay-per-CU is accepted as the most reliable indicator of a household's poverty, but is seldom available by per-person (or per-household) outlay classes. Then outlay-per-household might well rank households, in respect of welfare, in an order much closer than outlay-per-person to their "correct" ranking by outlay-per-CU. For example, suppose household A contains 8 persons and 5 CUs, and has $10 NEPF; B contains 6 persons and 6 CUs, and has $12 MEP. A's household outlay is $80, and B's $72; A's outlay per CU is $16, and B's $12. Then household outlay ranks A as richer than B, which is in accord with outlay-per-CU ranking; but NEP "wrongly" tells us B is richer than A. In practice, though lower-NEP households such as A do tend to have a somewhat higher persons/CUs ratio than higher-MEP (and generally smaller) households such as B - so that the correction in the example is in the right direction - the amount of the correction is much smaller, as the data on p. 60 suggested. It is in fact far too small to render significant numbers of households, better-off than others on NEP rankings yet not on per-household rankings, worse-off on outlay-per-CU rankings, so that households could be better ranked by outlay per household than by NEP. In general, we find a CU/person ratio decreasing only very slowly as NEP declines. These two effects, scale-economies in consumption and ranking disparities as between NEP and outlay per CU, are of course additive in their tendency to make household outlay a less bad poverty indicator, relative to NIP, than it might at first appear. However, even together, the two effects appear to be each so small, in the overwhelming majority of cases, as not to invalidate the comonsense conclusion. Normally, SEP appears to be very muoh better than household outlay as a poverty indioator. Indeed, the latter is almost worthless, unless the importance of scale-economies can be shown in Sec. (d) to be established, after all, by recent evidence. * 0 a Despite the above reassurances, it is easy to simulate conditions where NEP is a bad proxy, as a poverty-ranking device, for outlay per CU. Unfortunately, there is no universal way to weight males and females of different ages, in respect either of total "requirements" to attain a given level of health or performance, or of the "welfare" that each person would attain from a given income or outlay. Such weights could be different at 63 different levels, of household inoome-per-person (or per-CU) and fasily size, and in different cultural environments; this underlies the attempt to devise constant "equivalenoe soales" (pp. 64-5). Also, even if we could show that a man or 40 in a particular oase needed 1.3 times the outlay of a ohild of 10 to obtain the same welfare, there is no guarantee that intra-houwehold allocation would produce that ratio - or approximate it to the same extent in groups of households being compared. To some extent, however, all this oan be short-circuited in the oase of poor, or potentially poor, people in LICs. For example, in India in 1972-3, the 32 peroent of urban households, and the 76 percent of rural households, with lower levels of ME? - i.e. all MEP sub-groups up to 43-55 Rs./month - were, in each (often quite small) MEP sub-group, spending over three-quarters of MEP on food (Sarvekshana, Jan. 1979, pp. s354, 8423). Such concentration of outlay on food is almost always found among those who are ultra-, moderately, or even potentially poor. Among the three-quarters of Indian villagers - (and the one-third of townspeople) - with lower MEP, this food/outlay ratio is not only very high (75-80S), but falls only very slowly with rising MEP (and not at all for the poorest 10-15% or households). Also, all these groups spend the bulk of food outlay on oheap calories; and calories (dietary energy) "per unit" are usually the only requirement both at risk of serious deficiency and related in intake mainly to private inoome and outlay (see Lipton, 1983, for evidenoe for all the above). Therefore, oan we not establish the average oalorio requirements of males and females in different age-groups; use these to construct CU-equivalents for households of different total outlays; and - for the potentially poor - oonvert outlay to outlay-per-CU upon oalorie-based CU-weights? Suoh weights are widely available, and are used in Lusk, Indian and Taiwanese CUs. These three produce somewhat different poverty measures, but all appear to adjust crude MEP data in the "right" direction (see ftn. 46). What these adjustments suggest is that - because child/adult ratios rise quite sharply as MEP falls - low-ME? households tend to have relatively less inadequate outlay-per-CU, so that the inoidenoe and severity of poverty are less than they appear to be (although for various reasons the adjustments are very small). While this reduction may be ensible from other oontexts (see, for example, Srinivasan, 1980), the inferenoe from CUs and MP-ohild/adult relationships is not quite so olear. First, the inferenoe is actually strengthened by the faot that food/outlay ratios tend to rise not only with falling MEP but also (given MEP) with rising child/adult ratios (Deaton and Muellbauer, 1983), so that food requirements beoome better indioators among the poor than among total populations of a household's total "welfare size". Seoond, however, this very faot of a greater food/outlay ratio in households - together with the muoh greater risk of damage for children under five from a given proportionate shortfall behind dietary energy requirements - means that an 1% shortfall in outlay-per-CU does more damage to very poor households than to others. This, in turn, suggests that - even if "ME? below a given level" may somewhat overstate the incidence and severity of calorio risk - correction to per-CU indioators correspondingly understates them. Third, in parts of Bangladesh and North India (though probably not elsewhere), and in areas where women's prospeots in labor markets are partiolarly bad oompared to men's (Rosenzweig and Schultz, 1980), ultra-poor families in partioular (Carloni, 1981, oiting Levinson) appear to 64 be driven to discriminate against small girls - not other age-groups (Chen et al., 1981), nor boys - by selective underfeeding. This, onoe again, suggests that the move from MEP to outlay-per-CU "over-deflates" the requirements of the very poor. It follows that, when we ask "Has Area A more, or severer, poverty than Area B?", or "Does Project X benefit the poor more, or less, than Project Y?"p - we take an implicit decision by using per-person rather than per-CU measures. The decision is to weight low purchasing-power more highly in households with lower CUtMEP ratios, i.e., in general, with more small children per adult. Since it is in such households that poverty is likeliest to disable or to kill, this decision seems sensible in policy-making to reduce poverty by cost-effective alloeative decisions among projects or regions. (It would not be sensible in making individual clinical or diaxnostic decisions about which particular household or person needs what and why; but no economist (indeed no sane person, surely) would make individual decisions on the sole basis of grouped survey data about outlays, incomes and populations. 50/) We can be the more confident in using HEP data ranking in broad allocations, because the overlap, for relevant decisions, with per-CU data rankings is very close (p. 59); and because tolerably reliable data are more seldom available per-CU than per-person. There remains the question: should both per-person and per-CU data for outlay be replaced, as indicators of incidence and severity of poverty, by appropriate equivalence seales? These, if they could be constructed and if the relevant data were available at acceptable cost, would also take care of the problems of economies of scale in consumption (Sec. (d)). A deliberately loose statement would be that equivalence scales show the level of outlay (or income) that households of various sizes and compositions require in order to be on the same indifference surface. If we could then associate a particular ranked indifference surface, for each size and composition of household, with (i) poverty and (ii) a level of total household outlay (or income), we could then count the households and persons in poverty in a given area, or moved into or out of poverty by a particular project. 51/ Unfortunately, the steps needed to tighten up the above "deliberately loose statement" - while still leaving equivalence scales important for some purposes - remove much of their apparent potential usefulness as indicators of numbers of persons in need. First, some people would deny the possibility of comparing the impact of extra income on welfare among even broad groups of different persons; but this extreme position - apart from removing the point from almost all arguments about the projects, areas, etc. on which a cost-effective attack on poverty "ought" to concentrate - is surely pseudo-sophisticated and even a little silly. 52/ Second and much more plausibly, one can object to any attempt to compare the welfare - total, marginal with respect to extra income, or whatever - of groups of persons with distinct demographic structures: of, for example, childless couples and couples with three children. The only oredible solution 53/ is to ask: what extra annual outlay does a "representative" couple require, at a given time, to compensate it for the outlay foregone by meeting, at the level choosen by that couDle, the extra requirements of one, two ... children in various age-groups (or by divorcing into two one-member households; by taking an old, 65 incomeless person into the household; etc.)? The answers to these questions imply - if we assume that the oapacity for enjoyment, the "rights" to enjoyment, and the impaot on incentives of enjoyment are, among members of any set of demographically identioal households, statistioally independent of outlay - quite strong conclusions. In partioular, we could then isolate a level of outlay just sufficient to avoid *poverty" or "ultra-poverty" for (say) a couple with one ohild aged five, and work out welfare-equivalent outlay levels for all other sizes 54/ and age- and sex-structures of household. How to do this - and the surprising faot that maintenance of parental acoess to "adult goods" (mainly non-food), not the Engelian oriterion of a constant food share, is the most plausible indioator of welfare equivalenoe - has been explained very clearly elsewhere (Deaton and Muellbauer, 1983, esp. pp. 24-5). However, the method, even if data were oostless, seems unsuitable for our purposes. In order to oompare like with like (and to focus upon the indifferenoe surfaces of the decision-takers), we were logioally compelled to ask what extra outlay a couple needed to compensate it for welfare foregone by "adding", say, an infant to the family. Leaving aside some important conceptual issues - the time-stream of benefits to the couple from such an "addition"; the question of whether the couple would be compensated or merely given a windfall, if now "reimbursed" the streams of benefits deliberately sacrificed, through a decision to have more children, for expected benefits later; the possibility (Schultz, 1981) that the couple is making something like a (constrained) optimizing demographic-economic life-plan - there remains, at the "simpler" level of poverty planning, a more peremptory objection to any attempt to use equivalence scales as an indicator of how many households in Area A, or benefiting from Project X, are poor. This objection is that, although the most serious sufferers from poverty are children under five - who in extreme oases suffer nutrition-induoed mental or physioal damage, or even die - the logioal requirements of equivalence scaling oompel us totally to disregard these ohildren's level of well-being in deoiding how many households fall below a given level of "welfare" (ranked indifference surfaoe). "Perhaps this is unfortunate, because for many polioy purposes we are interested in the welfare of the children themselves. However, [equivalence scales] cannot be direotly interpreted in terms of children's welfare, although it would olearly be possible to add supplementary assumptions linking parental oosts to the welfare of their ohildren" (Deaton and Muellbauer, 1983, pp. 8-9). Even such assumptions could not dispose of the problem that under-fives cannot reveal their preferences by their deoisions as purohasers. The results from logioally aoceptable (Rothbarth-Barten) equivalence soales produce ohild-weights somewhat, but not dramatioally, below those of Lusk CUs. In such soales, "child costs oome out to be 30-40% of an adult" (ibid., p. 39); Lusk CUs are 0.3 of an adult male for children aged 0-1, 0.4 at 13, 0.5 at 4-6, 0.7 at 7-9, and 0.8 at 0-12. The Indian weights (Visaria, 1980, p. 200) would leave the average child aged under 12 about halfway between the Lusk and Rothbarth-Barten levels. Lusk and Indian weights also handle, as equivalence scales do not, the problem of adult women's "requirements" vis-a-vis men; but the soales oan deal with, and CUs ignore, economies of scale in consumption. Despite the advantages of Rothbarth-Barten 66 procedures here, however, and despite their logieal (and in other contexts empirical) merits, they seem to be unhelpful for LIC poverty measurement - and for allocations aimed at cost-effective poverty reduction - above all because their great advantage, logical rigor, implies the neglect of children's welfare. We thus need to look at economies of scale in consumption directly. How important are they? Could they mean that, although Area A or Project x looks much more oost-effeotive as a way of reducing poverty, the bigger households benefiting are in fact less poor than they seem - and the amall, poorer - so that more would be achieved in Area B or by Project Y? (d) Economies of scale in oonsumDtion An important analysis of a sample of 13,000 US households in 1970-71 (Lazear and Miohael, 1980) uses standard assumptions about optimizing consumer behavior to estimate how much must be spent on purchase commodities, by households of different sizes and structures, to generate a given flow of benefits-Der-Derson - say, that obtained from *10,000 annual outlay by a single person - from the services derived from purchased commodities (a flow hereafter called "eofare"). 55/ As household size and composition vary, so do purchases of all types of commodities. Hence Lazear and Michael use cross-sectional variations in expenditures, as household size and composition vary - plus independent estimates of "price elasticities - to infer changes in prices ... from which real (price deflated) levels of income can be inferred" for households with similar nominal income, but different size and composition (ibid., pp. 96, 104). In three main ways (ibid., pp. 92-3) larger households might get more welfare-per-person from a particular level of outlay-per-person. One is via family goods: "If $5 provides the man with a securely looked door and $5 provides the woman with the same ... then together its price [becomes] $2.50" to each when they form a joint household. A second is standard scale-economies, such as quantity discounts. A third is complementarity due to specialization within the household. The total effect of these three economies of scale in consumption (ESCs), at least in the USA, is astonishingly large. Each adult in a two-adult-household can enjoy, for $10,606, the same eofare as a single-person household enjoys for 10,000: i.e. "on average the prices of service flow units faced by couples are ... only 53 per cent ... as high as the prices faced by single-parent households" (ibid., p. 99). In other words, if two live as cheaply as one, they enjoy not 50% but over 94S of the level of living of "one"l A household of two adults and one child require to spend $13,342 to enjoy the same eofare per person that three single-adult households would get for $10,000 each: five of the latter ("costing" $50,000) would enjoy the same ecfare as one household with two adults and three children spending only $16,889. In other words, a nominal post-tax dollar appears to generate almost twice as much ecfare, at these outlay levels, for a childless couple as for a single person - and almost thrice as much for a couple with three children as for a single person (ibid., p. 97). Three other expert estimates of ESCJ for the USA are cited by Lazear and Hichael. They use different methods, but do not reach radically different results. Moreover, the higher levels of consumption-flows in larger US 67 households are not significantly offeet, in their effects on faully welfare, by higher doeastio work-inputs; only a small part of the greater "eofare-cost-erfeotiveness" of outlay in bigger households is traceable to "speoialization within the household" by labor, and the effeot of eztra domestic labor by tamtly members, suoh as the wife, appears as a proportion of this small amount (ibid., p. 102). Even for the USA, we do not yet know how these massive ESCa might operate differentially for households of similar numbers and ohild/adult ratios, but different age- and sez-oompositions. Nor need these results reduce the proportion of households below a Opoverty line". Howver, if these results carried over to LICs, the Impact on all existing analyses of the incidence of poverty in different tiDes of household - and hence in different regions, among sets of persons beneriting from alternative projects, eto. - would be devastating. Sufferers from poverty would then appoar to be muoh more oonoentrated in smaller households, Implying that anti-poverty strategies should redireot polioy and projeot benefits towards such households. This requirement of equity would clash with the requirement of efficiency if - as would surely be likely - signifioant parts of these average ESCa oontinued to apply to marginal outlays in bigger households, whioh however were now to lose some benefits from anti-poverty polioy to smaller households, preclsely because the latter were now counted as relatively "poorer" because less eotare-efficient. However, independent features of LIC data sets suggest that ESCs are muoh leas important - especially for low-ME groups - than in the USA. First, in LICs, espeoially among such groups, as the number of siblings rises, their proportionate death-risks rise (Birdsall, 1977, p. 75) muoh more sharply than in developed countries, and upon a higher (one-ohild family) base-level. Moreover, both high IHR (and ohild mortality) at low MEPs, and its tendenoy to increase with sibling numbers, are muoh more directly attributable in LICs than in the USA to "primary" poverty In Rowntree's sense: i.e. not due to household decisions to allooate resouroes away from basic foods, medicine, shelter, eto., but due to absolute lack of resources, per poor child, to combat undernutrition, infection, eto. This stark evidenoe of the effeot of low private MEP, as MNS rises, in poor LIC families militates against major ESCa for them. 56/ Seoond, the structure of outlay is much less favorable to ESCs in such families. Lazear and Miohael (ibid., p. 97) usefully indioate the types of comodity purchase for whioh, on their revealed-preference approaoh, the larger US household is espeoially efficient - compared to the smaller household - in tranasorming eaoh dollar of outlay into eofare. For example, overall eofare per dollar is about twice as high for a three-ohild couple as for a ohildless oouple - but 28% for shelter, 84% for food, 187% for clothing, 191% for transport, 184% for other goods, and 85% for other services. Much of the 84% on food is due to lower per-person costa when a larger party eats out. Food not eaten out (which oomprises 70-80% of total outlay for poor people in LICa, but - even for those olassified as "poorw --well below 35% in the USA) does offer some ESCJ in purohasing and preparation, but not very substantial ones. 57/ The methodology of US studies, too, could lead us to expect too muoh of ESCs for poor people. The figures on p. 66, while given to a norm of 68 $10,000 yearly for a one-person household, in fact refer to the relative puchasing-power of total, or average, outlays for households of eaoh size and adult-child ratio (Lazear and Niohael, 1980, p. 99). In the USA, and even more in LICs, there are a priori reasons to expeot poorer groups to show much lower ESCs, as indicated above. Nor is it clear whether similar ESCs should be expected whenever a household of given size and composition adds a member, whether young or old, male or female; and whether urban or rural. The lower incidence of the decision to live in complex households in LICs among the poor (Sea. II(h)), despite their presumably greater pressures towards the welfare optimization assumed by Lazear-Michael methods, might have at least two interpretations. It could be that ESCs there are in fact very amall. Alternatively, as argued on p. 32, a poor couple might have too little heritable eapital to induce married children to stay at home, even though if they did so there would be a larger stream of services, from eoonomic oommodities, to parent and offspring couples taken together. Equivalence-scales - apart from their major claims upon data - have basic drawbacks in assessing incidence or severity of poverty, especially if the main victims of poverty are children. Outlay-per-CU - for which outlay-per-person is a surprisingly good proxy, as regards poverty incidence and severity in large groups of households - has therefore to be used. In developed countries, however, scale-economies in consumption (at least for households around mean outlay-per-CU) may well be so great that an outlay-per-CU measure would lead to serious over-estimation of the relative poverty in big households, and serious under-estimation of it in small ones. Such scale-economies are probably much less significant for the typical poor household in a typical LIC. However, this last statement, in considering the impact of an extra person on a poor LIC household, averages far too many distinct circumstances, even holding outlay-per-CU constant - urban and rural, big and small households, extra child and adult members, etc. For instance, in India, low-MEP households' MHS is much bigger in towns than other households' MHS; in rural areas, the disparity is a good deal smaller (Table 2). The possible implications - that there is a higher ratio of urban to rural ESCs for poor than for other households; and that rural poverty is severer, relative to urban, than MEP-based comparisons would suggest - seem striking, but depend on absent information about urban and rural ESCs. The large apparent size of ESCs for the mean US household, in any oase, adds research and policy urgency to an improved understanding of ESCs for the LIC household. They could be much smaller in the latter, yet still could importantly refocus anti-poverty activities upon - for example - non-married women, especially where they have difficulty in remarrying or working outside the home, or where long-term male migration prevails. However, we are now unoomfortably ill-informed about the importance and location of soale-economies for poor LIC households in various circumstances. We return to research and policy implications in Sec. V. 69 V. DEKOGRAPHY. POVERTY AND POLICY (a) Policy for researoh, measurement and Drojeot planning Many of the oonclusions we have advanced in this paper are highly tentative. It is organised, first, around the oorrelates of differences among poverty-groups in MHS; as such, this topic appears to be largely unresearched in today's LICs, so that we have sought, in part, to assess the applioability there of findings from historioal demography in NDCs. The other central isaue considered here, the relationship of poverty to age- and sex-structure, is only beginning to yield generalizations, partly because that relationship has in the past been too readily assumed to be culturally so specific that they are impossible. This paper also limits its immediate policy applioations by seeking merely the demographic oorrelates of poverty, rather than, as yet, the direction of causes and effects. Nevertheless, a few policy suggestions emerge. The greatest oonfidenoe attaches to suggestions about polioy for research, measurement, and project preparation. Perhaps half the quantitatively based "findings" about poverty, and oertainly over three-quarters of the wrong ones, are still based on surveys that - openly or otherwise 58/ - rank households by "poverty" in order of their total income or outlay. Yet most households in, say, the lowest quintile by total household income are not in the lowest quintile by outlay-per-person; and the oommonsense observation, that per-person or per-CU outlay is a far better guide to poverty than per-household outlay, is fully borne out by all the evidence. Low MEP is strongly linked to high MRS in total populations; so, only slightly less, is low outlay-per-CU, despite the clear tendency, for low-MEP (and for big) households to feature high child/adult ratios. Our first policy conclusion - already implicit in Datta and Meerman (1980) - is that poverty lines, surveys, research, evaluations of the poverty impaot of policies (or projeots or tax systems), and project planning should never accept that numbers, or distances from a poverty line, of households with low total outlay (or income) are a remotely tolerable proxy for the inoidence, or severity of poverty in, poor households. Second, MEP is, for all its over-simplifioations, a surprisingly acoeptable guide to poverty. If seasonally averaged in a typioal year, and adjusted for price differences that may be faced by different groups, NEP does seem to rank xroups of households reasonably acourately. Consider two equally populous regions where 80% of households have MEP sufficient to command adequate diets. The region where the lowest-MEP 20% of households are 5% below an MEP sufficient (on normal spending patterns) to provide even 80% of average 59/ dietary energy requirements to those households is normally poorer than the region where 10% of households are 5% below that MBE, and the 10% with next-lowest MEP oan afford 80-100% of those requirements. Projeots benefiting the poorest 20% in the former region, given their oost-effectiveness, normally reduce both the inoidenoe and the severity of poverty more than in the latter region. Third, it is not certain that we oan find a better measure than MEP to assess a household's risk of poverty. Equivalenoe scales make aevere claims on data, cannot measure ohild poverty, and have other disadvantages. 70 Outlay-per-CU, on the other hand, superficially appears to be elearly superior to MEP in 'locating" areas, projecto, etc. especially important in polclies against absolute poverty; but is it? Where infant and child mortality is especially high, the apparent advantages of outlay-per-CU might seem especially big; for such high mortality conoentrates in the lowest-MEP households - as do infants and small ohildren with relatively low requirements and hence CU-weighting. - thus somewhat (though surprisingly little: p. 60) reducing CU/person ratios. Yet it is precisely in such ciroumstanoes that we would be most reluctant to adopt a weighting that relocated 'poverty", and thus social outlay to reduce it, in households with higher adult/ohild and therefore CU/person ratios; suoh reweighting would make adults less unhappy, but oause permanent harm, even death, for under-fives. The imDaot of a given shortfall of outlay behind 'requirements", especially if that shortfall means ultra-poverty, is greater in households where those ratios are low - although the size of the shortfall, given MEP, is greater where the ratios are high. We are conoerned with impact as well as size. So NEP may well be a more aoeptable poverty indioator than outlay-per-CU. Fortunately, the choice between per-C and per-person measures seldom greatly alters poverty rankings of major groups of households, or of big projeots. Fourth, we need to know muoh more about the possible impact of economies of scale upon household well-being. The evidence on the consumption side that suggests big ESCJ for average US households, and the hints and hunches that suggest such smaller ones for poor LIC households, have just been presented. But useful policy decisions require integrated research into scale-economies, for poor LIC households of different age- and sex-structures, in three areas: consumption, production (especially via self-employment), and job searoh. Different sorts of household enterprise (e.g. carpentry or hortioulture?) and of hired employment (e.g. casual or longer-term?) make sense for different sizes and oompositions of household. Of course, other issues - e.g. oapital and skill requirements, seasonality of incomes, and market structure - affoet the choice of appropriate aotivities to inorease poor people's incomes. *However, research into their effects on soale-economies in production and consumption is surely a high priority. Very little is known about these issues; yet they could well turn out to be central (saine poor households have sizes and struotures atypical of total populations) in deciding which policies work, and which fail, in advancing poor people's prospeots of self-suffilient esoape from poverty. Fifth, the findings on tMIS suggest a particular form of demogramhic as of the impact frm major policies - not just 'anti-poverty polioies' - on the poor; for policies helpful to the poor overlap oonsiderably with policies beneiting big households. The docile-wise MS averages (Visaria, 1980, p. 49) do not, perhaps, bring out sufficiently dramatically some of the size differenoes between poor and rioh households. In 1972-3, in rural Gujarat, only 2.2 percent of households in the bottom NIP decile had fewer than three mmbers, as against 29.5 percent of households in the top deoile. In urban Oujarat, the respective proportions were 2.7 percent and 43.2 peroent. Other Asian surveys showed similar features (ibid., p. 47). Just as really small households tend strongly to be better off, so really big ones tend strongly to be poor. Thus in rural N.E. Thailand in 1975-76, in the lowest NtP quintile, 42.2% of households had nine or more members, as against 25.0% in the middle and 12.2% in the lowest quintile (Meesook, 1979, p. 65); so about 60% of persons in the "poverty quintile" were in households of nine 71 or more. In urban Colombia in the 19708, a massive 78 pereont of households in the lowest income-per-person deoile oontained eight persona or more - as against 5.5 percent in the highest deoile, and 11.8 peroent overall (Birdsall, 1979, p. 132). In suoh oiroumstanoes there is normally a powerful Doditive relationshiD between the effeots of a DroJeot or Polioy on the poorewt and its effeots on big households. It is, of oourse, important not to enoourage higher fertility by the promise of bonefita to families that enlarge themselves. However, too many policies - notably in the setting of land-reform floors and oeilings - have been designed with little or no attention to household size. Sinoe larger holdings are almost always associated with larger households, it is not merely unjust, but oan sharpen poverty, if flows or entitlements are set in terms of land-per-household, instead of per person or per CU. 60/ More generally, a demographio assay, in order to estimate very roughly the impaot on different sizes of households that benefit (or lose) from alternative proposals, is on the above evidence a vital part of any assessment of the effect on poverty of those proposals. .9, The demographic assay should also look at the ohild/adult ratios in affeoted households. We have seen that these ratios tend to be bigger in (i) big households and (ii) poor householda; and that (iii) bigger households tend to be poorer and vice versa. Together with the dramatic eoess of ery big households (and deficienoy of very small ones) just disoussed, thia means that the poorest HEP deciles contain a hugely disproportionate number of children, often in very big multi-sibling households. Sinoe life-threatening undernutrition concentrates on ages 0-3, and is strongly linked both to extreme poverty and - even given sooial olass and poverty - to high birth-order and MHS (Birdsall, 1977, p. 75; Lipton, 1983), these ohildren will in turn, include disproportionately many infants and under-fivea, often doomed to early death, yet raising the dependency burden - and henoe the difficulty of esoape from poverty - in the households of which they form such a large part. Once again, our evidence of decile average ohild/adult ratios (pp. 43-5) perhaps understated the dramatio disadvantages at the interaeotion between severe poverty, large MHS, and high ohild/adult ratios. Around 1970-4, the quintile of households with loweat inoome-per-person oontained one-quarter of ohildren under 15 in rural India, and about 30 peroent in Colombia, Malaysia and Brazil (Birdoall, 1980, p. 39). In India's large National Sample Survey in 1972-3, in the pooreat 10.35 of rural households, 51.4 peroent of persons were aged under 15 (as against 41.55 in the remaining rural households), so that 16.85 of ohildren - supported by only 13.65 of adults - were in this most vulnerable rural household deoile. In the poorest 9.1% of urban householda, 49.8% were ohildren (as against 37.1% in other urban households), so that 15.35 of ohildren - supported by only 11.8% of adults - were in thia ultra-poor urban 95 of homes (Sarvokshana, Jan. 1979, pp. 3-305, 3-369). Under-fives - oontributing only demand and not supply to the household's eoonomio balanoo; and eopeoially vulnerable to poverty-related physioal damage - must have been oonsiderably more heavily conoontrated in very poor households. Any assasment of the iopact of projeots and polioles on poor people, then, needs to examine the effeot on large households with many ohildron. The 72 central issue of Part II, however, suggests a word of caution. Within a national population, area, or social group, high MHS and high ohild/adult ratios now tend in LICs to accompany poverty. But across areas, land-ownership classes, jobs, castes, sexes of household heads, or other status-linked groups, the opposite is true: as NDC history suggests, social norms associate larger households and many children with higher status, even though recent demographic trends and households' economic preferences associate high MHS and child/adult ratios with lower income. Hence a demographic assay must not be mechanical. An agricultural project might benefit big households with many children because it provided gains mainly for persons cultivating large acreages, or because it created labor-intensive production opportunities for poorer people within each holding-size group (including the landless). While the proposed demographic components in policy formation and property does therefore need to be fairly subtle, an elaborate exercise in applied statistics would be too time-consuming, too vulnerable to data limitations, and therefore a recipe for fudging and abuse. Required is simply an intelligent attempt to address the question: do the project or policy alternatives under review offer better, worse, or similar prospects of benefit to big and small households, and to households with high and low child/adult ratios, within given areas, status-groups, and total populations? That question implies fairly self-evident "second-round" questions about impact on poverty. The point stressed here is simply that the very sharp demographic differences among MEP groups strongly suggest that, to attack poverty effectively, major projects and policies require evaluation of the sizes, structures, and demographic types of households likely to gain (or lose) significantly. (b) Steering resources towards given, Doverty-prone. household demographies It is, then, clear that governments can improve their anti-poverty policies by incorporating relevant demographic knowledge into them, and often by undertaking new economic-demographic research. This apart, there are three - perhaps three-and-a-half - substantive ways to adapt anti-poverty strategy to allow for the speoial demographic oharacteristics of the various MEP groups. The first two ways accept those characteristics as given. First, public-sector action may seek to raise incomes, cut costs, or in other ways provide resources, for households with sizes, structures, and associated demographic parameters strongly linked with poverty. Second, big families with high child/adult ratios - which include vastly disproportionate numbers of the actually or potentially poor and ultra-poor - might be helped or induced to alter the nature or timing of their non-demographic behaviour (mainly job search and choice, asset size and structure, oonsumption pattern and intra-household distribution) in ways that reduce the risks of and from poverty, and, espeoially, ultra-poverty. Third, public aotion can seek to ohange attitudes, incentives, laws, technologies (inoluding contraceptive technologies), or delivery systems, in the hope of causing potentially poor households to ohange size, structure, or associated demographic parameters in ways tending to reduce the risk of poverty. Finally, halfway between the first and third option, governments might "redistribute" aocess to resources, including loans, so that poverty 73 becomes (1) briefly life-cyclioal for many rather than prolonged for few, and/or (ii) concentrated on parts of the family cycle where it does relatively little permanent harm. There is reason to expect significant, though limited, scope for the first two types of polioy, which accept the given demographic struoture in all MEP-groups but steer resources towards the sorts of household size, composition, etc. more typical of poor people. Large MRS and high child/adult ratios oharacterize the poor sufficiently to define olear demographio target groups (subject to the caution mentioned on p. 72). In reaohing these decisions, households are sufficiently rational, even perhaps optimizing (Schultz, 1981; Cassen, 1978,) to suggest that the polioy-makers ought indeed to adapt to such decisions - to attempt to dissociate them from poverty - rather than trying to persuade people to ohange them. However, suoh dissooiation may be an uphill task, because townward movements, industrialization, and increased status do not appear to reduce poor people's absolutely high MHS and child/adult ratio - indeed, may raise them relatively to the better-off. The extremely high propensity of poor households in LICs - and, of course, even more so of poor people and especially children - to concentrate in high-MHS groups, and oonversely the extreme rarity of poor single-person or two-person households, has imnediate consequences for horizontal and vertioal equity. In NDCs one would seek suoh consequences mainly via direct taxation. Horizontal equity - taxing equally poor people equally - would normally be advanced, whatever the degree of progressivity of direot taxation, by taxing households with many CUs more lightly per CU than households with few, thus placing a larger share of tax burdens on smaller households. Vertical equity - taxing poor people less heavily than rich people (on the assumption that suoh a tax-structure is needed to equalize sacrifioe, because income has diminishing marginal utility) - would normally be advanced by even greater shifts of the tax burden to small and "adult-intensive" households, in particular by using a per-person rather than per-CU basis for reliefs or allowances. In LICs, direot taxes are normally much smaller parts of the tax burden than in NDCs, espeoially for poor and even middle-inoome taxpayers. Attempts to use "poor people's demographies" to select a balanoe of taxation that improves equity - or reduces poverty - must therefore rely mainly on indirect taxes, positive or negative (subsidies). Food looms larger in the spending of larger households, because they tend to have higher proportions of under-fives; and simple foods, eaten at home, appear to offer few ESCs. It is well known that oonsumer subsidies on food - if fiscally sustainable, and if not supported by offering below-market prices to the farm sector (especially small farmers) - offer moat benefit, proportionately to income, to poor households of a given size and composition, because their food/income ratio is relatively high. It is less well recognized that food subsidies also offer most benefit, proportionately to inoome, to big households with high child/adult ratios (partly because children's consumption is more oonoentrated on food than is adults' oonsumption, given outlay per CU). Such households not only tend strongly to be poor; they are also households where poverty carries the most serious health hazards, particularly to children and most particularly to underfed ohidren. 74 The serious problems of direct food supplementation are familiar (see, for example, Beaton and Ghasseimi, 1982). So are the diffioulties of targeting food subsidies to the needy (and away from well-heeled urban civil servants, soldiers, etc.), and the apparent unpopularity in LICs of food stamp schemes. Most seriously, perhaps, the "fiscal crisis of the State" in the 1970s showed in LICe from Sri Lanka to Egypt that more general food subsidies, in hard times, get squeezed between rising offtake (and henee public outlays) and lower tax revenues. Indeed, this "fiscal crisis" was also a foreign-exchange crisis, further squeezing food subsidies between growing food imports, discouraged domestic food producers, and recession-hit exports.61/ We nevertheless stress the demographic and health gain, to the poorest quintile or so, of measures that (without negative impact on farm laborers) render purchased food cheaper - especially if they can be directed, even more than is already achieved by the high food/outlay ratio of ultra-poor multi-child households, towards such households. The dramatically higher incidence in such households of infant and child undernutrition and mortality underpins the demographic evidence. There is a case, in anti-poverty polioy, for food subsidies, stamps and supplementation. However, such measures - even if, rightly, they are concentrated on cheap cereals and roots grown and eaten by the poor, and are not paid for by farm-price repression - are known to be difficult, vulnerable, and often ill-targeted remedies. Certainly, it is not likely that tax or subsidy measures affecting non-foods can as effectively home in on poor people's demographies. Kerosene, for cooking and heating, is often selected for pro-poor subsidization, but is probably even harder to target than food subsidies. Large families need less cooking, and much less heating, per member than do small (and normally less poor and vulnerable) ones; many of the poorest cook and heat with wood or dung, collected by family labor and much cheaper for them than even subsidized kerosene; and even the average LIC family spends a much smaller share of outlay on cooking and heating than on raw foods. Probably the standard tax-subsidy area is too narrow a field for publio action that seeks to steer resources towards "poor people's demographies". Thus the point is not that, because food subsidies are difficult to handle, one should look to non-food taxes and subsidies. Rather, the "food-centredness" of poor people's demographies provides yet another good reason to steer public-sector activity, as a whole, more toward helping or encouraging farmers (i) to raise their output of low-oost oalorie sources - root-crops, millet and sorghum, maize and cassava - and/or (ii) to supply such products to poor African and Asian people, rather than to European and American cattle. The case for, difficulties of, and comparative-advantage limitations upon such consumption-orientated policies are much disoussed and will not be rehearsed here. On the produotion side, the policy implications of poor people's heavy concentration - especially within urban populations - into big, multi-child households are less familiar. They go beyond mere supporting arguments for orientations (e.g. towards food) already suggested by other, non-demographic features of people. In particular, Droduction in or near the home, oomplementary with ohild care (espeoially for women) and if possible not acutely seasonal, offers special opportunities to the typioal large. multi-child poor household. It is quite cozon to dismiss family modes of 75 production as either old-fashioned, ineffioient and doomed, or else inevitably "exploited" by big enterprises (espeoially through outwork). Indeed, the daily rewards to outworkers usually oempare badly with faotory-wages for similar work. However, for large poor households, "threatened" during and after pregnancies by reduoed female labor participation - and above all for the health of small children in female-headed households - home or homestead, including artisan, self-employment (including outwork) can made the cruoial difference between poverty and ultra-poverty. If measures to improve the position of outworkers (perhaps especially in the case of women) concentrated only on raising the price of domestic labor to an unchanged structure of (price-sensitive) purohasing enterprises, such measures would seriously damage the very groups of demographically vulnerable poor people they sought to help. Public action, instead, should seek to increase the bargaining and market power of outworkers, by assisting them (i) to own a larger proportion of their productive assets (raw materials and/or capital) individually or jointly; (ii) to control the marketing and improvement of their produots; and (iii) to negotiate with a more competitive structure of product buyers and materials suppliers. The demographio nature and dynamics of poverty in female-headed households - though such poverty is more complex, less static, and perhaps somewhat less severe than is usually claimed - add special importance, in anti-poverty policy, to the largely neglected area of domestic, especially rural, artisan and hawking enterprises. Many examples could be given of the general principle here suggested: that, in assessing how any policy of proposal affects the poor, one should consider carefully its impact on large, multi-child households. (Conversely, transfers of resources to small households in LICs - though frequently not in NDCs - are likely to harm the poor). Measures affecting the demographic parameters that underlie MHS - migration, fertility, complexity, migration, etc. - are considered in Seo. (d) below. However, age-sex-status interaotions affeot even the part of anti-poverty policy that consists of statio resouroe allocation among households with different demographic structures. In Sec. III we saw that female exposure to poverty risk was associated with (i) certain sorts of (usually unmarried) civil status, and sex-imbalance in the household, which both also damaged males, though rather less; (ii) difficulty - not shared by men - in increasing income with age after about 30, and reflecting difficulty (partly due to discrimination) in acquiring assets and education; (iii) following this, not so much static disadvantage vis-'a-vis (i.e. greater poverty than among) men, but rather a lesser prospeot of escape, and less confinement of poverty to a shortish period of the life-cycle. These three facts add relevanoe to policies on self-employment, and on very small-soale, household-level ownership of appropriate assets, in any attempt to redistribute resources to households which - notably because of age or sex or HH - may be at especially high risk of poverty. (c) Non-demoxraphic behavior of big. poor households The concentration of poverty and ultra-poverty upon big households with high child/female and child/adult ratios (and youngish heads) is comon to many LICs. Yet not all such households, at all times, are poor. In 76 particular, the transition from poverty to extreme poverty is often not accompanied by a further rise in MHS, child/female, or child/adult ratios. Moreover, in NDCs, these parameters tended actually to fall with increasing pressure of poverty; and, even in LICs today, at each level of MEP a low "status" (asset ownership, landholding, job type, location, etc.) accompanies low MHS. All this suggests that the nature or timing of household behaviour can "de-link" inherently unfavorable demographic characteristics from the poverty normally associated with them. How, if at all, can policy help in this de-linking? The question acquires special importance from the high incidence of especially vulnerable persons - children under five - in demographically typical poor LIC households. One option is in the area of job search and choice. Women and older siblings are often prevented from earning by the need to care for under-fives. Apart from appropriate artisan or trading activities based on the home (Sec. (b)), rural works programs such as Maharashtra's Employment Guarantee Scheme - programs that guarantee work near home, and/or by piece-rates or otherwise permit substitution among family members - can greatly assist such households to raise their participation rates: can de-link high child/adult ratios from low participation. So can information, or transport, to help the process of short-term (or even daily) intra-rural migration. In urban areas, of course, legislation that restrains retailing (including hawking), rickshaws, poultry-keeping, repair work, and other home-based enterprise is especially damaging to big, poor families with high proportions of females in the potential workforce. Such females are especially likely to be able to combine income-earning work with family obligations only through home-based "informal" activity. The higher female/male ratio (yet also much higher MHS) among the urban poor, as compared to other urban households, renders building, zoning, licensing, labor, and other controls on urban informal activity - controls nominally designed to help poor people - counter-productive in the context of anti-poverty policy, unless needed for health and safety. More usually, such controls are merely concealed protectionism for big enterprise, public and private, that both competes with informal family firms and intervenes in the administration of controls that restrict their activity. Options also occur in respect of asset size and structure. As for human assets, the familiar linkage of secondary female education to low fertility implies that families with many children are likely to have mothers with, at best, primary schooling; neither such schooling nor post-primary terminal or adult-education courses typically help working mothers to combine family and work management better. It is often rightly stressed (Lloyd, 1979; Bromley (ed.), 1978) that poor, big families usually cope very well with horrifying difficulties; however, their very different success (given the level of poverty) in avoiding undernutrition or ill-health (C. Shah, 1979) shows that there is something to learn. As regards productive physical assets - land, craft machinery, durable consumer goods, or bicycles fungible between production and consumption uses - the very poor usually own very little; but, as indicated aboves much more needs to be known about which assets are critical for poverty-prone large families. Consumption patterns of poor families with many small ohildren may well account for their dramatically greater vulnerability to nutritional 77 stress and even death (Ruzicka, 1982; Birdsall, 1974, 1980). The combination of childbirth, a normally hungry season, and an atypically bad year can impose "breaking strains" upon families for whom even their normal MHS and child/adult ratios mean some risk. It is not known whether such high-risk families (and seasons) overlap with high, or low, availability and use of primary health facilities; once more, in health policy too, appropriate effort to mitigate the effects of given "demographies of poverty" could achieve as much, in reducing at least short-run ultra-poverty, as ambitious attempts to change them. On the other hand, exhortation to poor parents to "improve" allocation of food or health-care, towards small children or females within their large families, might achieve rather little. Intra-family "misallocations" are (i) perhaps less common than is often alleged, (ii) often, rational responses to terrible dilemmas (Lipton, 1983, Sec. III(a)). Appropriate assets and work chances, not lectures on ethics, are what is required. (d) Policies to change the demographies of the poor Historically, poor households tended to be smaller, and to have lower child/adult ratios, than better-off households. This reduced both the number of persons affected by a given incidence of household poverty, and the proportion of those poor persons consisting of children under five - the most vulnerable to permanent harm, even death, from poverty-induced undernutrition or illness. Unfortunately, the relationship has been reversed in most LICs. Is there scope for policy to help restore it - to de-link high MHS, and high child/adult ratios, from the risk of poverty and especially of ultra-poverty? 1. Fertility Fertility probably plays the main role in the reversal of historically positive affluence-MHS correlations in today's LICs. It is better-off women who - in four interlocked processes - have come to adopt modern contraception, to receive post-primary education, to marry later, and to reduce marital fertility. Can and should policy attempt to distribute these forms of behavior differently among income-groups? We cannot here review the massive literature on family planning. Clearly, however, policies to increase the use of modern means of contraception have been motivated mainly by the quest for growth in GNP per person - by the wish to divert personal savings and public investment from "widening" uses that merely maintain, for a larger population, existing levels of capital, land, education, etc., towards "deepening" uses that increase per-person availability of such sources of growth. Accordingly, as almost all KAP studies confirm, access to and offtake of modern means of contraception have in most LICs concentrated on the readier customers, notably better-off, better-educated urban women. Even from the standpoint of growth, this emphasis on low-oost aoceptors - apparently such good economio sense - may have been misplaced. Many suoh acceptors had probably already attained their family size norm. By adopting modern oontraception, they were shifting to more aesthetic methods of 78 contraception, rather than reduoing age-specifio fertility. Certainly, in any event, modern contraoeption in most LICe - there are familiar, striking exceptions - neither got far down the scale of income-per-person nor achieved the hoped-for big reductions in overall birth-rates. From our standpoint - that of delinking poverty from the riskiest sorts of demographic structure - the policy of concentrating attempts to spread modern oontraception upon ready acceptors was plainly unhelpful. Absolutely, poor families gained little: too little, for poor ohildren, to meet the new need. This need was to cut fertility (so as to free household resources for existing children) enough to counterbalance the effects of public-health measures - especially malaria oontrol - in increasing infant and child survival rates, and thereby reduoing resources-per-person. Relatively, poor families probably lost out as contraception spread, because their MHS was reduced much less than was that of better-off families; this probably raised the supply of unskilled labor (especially from the poor) relative to demand (especially from the better-off), rendering poor people's unskilled wage-rates, especially in domestic and personal service, lower than they would have been in the absence of "rich-selective" contraception. With all due respect to the view that the poor choose their fertility and MHS - a view supported by much, but not all, evidence on traditional fertility controls - such choice is bound to inoorporate some insurance against childlessness, especially given the risks to women of unsupported widowhood. Insurance implies produoing a larger expeoted number of children than the family size norm, so as to avert some of the very serious risks involved in ending up with a smaller number. Better oontrol of mortality and fertility for the poor, therefore, could be substantially MHS-reduoing. While attempts at compulsion are probably unethical and certainly counter-productive, there is a strong welfare case for improved parental control of, information about, and acoess to contraception for the poor, accompanied by appropriate changes in incentives. Forms of insurance alternative to high child/adult ratios - from reduced infant mortality via better access to health care as in Kerala, to employment guarantees or even social old-age support - would, in respeot of impact on poor people's fertility and 1HS, at least justify closely-monitored experiments. Another indirect approach is relevant too. The family-planning literature is replete with evidence that, even holding earnings constant, female post-primary education delays marriage and reduces marital fertility. Yet suoh education is concentrated on the better-off, whose MHS is already smaller, and whose children are much less likely to suffer lasting damage from high MHS. It is often argued that a very wide spread of primary education is needed first, (i) on efficiency grounds in order to improve the intake (and hence efficiency) of post-primary education , (ii) on equity grounds, so that more should obtain basic skills before expanding privileged aocess to further eduoation. These arguments have foroe; however, primary education - especially if incomplete - seems to do little to out fertility or delay marriage. These effects do appear to improve the equity case for continuing to concentrate some expansion upon post-primary schooling, if the access to it can be saread to poor DeoDle. If so, such schooling is likely to cut their fertility and 141, to ease the escape from poverty, and to soften the oonsequenoes of ultra-poverty. In any case, it is questionable whether laws and exhortation can do nearly as much as mass post-primary education to 79 spread, into poor households, the oase for later marriages - as a oomparison of Sri Lanka and India shows. The effects of most polioies regarding fertility, howeverp are clouded. It appears to show a -,~ -shaped relationship to affluence, as well as to education. Improvements in the lot of "the poorest of the poor" may well raise fertility and 4HS, at least in the short term. This is associated with the flattening-out observed among these households in the rise, with falling MEP, in MHS and child/adult ratios. However, the top of the fertility-affluence .- N~ (like the flattening-out in the poverty-MHS curve) appears to occur at a very low level of well-being; all but the very poorest 5-10% of Indian households seems to show declining fertility, MHS and child/adult ratios as cross-section MEP rises. 2. Mortality Since 1945, malaria control has spread through most LICs, alongside major improvements in food security during famine. Both malaria and famine have recurred in some areas, but their reduction has brought big reduotions in IMR and child mortality. For better-off parents, it increasingly looked both safe and rewarding to restrict fertility in response, completing a demographic transition; for most people, and especially the poorest 30-50 percent of parents, it did not. Hence falling IMRs, together with selective fertility responses, have in LICs played a big part in reversing the traditional link of low MHS to poverty. This aocount, however over-simplified, may remove the fears that "harsh realism", in neo-Malthusian guise, destroys the benefits of better child health for poor people, because such "improvement" brings a higher household size, dependency burdens, and thus for the ultra-poor more undernutrition, disease and death. Any such argument, let alone an inferenoe that "health for all" might not do the poor any good, is quite unjustified. "Thinking the unthinkable" is not laudable if the thought-prooesses are muddled or wrong. Several pieces of cross-section evidence from Seo. II(e) reinforce the policy conclusion that better health and lower mortality for poor children, except perhaps in the very short run, will help to dissooiate high 14S from poverty. (Juat one example: in India, urban areas - and the State of Kerala - already show both lower IMRs and lower fertility.) It is, however, the poorest and remotest households that, discontinuously, suffer much higher infant and child mortality - but do not over-insure, or perhaps even respond fully, by raising their fertility. If the reasons are partly physical, it is possible that better nutrition (via higher household inoome) will raise fertility as well as cutting mortality. Since the groups affected are the very poorest, however (Table 12) - by no means a majority even of the nutritionally at-risk ultra-poor - this would be a short-term effect, while the bottom 5-10 percent of people reaohed the health and/or income levels of the next 10-15 percent. However, it does suggest an addendum, or converse, to the (probably correct) conventional wisdom that seeks to inoorporate family planning programs into wider maternal and child health care; it also makes sense to accompany programs of preventive medcline with arrangements to make appropriate means of contraception readily available. 80 3. Complexity Whatever the possible entrepreneurial drawbacks of complex producer-consumer families, they provide mutual insurance, some ESCs, and perhaps assistance in some forms of domestic production. As explained in Sec.II(h), it is better-off households that often seem to have the explicit option of complex status. Poor households mostly enlarge, perhaps via implicit choice (Schultz, 1981), but by only partly controlled excesses of births over deaths. Can policy ease the path of poor households towards choosing complex status where appropriate? At present, this often fails to happen because, though household members as a whole would stand to gain, individuals might lose, and "veto" complexity by setting up independent homes. Direct intra-household intervention by Government (here as with food distribution between daughters, sons and parents) seems infeasible; but can the non-zero-sum game be made easier to play? Provision of competitive (rather than subsidized) credit for domestic production activities that offer advantages to parent/child or multi-sibling co-operation - such as trading, dairying or some forms of artisanship - could be one way forward. It depends, however, on steering such loans to poor people who would thereby be helped to choose household complexity: a tall order for a government agency. More credible, perhaps, might be measures to help groups of houses, in slum areas or (in India) scheduled-caste quarters, to use common facilities requiring several users to be economic: standpipes, even sometimes small bio-gas plants. That way, the advantages of complexity might become more accesible to the very poor, even if the mutuality required for joint household formation remains rare among them. 4. Migration Once again, an explicit and clearly voluntary route to adjustment of MHS appears to be relatively less open, or less rewarding, for very poor households, leaving such households with fewer options than others in responding to unexpected deaths, conceptions, or income changes. Poor people's migration is shorter-term, shorter-range, less "urbanizing" and educated, and less prepared (i.e. more prone to painful learning-by-doing) than the migration of the better-off. The ultra-poor, except in desperation, usually lack the resources for prolonged job search (or income delay), and migrate only seasonally, along familiar but very low-yielding tracks. Better information about rural work chances, possibly through registers of available work and skills, could "piggyback" very cheaply on mobile government services - input delivery, health, extension, etc. - within an area. Most discussions of migration policy seem to concentrate upon whether and how, if at all, it should seek to change the scale of migration. But the structures of migration (by age and sex, origin and destination, duration, MEP decile, etc.) - even if chosen optimally, given the often high and price-discriminating search costs, by each household - involve major negative externalities (i) for individuals within some households, (ii) for groups of households (migrant, resident, low-MEP...) taken together. Townward migration structures in Asia are such that the poorest two or three urban deciles at least escape the hugely unbalanced male/female ratios that characterize better-off urban groups - but suffer correspondingly high MHSs and child/adult ratios. Customary areas of origin, using special information (normally from 81., relatives), monopolize net gains, though also some (gross) losses, from emigration; if, as one would expect, the gains carry diminishing returns and the losses increasing costs, then a strong case can be made for adaptive regional policies on information, transport flows, even education. In any case - as with complexity, so with migration - more thought is needed about policies to widen the range of poor and ultra-poor households' explicit and well-informed demographic responses. 5. Non-family household members In Sec. II(k), following Hajnal (1982), we emphasized the importance of yet a third demographic option - "going into service" with the better-off - as a link, now cut, that was historically chosen by many poor households in order to attain a tolerably low MHS, both directly and as a means to deferred marriage and hence reduced lifetime fertility. In this choice must be included many sorts of resident apprenticeship - which, as readers of ohs. 8-10 of Oliver Twist will recall, often involved poor children in dubious as well as serf-like activities, not only in LICsl Certainly there can be no question of policies that re-create resident "service" as a major option for poor teenagers (though some would argue that military service performed an analogous role). At most, it might be argued that, in locating public-sector construction and other activities demanding a locaily resident workforce, authorities might give preference, ceteris paribus, to areas with many big, poor households. Also - as in the case of outworkers - resident domestic servants should be protected, not by wage or other legislation that prices them into unemployment (or back into overcrowded and poor homes), but by information, options, and access to publicly-provided health and education, at the expense of the general taxpayers and not as a specific tax against forms of employment that ease the burdens of big and poor households. The main point about resident service, apprenticeship, etc., is that it represents just one example of the decline of traditional social-security systems that once protected the ultra-poor. With both labor and labor-saving devices more plentiful, relative to other capital, in today's LICs, than during early industrialization of the NDCs, neither "moral community" nor class interest is so likely to persuade the better-off to preserve the very poor in hard times. Resident service and apprenticeship are further threatened by greater mobility - of homes, labor, and capital - and by research-based scale-economies in mass-produced goods and services (including processing and distribution) that compete against small-scale, little-researched traditional and domestic skills. Yet the demography of poverty feeds on itself: if very poor couples cannot be sure of placing a child in resident (and probably income-remitting) work outside the home, and if security in old age is also lacking, they are driven to provide their own security by procreating larger, child/intensive, and hence persistently poor households. Unless some form of publicly provided security system or asset-provision or employment-guarantee substitutes, in our new labor-surplus world, for the old and imperfect (yet real) social security of traditional communities, it is hard to avoid the conclusion that each poor household must - if it is rational - insure itself by procreative behaviour; this both impedes "capital deepening" and - by raising poor households' MHS and dependency ratios and ultimately by raising unskilled labor supply ahead of 82 demand - perpetuates poverty. (e) Poverty, policy and the family cycle The role of the family cycle in poverty in most LICs is too small; if it were larger (provided chidren oould be safeguarded) poor people would be better off. This is only a pseudo-paradox. Family-cycle poverty comes and goes. Its temporary nature (i) reduces the welfare loss, (ii) makes it victims more creditworthy. The evidence tentatively suggests that a given "amount" of poverty in a typical year (i.e. say, of headcount below a fixed poverty-line, times average shortfall, modified to allow for intra-poor distribution a la Sen, 1981) is much more life-cyclical, and much less concentrated upon a few lifelong victims, (i) in LICs than elsewhere, and, relatedly, (ii) in micro-societies, to the extent that status, jobs, assets, and earning power are ascribed rather than achieved. This is an empirical statement only - it need not be true. If all jobs and assets are allocated in perfect markets, there will be a lifelong poor underclass unable, in such achievement-orientated societies, to achieve much. It is contingent, not necessary, that the lifelong poor group of households ascribed low-earning assets, jobs, etc., in asoription-orientated societies constitutes a much larger proportion of the poor than does the "underclass" in achievement-orientated ones. However, it does seem to be clearly the oase. Hence policies to improve the functioning of labor markets, and to equalize access to them, are likely to reduce the "illfare" created by poverty, because such policies render poverty brief for many instead of durable for few. The proviso that children must be protected is, however, important. Otherwise, a shift from lifelong to life-cycle poverty could harm the poorest. That is because life-cycle poverty in LICs - unlike NDCs, - concentrate much more heavily on youngish couples with several small children than on, say, old single persons. More spending on child health and nutrition, especially on helping the near-landless to feed their children enough calories, is therefore necessary if the poor are to be helped by policies making poverty more life-cyclical. Such policies would consist, above all, of steps to improve the access of currently weak groups - in different sooieties they might be women, over-fifties, countryfolk, scheduled castes, minority religions, etc. - to the means of exit from poverty. The apparent difficulty of Batawana women in acquiring assets after age 30 is a case in point: almost oertainly, oapital markets are less open to them than to men. Laws, inoentives, and market organization - all combined in suoh prejudices as that against labor as loan collateral - all combine to create artificial barriers against escape from poverty; polioy should seek to remove those barriers. If growth is slow and asset distribution fairly fixed, such policy will not normally reduce poverty. It will, instead, shift it from being mainly a lifelong condition for some, to being a temporary life-oyclical condition for many. That is a limited achievement. To do more almost certainly depends on asset redistribution and rapid growth. In any context, policies that neglect the quite distinotive demographic circumstances of the 83 poor, however well designed in other respeots, oan have serious unintended effects on poor households in general, and in particular on the vulnerable children so severely over-represented among the ultra-poor. Demographic assay of the anti-poverty impaot of major proposed policies (Seo.(a)) io, therefore, a clear-out need in eaoh speoific situation, whether or not the inevitably general and tentative polioy proposals advanced elsewhere are deemed appropriate. 84 ABBREVIATIONS AERC Agro-economic Research Centre (normally Sardar Vallabhbhai Patel Vidyapeeth, Vallabh Vidyanagar, India) AJCN American Journal of Clinical Nutrition BSPS British Society for Population Studies CU consumer unit (see fn. 46) DR dependency ratio EPW Economic and Political Weekly (Bombay) iwes economies of s¢ale in consumption HH head of household IHR infant mortality ratio IVS Indian Village Studies (of AERC) KAP knowledge, attitude and practice (family planning surveys) LIC low-income countries LINUM long-run, individual, non-marriage, rural-to-urban migration DEP monthly expenditure per person (normally averaged over a year) MHS mean household size NDC now-developed country NSS National Sample Survey (India) PDR Population and Development Review PI-NDCs pre-industrial NDCs r. reference period B. survey period SRS Sample Registration Survey (India) WBSWP World Bank Staff Working Paper Table 1: SIMPLE C ERATICNS (r): YXtEMSEWD SIZE, HJSECHoD INCX)E (OR aILAY), AND HOUJSEHM PER PERSC IN mE (OR CUMAY) Cbumtrv/State Gijarat Mbharashtra Gujarat Maharashtra Nepal Nepal Sri Lanka Taiwan Taiwan Rural (R) Urban (U) R R u U 11 tojns 7- tons National Natianal NatiCoal Date 1972-3 1972-3 1972-3 1972-3 1973-4 1974-5 1969-70 1968 1974 Measure Outlay Outlav outlay Outlay aCtlay Outlay Inoome Inome Inome Simnle corre- Decile rank lation of by ner head _.9503 -.9551 -.9731 -.9837 -.9496 -.9301 -.9868 -.9652 -.9889 average size welfare nea- of household sure in decile upco: Decile rank +.9779 +.9769 +.9668 +.9806 +.9763 +.9776 +.9864 +.9393 +.9408 by total- hasehold resoumre c flow mea- sure source: caiputed Fra Visaria (1980a), Table 3; see also Visaria (1980). Notes: (a) Latter saorce gives mbers of sample households. Sample sizes range frua 2254 (Nepal, 7 towns, 1975-5) to 11103 (Maharashtra, urban 1972-3). (b) Fbr n = 10 (there being 10 deciles), the .1% significance level for r (with n-2 = 8 d.f.) is plus or minus 0.862. Tabls 2: 1U )D SIZE, C1CPTrrID A4ID U!AY FCMW1Y', W. INDIA, 1973-74 * of househoids Hmnao uize in Chdrlen (0-14) as Adult sals pEfr I of poerma with utlaVy per w cdlay per P of peI m'q in a< fale in *4th OcdLay. nonth oar pFonm 0%) qnaq ctLAy grouP outLy gr Ftun3a 0-34 34-43 43-150 150- 0-34 34-43 43-150 0-34 34-43 43-150 150- 0-34 34-43 43-150 150 0-34 34-43 34-150 MC04 Quaindat R 14.2 18.9 65.7 1.3 6.8 6.0 5.2 4.1 48.4 45.5 41.7 43.4 1.07 0.84 1.13 l.OD 17.3 20.4 61.3 1.0 hwiazdktra R 22.5 18.7 56.6 2.3 5.9 5.6 5.1 3.2 50.0 46.7 39.2 31. 0.96 0.97 1.03 0.96 24.9 19.6 S4.1 1.4 jasitham R 13.9 13.2 69.0 3.9 6.2 6.3 5.6 3.7 49.1 46.0 42.4 38.2 1.01 0.95 1.04 1.04 15.1 14.6 57.8 2.5 Qalarat U 5.6 12.6 75.1 6.7 7.3 7.1 5.0 2.4 52.1 46.0 38.9 24. 1.CD 0.91 1.03 2.15 7.8 17.1 71.9 3.1 ahardni7tr U 9.0 8.8 63.2 19.0 6.6 6.2 4.8 2.6 50.9 44.6 36.6 19.2 O.99 0.95 1.20 2.10 12.9 U.8 55.8 9.5 Riansom U 6.5 12.7 n.8 9.0 8.0 7.0 5.1 2.7 49.1 50.9 40.2 21.9 0.91 0.87 1.15 1.761 9.8 16.7 58.9 4.6 Eturro: cal1abe frnm Govt. of Ihdla, Watisial Saiple Survey; 28th Ptnd Oct. 1973 - Seabotber 1974, no. 240, oo. 11,13,81. Notes: 'Icnest' outlay gazm, baelw RsIM3/erozss zxxth; 'low' 34-43; 'tycici', 43-150. Use of the swe borderline in rural and abm areas Jliss, in vifiW of semamt highar urwan =ries, that direct conmrison of rural and umban om.latiu. in 'lowest' and 'low' outlay-rer-verecn gr9uX somnewat undalrstates zma vrcorticrs in overty relative to rural osartiras (first and last 3 cola. of figures). Ftwal-urban omxariscres in the other 3 sets of columw ns t also be awaroedud cautiouly, beCame proortiouI of hobue- holds, versons, etc. in I lowest' and 'low' gzouIs are amdsderibly lcwer in urban areas - partly reflecting hiohr urban prics, and partly higher urban real inscms. Table 3: MAIN IiNCOM SOURCE, ANNUAL INCOKC-PER-PERSON, AND FAMILY SIZE, N.W. INDIAN VILLAGES A. GUJARAT AFAWA ANEODIA Main Income Sources House- Per- P/H Main/Total Income/H lIncome/P House- Per- P/H Main/Total Incoae/H Income/P holds sons Income (S) (Rs.) (Rs.) holds sons Income () (Rs.) (Rea.) Cltvtn., Operator 62 385 6.21 80.9 2,330 375 102 779 7.64 - 2,957 384 Amlis., Non-op. owner 1 2 2.00 51.3 828 414 10 29 2.90 - 951 328 Dairy Laboror 145 616 4.25 95.7 557 131 107 499 4.62 - 582 126 Non-ag. laborer 9 33 3.67 84.2 523 143 Arts, crafts 9 34 3.78 87.4 765 302 8 30 3.75 - 844 225 Services 13 45 3.46 74.5 1,155 333 22 111 5.05 - 1,888 235 Remlttances 13 76 5.85 69.6 2,324 398 - - - - - _ Trade 4 20 5.00 - 477 498 Transport 3 24 8.00 46.6 2,855 357 - - - - - - Profession 7 36 5.14 - 1,511 294 Others - - - - - 8 29 3.62 - 365 101 Total 255 1,215 4.76 82.7 1,143 240 268 1,533 5.72 - 1,597 279 WDTI KHADOL NARGOL Main Income Sources House- Per- P/H Main/Total Income/H Income/P House- Per- P/H Main/Total Income/H Income/P holds sons Income(%) (Rs.) (Rs.) holds sons Income(%) (Rs.) (Rs.) Cltvtn., Operator 76 390 5.13 92.3 1,876 370 60 397 6.62 69.3 3,020 45 ADmls., Non-op. owner - - - - - - 13 83 6.38 82.6 1,627 25 Dairy Laborer 47 173 3.68 79.4 595 162 10 54 5.40 76.7 844 15 Non-ag. laborer 2 6 3.00 55.6 630 210 8 40 5.00 52.5 1,421 28 Arts, Crafts 6 25 4.20 85.3 1,166 280 - - - - - - Services 11 57 5.18 78.8 1,868 361 2 10 5.00 82.8 1,956 39 Remittances 3 17 5.67 61.5 845 149 - - - - - - Trade 6 - 17 2.83 60.6 1,862 657 5 40 8.00 50.6 4,207 52 Transport 2 26 13.00 51.4 10,055 773 - - - - - - Profession 5 20 4.00 73.3 607 152 1 6 6.00 70.7 870 14 Others - - - - - - - - - - - - Total 158 731 4.63 83.7 1,494 323 99 630 6.36 68.5 2,505 39 OON RAMPURA Main Income Sources House- Per- P/H Main/Total Income/H Income/P House- Per- P/H Main/Total Income/H Income/P holds sons Income(%) (Rs.) (Rs.) holds sons Incone(%) (Rs.) (Rs.) Cltvtn.,[ Operator 52 391 7.52 82.7 2,466 328 62 362 5.83 89.8 1,135 19 AMIlS., Non-op. owner - - - - - - _ _ _ _ _ _ Dairy Laborer 48 244 5.08 85.6 760 149 22 103 4.68 79.8 495 10 Non-ag. laborer 17 92 5.41 70.9 1,086 201 Arts, Crafts 1 3 3.00 80.8 743 248 6 50 8.33 52.8 4,663 56 Services 53 332 6.26 79.0 1,704 272 7 39 5.57 64.3 1,129 20 Remittances 9 36 4.00 71.0 1,042 261 27 135 5.00 76.9 952 19 Trade 3 20 6.67 53.7 2,917 438 14 73 5.21 56.8 566 10 Transport 3 10 3.33 79.9 457 137 - 44 4.89 44.3 814 16 Profession 2 9 4.50 100.0 1,950 433 _- - - - - Others 1 2 2.00 117.7 765 383 - - - - - - Total 189 1,139 6.03 80.1 1,579 262 147 806 5.48 64.3 1,075 19 SANALDIVI PATADIA Main Income Sources House- Per- P/H Main/Total Income/H Income/P House- Per- P/H Main/Total Income/H Income/P holds sons Income(b) (Rs.) (Rs.) holds nons Income(%) (Rs.) (Rs.) Cltvtn., Operator 133 873 6.56 - 2,003 305 74 503 6.80 - 1,212 178 Anmls., Non-op. owner - - - - - - _ - - - Dairy Laborer 5 23 4.60 - (560) (122) 8 40 5.00 - 629 126 Non-ag. laborer 20 122 6.10 - 1,209 198 - - - - Arts, Crafts - - - - _ - - _ _ _ _ _ Services 22 178 8.09 - 1,580 195 1 1 1.00 - 1,687 1,687 Remlttances 10 51 5.10 - 985 193 - _ _ _ _ _ Trade - - - - - - 1 4 4.00 - 3,080 770 Transport - - - - - - - - Profession 1 6 6.00 - (625) (104) _ _ _ _ _ _ Others 1 3 3.00 - (1,250) (417) _ _ _ _ _ _ Total 192 1,256 6.54 - 1,802 276 84 548 6.52 - 1,184 363 JAMBUA Main Income Sources House- Per- P/H Main/Total Income/H Income/P holds sons Income(% (Rs.) (Rs.) Cltvtn.,f Operator 118 834 7.07 - 2,819 399 Anals., Non-op. owner - - - - - Dairy Laborer 38 274 - - 2,998 416 Non-ag. laborer 6 59 9.83 - 2,518 256 Arts, Crafts - - - - - - Services - 28 7.00 - 5.065 724 Remittances - - - - - Trade 2 42 21.00 25,821 1,230 Transport - - - - - Profession - - - - - Others 4 13 3.25 - 7,078 2,178 Total 172 1,250 7.27 - 3,267 450 4 VILLAGES, 1970-71 ALL 13 GUJARAT VILLAGES Main Income Sources House- Per- P/H Main/Total Income/H Income/P House- Per- P/H Main/Total Income/H Income/P holds sons Income(%) (Re.) (Rs.) holds sons Income(%) (Re.) (Re.) Cltvtn., § Operator 31 187 6.03 - 4,043 670 770 5,101 6.62 - - - Aumls., Non-op. owner - - - - - - 24 114 4.75 - - - Dairy Laborer 31 154 4.97 - 1,261 254 555 2,679 4.83 - - - Non-ag. laborer 32 147 4.46 - 1,446 315 Arts, Crafts - - - - - - 30 142 4.73 - - - Services - - - - - - 135 801 5.93 - - - Remittances - - - - - - 62 315 5.08 - - - Trade - - - - - - Transport - - - - - - 68 397 5.84 - - - Profession - - - - - - _ _ _ Others - - - - - - 14 47 3.36 - - - Total 94 488 5.49 - 2,241. 432 l,f658 9,596 5.79 - - - B. RAJASTHAN HASTEDA BA8NUSAR Rain Income Sources House- Per- P/H Main/Total Income/H Income/P House- Per- P/H Main/Total Income/H Income/P holds song Income(M) (Re.) (Rs.) holds sons Income(S) (Rs.) (Rs.) Cltvtn., Operator 105 767 7.30 83.3 1,076 148 30 198 6.60 73.8 1,553 235 Ali.., Non-Dop. owner - - - - - - - - - - - _ Dalry Laborer 4 4 1.00 100.0 801 601 54 268 4.96 56.4 509 103 Non-ag. laborer _ _ _ _ _ _ Arts, Crafts 63 361 5.41 76.7 612 113 _ - - - - Services 30 148 4.93 84.2 1,159 23 2 2 1.00 1.0 1,134 1,134 u BemIttances 41 258 6.29 79.2 1,186 22 - - - - - _ Trade 17 119 7.00 74.9 1,732 24 _ _- - - - Transport 7 52 7.43 56.4 1,508 20 - - - - - _ Profession 23 156 6.90 67.7 1,038 15 - - - - - _ Others 16 89 4.31 22.4 361 82 - - - - - _ Total 356 2,176 6.11 77.7 956 15 36 204 5.67 76.2 1,424 251 UMEDPUR ZAWM Main Income Sources House- Per- P/H Main/Total Income/H Income/P House- Per- P/H Main/Total Incose/H Income/P holds sons Income(%) (Rs.) (Rs.) holds sons Incoae(%) (Rs.) (Rs.) Cltvtn., Operator 22 107 4.86 80.9 1,319 271 70 392 5.60 70.2 898 160 Anals., Non-op. owner - - - - - - - - - - _ _ Dairy Laborer 7 31 4.43 70.6 436 99 - - - - - - Non-ag. laborer 3 11 3.67 94.0 413 113 97 562 5.59 73.4 1,343 241 Arts, Crafts 9 43 4.78 81.9 869 182 - - - - - - Services 34 92 2.71 91.0 1,172 433 3 12 4.00 79.9 1,659 415 Remittances 25 115 4.60 96.9 2,180 474 _ _ - - -. Trade 1 2 2.00 100.0 475 237 3 8 2.67 79.3 2,064 774 Transport - - - - - - 1 5 5.00 53.0 472 110 Profession - - - - - - 4 13 3.25 88.1 358 110 Others - - - - - - - - - - - - Total 101 401 3.97 90.0 1,346 339 178 972 5.46 72.7 1,162 213 DINGRI PALNA Main Income Sources House- Per- P/H Main/Total Income/H Income/P House- Per- P/H Main/Total Income/H Income/P holds sons Income(M) (Rs.) (Rs.) holds sons Income(S) (Rs.) (Rs.) Cltvtn., Operator 86 486 5.65 89.9 915 162 99 485 4.90 76.7 1,289 266 Anmls., Non-op. owner - - - - - - - - - - - - Dairy Laborer 2 11 5.50 49.4 605 110 77 368 4.78 84.0 688 144 Non-ag. laborer 19 110 5.79 56.4 821 142 - - - - - _ Arts, Crafts 1 6 6.00 46.0 670 112 25 142 5.68 75.7 1,056 186 8ervlces 2 6 3.00 89.9 1,355 445 12 59 4.92 73.6 1,315 267 Remlttances - - - - - - 22 82 3.73 74.4 793 213 Trade 4 10 2.50 98.8 1,056 423 8 45 5.62 83.6 1,056 188 Transport - - - - - - 5 18 3.60 96.5 612 170 Profession 4 24 6.00 72.5 801 133 12 52 4.33 86.7 591 136 Others - - - - - - - - - - - - Total 118 653 5.53 84.1 901 163 260 1,251 4.81 77.9 995 207 AIDAN-KA-WAS ALL 7 RAJASTHAN VILLAGES Main Income Source House- Per- P/H Main/Total Income/H Income/P House- Per- P/H Main/Total Income/H Income/P holds sons Income(%) (Rs.) (Rs.) holds sons Income(%) (Rs.) (Rs.) Cltvtn., Operator 30 227 7.57 - - - 442 2,662 6.02 - - - Anmls., Non-op. owner 2 8 4.00 - - - 2 8 4.00 - - - Dairy Laborer - - - _ _ 263 1,345 5.11 - - - Non-ag. laborer - - - - - - Arts, Crafts 8 48 6.00 - - - 106 580 5.47 - - - AD Services 2 4 2.00 - - - 85 323 3.80 - - - o Remittances 2 7 3.60 - - - 90 462 5.13 - - - Trade - - 33 184 5.58 - - - Transport - - 13 75 5.77 - - - Profession 3 12 4.00 - - - 40 255 6.38 - - - Others - _ - - - 16 69 4.31 - - - Total 47 306 6.51 - - ,096 5,963 5.44 - - - SOURCBS AND NOTES Sources: Anon. (1970), pp.55, 124; Anon. (1971), pp.31, 102; Anon. (1980), Table 8.10; Bambal (1968), p.78; Bhat and Pichhollya (1967), pp.280,184; Brahmbhatt (1977), pp.45, 104; Choudhary (1964), p.50; B. Desald (1967), p.175; M. Desal (1966), pp.36, 171; R. Patel (1964), pp.97, 150; R. Patel (1964a), pp.43, 175; V. Patel (1973), passim; Purohit (1974), p.l65; Saxena (1968), pp.90-1; Saxena and Charan (1973), pp.148-50; N. Shah (1968), p.164; Shetty (1963), p.134. For references and survey dates, see bibliography. Notes: Normally, operators "in cultivation, etc." all derived their main income from farming. An unstated number in Afawa and Dingri derived It from animal husbandry or dairying, as did one of the 76 households in Moti Khadol (2 persons; 51.3% of income from primary source; 787 Rs. per household yearly), one of the 133 in Samaldevi (2; percentage not available; about 875 Rs.), 9 of the 99 in Falna (55; 86.3%; 855 Rs.), and 9 of the 105 in Hasteda (50; 76.2%; 634 Rs.). In Nargol, all the "cultivation, etc." operators were fishing families; the "agricultural" labor all derived its main income from fishing employment; and the non-operating owners, described as having their main income as rentlers, pre- sumably own and rent out boats. For Samaldevi, the last two columns are interpolated, and very approximate for small samples (indicated by brackets). In Uoedpur, three households with no apparent income source were omitted from Table 17. Other minor differences between a few population and household totals (especially for population in Hasteda), between - Tables 16 and 17, reflect the sources, presumably because household sizes or numbers varied between visits to measure occupational and caste data. Data sets for each village are as complete as the sources permit. However, the last three columns were deliberately estimated for the total of villages, because this would involve adding up incomes for different years (of the various surveys)and different villages (with different price-levels), to misleading effect. Table 4: HDSEEID SIZE AND WEIEA, ZARIA VILLAGES (y = a + bx; y = household outlay per CU, sh/week; x = size of household in Cas). Village a h r n Sig. level (r) Doka +.1306 -.7753 -.5283 40 0.1% Dan Mzhwayi +.1409 -.5694 -.2614 32 10% Hanwa +.1821 -.6896 -.5216 32 1% Sources: Norman (1976); Simmmis (1976); Lipton (1983), as Tables 3, 4. Farm expenses deducted fram total outlay, and obviously inaccurate or exceptional cases exclixied. TABLE 5: SOCIAL STATUS AND HOUSEHOLD SIZE, RURAL W. INDIA, 1961-65 GUJARAT vILLAaO AFAWA ANKODIA MOTI KHADOL OON RANPURA SAKALDEVI ALL 6 VILLAGES Por- HouseP/H P H P/H P H P/H P H P/H P H P/H P H P/H P H P/H sons holds Caste Group Hlgh Hindus 13 3 4.33 30 7 4.29 38 9 4.22 7 1 7.00 3 1 3.00 91 25 4.33 Intermediate 428 68 6.29 1116 75 6.38 371 72 5.15 107 12 8.92 439 75 5.85 173 179 6.55 3634 581 6.25 Other Caste 695 166 4.19 148 33 4.48 110 24 4.58 225 40 5.62 226 46 4.91 6 1 6.00 1410 310 4.55 Hindus Scheduled Castes 79 16 4.39 \ 34 9 3.78 150 29 5.17 134 24 5.58 80 12 6.67 6 Schdued ries 39 226 51 4.3 4…9250 48 5.21 -- - -953 191 4.99 Non-Hindu 13 2 6.50 177 44 4.02 397 59 6.73 4 1 4.00 591 lO 5.58 VILLAGB TOTAL 1215 225 4.76 1533 268 5.72 730 158 4.62 1138 189 6.02 806 147 5.48 1259 192 6.56 6681 1209 5.53 RAJASTHAN AIDAN-KA-WAS DINGRI FALNA HASTEDA UVEDPUR ZAWAR ALL 6 VILLAGES VILLAGE Pron- hous p H P/H P H P/H P H P/H P H P/H P H P/H P H P/H Caste Group High Hindus 9 3 3.00 17 6 2.83 101 25 4.04 549 111 4.95 160 42 3.80 33 9 3.67 864 196 4.41 Intermediate 224 32 7.00 360 62 5.81 566 121 4.68 328 42 7.81 23 11 2.08 9 2 4.50 1510 270 5.59 Other Caste 41 7 5.86 17 3 5.40 382 71 5.10 510 87 5.88 139 29 4.79 52 11 4.73 1121 208 5.30 Hindus Scheduled Castes 29 5 5.80 30 5 6.00 219 42 5.21 268 48 5.58 58 13 4.4f 9 3 3.00 613 116 5.28 Scheduled Tribes - --29 42 5.45 __ __ __ 139 22 6.32 74 4 3.50 869 153 5.68 1251 221 5.66 Non-Hindu -- - -- - 3 1 3.00 254 46 5.52 10 5 2.00 -- -- - … 267 52 5.13 VILLAGE TOTAL 303 47 6.45 53 118 5.53 1251 260 4.81 2048 356 5.75 404 104 3.87 972 178 5.46 5631 1063 5.30 Sources: Patel (1964), p.44, and (1964a), p.42; B.Desal (1967), p.53; (1968), pp.50, 72; N. Shah (1968), pp.50, 72; Anon. (1970), p.37; Brahkblatt (1977), p.30; Choudhary (1964), p.49; Bazena and Charan (1973), p.27; Bhat and Pichhollya (1967), p.49; Anon., (1971), p.32; S.D. Purohit (1974), p.45; M. Desal (1966), p.18. Note: High Hindus: Brahains, Jains (classified as High Hindus by all these authors), Bania, Maheshwarl, Garg. Intermediate: the great majority are Rajput, Patel and Jat. TABLE 6: i1S AND POVE : HIGHILY TEITAIVE SU[UQRY OF LINlEl' AND DELINKER VARIABLES Variable Explains positive poverty Explains negative poverty-NES Explains 'paradoxical' negative Explains fade-out among very poorest i link in LICs now? link in PI-NiCs,and reversal? status-NUS link(LICs and PI-NDCs)? of today's positive poverty-B links? Nortality IR, child: more for peor. I1R, child: much more for poor. XI, child: rises very sharply. Rural/remote 'status-groups": higher Hinders explanation un- Helps explain link. If less only among ultra-poor: helps adult (and child) mortality. Unakilled less replaceent fertility difference now, hinders explain. assetless labor's high IXR, given exceeds sib crowding. explanation of reversal. NEP(?), helps explain 'oddly' low N8 C>uple Falls briefly, then rises Female ed. - less important; Helps, as lowest status, bad Probably helpsexplain - physical Fertility with rising EP. Linker if fertility adapted downwards health and hygiene, high INER, fertility reduction perhaps outweighed and only if populations to times, conditions of lower replacement fertility by more 'insurance" births. mainly poorer than turning- poverty: linker, linked. point. NBre post-primary female education for less- poor cuts fertility linker. Ybration Poor marry earlier (less Poor married latcr (cf. High status goes with longer Underfed ultra-poor may have later and duration educational delay)slinker couple fert., non-lfmily educn. A (delinker) and less menarche, earlier menopause, more of unions members): linker, marital disruption (linker), early widowhood: helps explain. Cplezity Somewhat more for better- Usually affected only Clear positive status-complexity iMuch less complexity among landless (non-nuclearity) off: delinker, if linked gentry: stronn 1 Inker, link. Major part of explanation. and assetless (no will-shaking) could to WSP rather than atatus. Possible bicjqer spread help explain. "down" now in i.i;; w,o.jlrl help) expl.i: reversal iigration Probably rural linkers Probably urbain linker; Too little researched to guess. Ultra-poor do little long-term urbanposition not clear, rural position riot clear individual migration; this would help explain any fade-out of links or de-links. laily cycles Very sial incidence of Low ased siLnlleness Rigid status barriers to child- Too little research to guess. aged singleness among amonq poor delinked; ren's careers could push low- poorl high child/adult lower child/adult ratLos, status couples to cut childbirths ratio vs poverty "stage" with poverty, i:s.'rfnlly in early family-cycle (linker). of cycle. Both linkers, linked. mon-family Decline of resident "Massive trdnsfer" of Low-status teenage lodgers, Too little research to guess. household service/apprenticeship survonts from poor to less- servants, in high-status homes: members helps explain reversal poor honls: holds, major linker, but weaker in LICs than of traditional NiS-aff- linker, directl y .nd by PI-MDCs. luence link. d'lI.yLnj murriaqe dmonq poor. 101 TABLE 7: AGE-DISTRIBUTION BY HOUSEHOLDS (GROUPED BY OUTLAY PER PERSON): W. INDIA, 1972-3 POOREST POOREST SBCOND RICHEST DECILE QUINTILE POORQST QUINTILE RURAL GUJARAT: % 0-4 17.0 15.7. 15.2 10.0 % 5-14 33.6 33.7 30.8 23.6 % 60+ 3.2 3.5 4.2 7.0 Dependency Ratio 1165 1197 1007 682 RURAL MAHARASHTRA: % 0-4 17.5 16.8 15.7 12.4 % 5-14 33.3 33.4 30.0 24.2 % 60+ 5.7 5.4 6.6 6.9 Dependency Ratio 1304 1253 1100 793 URBAN GUJARAT: % 0-4 14.8 14.4 13.3 7.4 % 5-14 35.3 33.7 29.4 18.4 % 60+ 3.4 3.5 4.4 5.9 Dependency Ratio 1150 1065 890 469 URBAN MAHARASHTRA: % 0-4 15.3 15.2 13.9 6.8 % 5-14 34.1 32.4 28.5 13.7 % 60+ 5.2 5.2 4.6 6.4 Dependency Ratlo 1203 1120 887 370 Source: N8S, 27th Round, reported In Visaria (1977), Appendiz, Tables 1-3. TABLE 8(a): CHILD/ADULT RATIOS AND POVERTY, INDIA 1972-3 RURAL MDNTHLY GUJARAT MAHARASHTRA RAJASTHAN ALL-INDIA EXPENDITURE Z of Ave. Child % of Ave. Child x of Ave. Child x of Ave. Child PER PERSON . house- Size h/hold house- Size h/hold house- size h/hold' house- size h/hold (Rs) holds of h/h ratio holds of h/h ratio holds of h/h ratio holds of h/h ratio 0-13 0.4k 7.6 .56 1.3 7.0 .51 0.7 4.7 .54 1.5 5.6 .54 13-15 0.3* 4.9 .48 1.1 5.9 .56 0.9 5.8 .62 1.2 5.8 .52 15-18 1.2 6.3 .53 2.9 6.1 .52 1.9 5.5 .53 3.0 5.9 .52 18-21 2.5 6.7 .52 4.6 6.2 .53 4.0 6.0 .51 4.6 5.8 .50 21-24 3.4 6.2 .51 6.1 6.0 .51 3.7 5.9 .51 5.9 5.7 .49 24-28 6.6 6.8 .52 9.8 5.9 .47 8.2 6.2 .44 9.4 5.6 .47 28-34 13.2 6.5 .49 15.5 5.7 .46 12.2 5.8 .46 14.9 5.4 .45 34-43 19.4 6.1 .44 19.9 5.3 .42 16.6 6.0 .46 18.7 5.3 .42 43-55 17.7 5.7 .43 17.0 5.2 .41 17.8 5.7 .45 16.5 5.1 .40 55-75 18.0 5.4 .39 12.7 4.9 .38 16.3 5.4 .41 12.9 4.8 .37 75-100 9.4 4.6 .36 5.2 4.1 .35 8.9 5.2 .39 6.1 4.5 .35 100-150 5.5. 5.1 .31 2.7 4.4 .35 6.6 4.8 .36 3.5 4.3 .33 150-200 0.9 2.4 .21 0.6 2.6 .22 1.0 2.6 .24 0.9 3.2 .26 200+ 1.5* 4.8 .16 0.6 3. 1.2 4.8 .33 0.6 3.9 .29 ALL 100.0 5.8 .43 100.0 5.4 100.0 5.6 .44 100.0 5.2 .43 (3381) (5249) (2285) (72270) Source: Sarvekshana, Jan. 1979, pp. S-294, 300, 305, 357, 360, 364, 369. Notes: (1) Sub-sample sizes for rural Gujarat: 11 in per person outlay group, 0-13 Rs. per month; 7 in group, 13-15 Rs per month; and 15 in group for 200 Rs. and above. All these small sub-samples are marked * above. All other cells contain at least 18 households. (2) Entries in columns "Z of households" are for the whole State (or all-India), not for the sample. TABLE 8(b): CHILD/ADULT RATIOS AID POVERTY, INDIA 1972-3 URBAN MONTHLY | GUJARAT MAHARASHTRA RAJASTHAN ALL-INDIA EXPENDITURE Z of Ave. Child Z of Ave. Child Z of Ave. Child % of Ave. Child PER PERSON House- Size h/hold house- Size h/hold house- Size h/hold house- Size h/hold (Rs) holds of h/h ratio holds of h/h ratio holds of h/h ratio holds of h/h ratio 0-13 i 0.1 1.00 .00 0.6 4.9 .45 0.3 3.4 .36 0.3 4.8 .49 13-15 -* - - 0.3 6.6 .51 0.1* 3.5 .26 0.3 5.9 .51 15-18 0.5 5.2 .62 0.7 6.4 .54 0.7 6.7 .55 0.8 6.1 .54 18-21 0.4 7.5 .52 1.1 7.1 .49 i 1.2 6.7 .52 1.4 6.4 .51 21-24 0.5 7.2 .50 1.7 6.9 .48 1.8 6.9 .55 2.1 6.5 .51 24-28 2.0 6.3 .46 3.7 6.4 .45 3.8 7.2 .50 4.2 6.4 .48 28-34 7.3 6.8 .50 5.7 6.3 .47 7.8 6.0 .47 8.4 6.2 .47 34-43 15.5 6.3 .45 10.5 5.9 .43 15.1 6.1 .48 13.8 5.8 .44 43-55 21.5 5.9 .39 13.4 5.5 .41 17.3 5.8 .44 15.9 5.4 .40 55-75 23.2 4.9 .36 17.2 5.2 .38 17.5 3.3 .39 18.0 4.7 .36 75-100 15.1 3.9 .29 14.0 4.2 .32 13.8 4.1 .37 13.4 3.8 .31 100-150 8.4 2.9 .22 17.3 3.3 .25 12.2 3.4 .32 12.5 3.1 .26 150-200 2.6 2.3 .11 7.0 2.9 .20 4.3 2.7 .35 4.6 2.7 .21 200+ 2.0 1.5 .12 6.7 2.8 .20 4.0 2.6 .26 4.3 2.4 .19 ALL 100.0 5.1 .38 100.0 4.7 .37 100.0 5.0 .42 100.0 4.7 .39 (1990) (6181) (2392) (52820) Source: Sarvekshana, Jan. 1979, pp.S-294, 297, 300, 305, 357, 360, 364, 369. Notes: (1) Sub-sample sizes for urban Gujarat: 2 for 0-13 Rs; zero for 13-15; 6 for 18-21; 11 for 21-24. Urban Rajasthan: 7 for 0-13 Rs. month, 3 for 13-15, 17 for 15-18. All these small sub-samples are marked * above. All other cells at least 18 households. (2) Entries in columns "1 of households" are for the whole State (or all-India), not for the sample. TABLE 9(a): URBAN FEMALE DEMOGRAPHY BY INCOME CLASS, INDIA, 1972-3 MONTHLY YAHARASHTRA GUJARAT RAJASTHAN ALL-INDIA OUTLAY % of males Child- % of Males Child- % of Males Child- % of Males Child- PBR80N House- per ren H/h per ren H/h per ren H/h per ren holds Female per in Female per in Female per Female per (Rs) In (Adults) Adult State (Adult) Adult State (Adult) Adult (Adult) Adult State Fe- Fe- Fe- Female male male male 0-13 0.6 1.09 1.74 (0.I)(b)(l.00) (0.00) (0.3) (0.61) (0.91) 0.4 0.94 1.84 13-15 0.4 0.81 1.86 ( - ) ( - ) ( - ) (0.1) (0.81) (0.63) 0.3 0.94 1.97 15-18 0.7 0.83 2.14 (0.5) (0.77) (3.53) 0.7 0.96 2.35 0.8 0.83 2.12 18-21 1.1 0.83 1.74 (0.4) (0.86) (2.05) 1.2 1.07 2.29 1.4 0.80 1.86 21-24 1.7 1.02 1.88 0.6 1.10 2.10 1.8 0.91 2.30 2.1 0.94 1.99 24-28 3.7 1.01 1.68 2.0 1.04 1.72 3.8 0.89 1.86 4.2 0.96 1.84 28-34 5.7 1.05 1.81 7.3 1.05 2.04 7.8 1.01 1.77 8.4 1.01 1.74 34-43 10.5 1.04 1.54 15.5 1.01 1.66 15.2 1.04 1.85 13.8 1.00 1.56 43-55 13.3 1.10 1.48 21.5 1.07 1.30 17.3 1.04 1.62 15.9 1.06 1.37 55-75 17.2 1.16 1.32 23.1 1.04 1.12 17.5 1.09 1.35 18.0 1.13 1.20 75-100 14.0 1.35 1.10 15.1 1.11 0.86 13.8 1.23 1.31 13.4 1.32 1.03 100-150 17.3 1.61 0.86 8.4 1.32 0.65 12.2 1.23 1.06 12.5 1.55 0.88 150-200 7.0 1.65 0.67 2.6 1.26 0.27 4.3 1.51 0.81 4.6 1.67 0.70 200+ 6.7 1.65 0.67 2.1 2.07 0.42 4.0 1.76 0.96 4.3 1.76 0.65 ALL 6181(a) 1.23 1.32 1990(a) 1.08 1.29 2392(a) 1.10 1.54 5282da) 1.15 1.35 Source: Sarvekshana, Jan. 1979, Tables 1/U. Notes; (a) "% of all households" is clearly 100; entry shows number of households sampled. (b) Bracketed figures indicate sub-sample below 10. This table and table 12 allow all sub-sample sizes to be calculated. TABLE 9(b): RURAL FEMALE DEMOGRAPHY BY INCOME CLASS, INDIA, 1972-3 VONTHLY MAHARASHTRA GUJARAT RAJASTHAN ALL-INDIA OUTLAY % of males Child- % of Males Child- 5 of Males Child- 9 of males Child- PER House- per ren H/h per ren H/h per ren H/h per ren PZRSON holds Female per In Female per in Female per Female per (as) in (Adults) Adult State (Adult) Adult State (Adult) Adult (Adult) Adult state Fe- Fe- Fe- Female male sale male 0-13 1.3 0.55 1.62 0.4 0.84 2.35 0.7 0.85 2.16 1.5 0.79 2.10 13-15 1.1 0.83 2.34 (0.3) (1.77) (2.58) 0.9 0.75 2.80 1.2 0.90 2.07 15-18 2.9 0.94 2.09 1.2 1.03 2.29 1.9 1.01 2.28 3.0 0.91 2.05 18-21 4.6 0.95 2.21 2.5 0.94 2.14 4.0 1.01 2.10 4.6 0.94 1.94 21-24 6.1 0.93 2.03 3.4 1.00 2.05 3.7 0.93 2.01 5.9 0.94 1.84 24-28 9.8 0.96 1.74 6.6 1.05 2.22 8.2 0.98 1.57 9.4 0.96 1.72 28-34 15.5 0.92 1.64 13.2 1.02 1.92 12.2 1.02 1.72 14.9 0.97 1.61 34-43 19.9 0.96 1.43 19.4 0.99 1.60 16.6 1.00 1.70 18.7 1.00 1.47 43-55 17.0 0.97 1.37 17.7 1.04 1.56 17.8 1.04 1.65 16.5 1.03 1.36 55-75 12.7 0.99 1.21 18.0 1.02 1.27 16.3 1.12 1.50 12.9 1.06 1.21 75-100 5.2 1.00 1.06 9.4 0.89 1.05 8.9 1.05 1.33 6.1 1.10 1.14 100-150 2.7 1.09 1.12 5.5 1.20 1.01 6.6 0.97 1.10 3.5 1.19 1.08 150-200 0.6 1.61 0.75 0.9 1.92 0.77 1.0 1.61 0.86 0.9 1.27 0.81 200+ 0.6 1.42 1.21 1.5 0.99 0.37 1.2 1.53 1.25 0.1 1.30 0.93 LL(a) 5249 (a) 0.96 1.54 3381(a) 1.02 1.55 2285(a) 1.03 1.62 2270(a) 1.01 1.50 Source: Sarvekshana, January 1979, Table 1/U. Notes As Table 9(a). TABLE 10: DISTRIBUTION OF HOUSEHOLDS BY PER CAPITA INCOME AND HOUSEHOLD INCOME, KERALA, 1978-9 ROUSEHOLD INCOME (Rs/MONTH) Per Capita Income TOTAL (Rs/Month) 99 100-240 250-499 500-749 750-999 1000+ 0-19 95 38 1 0 0 0 134 20-39 55 494 56 0 0 0 605 40-59 0 259 197 23 5 1 485 60-79 27 0 241 70 7 3 348 80-99 0 54 93 59 24 4 234 100-149 0 0 36 118 52 34 240 150-199 0 22 29 50 26 43 169 200+ 0 0 7 32 30 109 178 TOTAL 177 887 660 352 143 194 2393 Source: Data supplied by the authors, and used by them to prepare Mathew and Scott (1980), a survey of 3 urban and 9 rural communities. TABLE 11: TOTAL HOUSEHOLD EXPENDITURE AND SIZE, ZARIA, 1970-1 Total Household Persons peraverage household size (no. of households in brackets) expenditure (sh/wk) Doka Dan Mahawayi Hanwa All three 0-30 4.7 3.8 3.4 4.2 30-60 6.9 6.4 6.1 6.7 60-90 11.0 7.6 10.1 9.3 90-120 5.1 9.8 10.7 10.1 120-150 - 13.6 10.2 11.0 150+ - 9.9 12.4 11.0 TOTAL 6.9 7.3 10.0 8.0 Source: Simmons (1976), pp. 84, 86. Notes: Consumers averaged over two survey weeks about six months apart. TABLE 12: MHS-NEP RELATIONSHIPS: INDIA, 1972-73 Rural households Urban households State Sample Percentage of households for which: Sample Percentage of households for which Size At poorest Poor and At top Size At poorest Poor and At top end: no above: end: no end: no above: end: no relationship relationship relationship relationship relationship relationship Andhra P. 5887 10.9 - 17.6 82.4 - 89.1 - 4654 9.2 - 19.7 80.3 - 90.8 - Assam 2591 2.1 - 5.5 92.7 - 96.1 1.8 1500 1.4 - 3.5 96.5 - 98.6 - Bihar 5739 See note (a) 0.5 3693 5.7 - 10.1 89.9 - 94.3 - C-ujarat 3381 7.9 - 14.5 77.6 - 84.2 7.9 1990 3.5 - 10.8 89.8 - 96.5 - |Haryana 2234 See note (b) - 1676 1.8 - 3.1 96.9 - 98.2 - Himachal P. 1190 2.4 - 6.9 93.1 - 97.6 - 702 Small sample I Jaa u, Kashuir 4307 10.6 - 21.5 78.1 - 89.0 0.4 2688 0.1 - 1.5 98.5 - 99.9 - i Karnataka 3333 13.1 - 21.0 75.2 - 83.1 3.8 2466 10.9 - 20.9 79.1 - 89.1 - Kerala 3789 19.2 - 29.2 70.8 - 80.8 - 1407 12.5 - 20.0 80.0 - 87.5 - Madhya P. 5739 2.8 - 4.3 90.7 - 92.2 5.0 3213 2.2 - 4.2 95.8 - 97.8 - Maharashtra 5249 5.3 - 9.9 86.8 - 91.4 3.3 6181 1.7 - 2.7 97.3 - 98.3 - Meghalaya 1025 6.5 - 19.8 79.3 - 92.9 0.6 - 0.9 502 Small sample Orissa 3312 4.9 - 8.0 76.9 - 88.1 7.0 - 15.1 1861 19.2 - 32.5 67.5 - 80.8 - Punjab 3064 4.5 - 9.6 87.3 - 92.4 3.1 1697 2.4 - 5.4 94.6 - 97.6 - Rajasthan 2285 See note (c) 1.2 2392 15.7 - 30.9 69.1 - 84.3 - Tamil Nadu 5984 6.3 - 12.3 87.7 - 93.7 - 3541 7.0 - 12.5 87.5 - 93.0 - Tripura 1085 10.8 - 25.1 73.3 - 87.6 1.6 647 Small sample Uttar P. 7985 5.3 - 10.7 88.7 - 94.1 0.6 5982 3.0 - 6.0 94.0 - 97.0 - W. Bengal 4895 6.4 - 12.2 87.5 - 92.3 0.3 4465 8.1 - 14.3 85.7 - 91.9 - Union Terrs.(d) 527 Small sample 2286 2.7 - 7.2 92.8 - 98.3 - All-India (e) 72270 2.5 - 3.8 95.5 - 96.8 0.7 52820 2.8 - 4.9 95.1 - 97.2 - Source: See fn. 11. Notes: The entry for rural Uttar Pradesh for example, should be read: "The MEP groups comprising the poorest 10.7% of households showed no systematic fall in MHS group means as MEP rose. From the highest of these very poor groups (i.e. the poorest 10.7% of households minus the very poorest 5.3%), to the group immediately below the very richest 0.6% of households, MHS consistently fell as group mean MEP rose. This relationship therefore applied to the 'middle' 88.7%-94.1% of households." There are 14 MEP groups in each State. (a) Nearest approximation to entries: 2.6-6.3; 93.2-95.9; 0.50. However, among the eleven "middle" MEP groups, two show slight reversals of the otherwise steady downtrend of group mean MRS with rising NIP. (b) No clear trend. (c) As (a), 11.2-19.4; 79.7-87.6; 1.2. However, one slight reversal takes place among the eight "middle" NEP groups. (d) Delhi, Goa-Daman-Diu, Pondicherry; also (urban only) Chandigarh. 110 FOOTNOTES 1. Poverty and ultra-poverty are here treated as oharaoteristios of households, for three reasons. First, it is by the household that incomes and spending are pooled and surveyed. Second, policy leverage on intra-household distribution is small. Third, severe deprivation of individuals - e.g. girls - within households sufficiently well-off to feed all their members adequately is probably muoh rarer than is usually argued - and where present is not readily amenable to polioy, except by enriching the household overall. See Lipton, 1983, Sec. III(a). 2. "Poverty" here means insuffioient income (or outlay) to provide household members with 100% of 1973 average FAO/WHO caloric requirements of their age, sex and activity groups, when the household allocates inoome (or outlay) among foods, and betweeen them and non-foods, typioally for households with its size, age- and sex-structure, and income (or outlay). "Ultra-poverty" replaces "100%" with "80%" in the above definition. Poverty and ultra-poverty may be assessed by surveying (per-person or per-CU) household income or outlay, food oonsumption, food/outlay ratios, caloric intake relative to requirements, or anthropometrio status. See Lipton, 1983, Sec. I. 3. "Denominator" in that household resouroes might be divided by number of members, by number of adult-equivalents, or not at all. "Resouroes" are used to oover any of several alternate measures of family access to the means to avoid poverty: to income, outlay, or - quite plausibly where 70-80% or more of outlay is on food - calories. 4. For classifioation of households, see Laslett and Wall, 1972, pp. 31, 41-2, and below, fn. 16. 5. These two offsets are not universal. Until recent times (pp. 7-8) - and probably still, where neither publio health nor income levels have started to improve - bigger households tended to have higher adult/child ratios, and to be better-off. In such ciroumstances, ranking of households, household deciles, or areas in rising order of resources-per-household would give even less guidanoe, if possible, to ranking in terms of poverty incidence. 6. However, even if we know nothint about oausality, we shall looate "poverty projeots" better ir we know the demographic charaoteristics of the poor, and where those oharaoteristics are to be found. See Sec. V. 7. Wachter and Laslett, 1978, pp. 76-7, argue that the increase is only "at first flush" and is shown, by simulation of evolving populations, to be statistically unsurprising. However, the simulation predicts ranges of oontinuous change in MHS; Laslett'J data appear to show a marked disoontinuity around 1750 or so. See also Nakane, 1972, p. 531. 8. This implies that positive linka of poverty to household size within a status-group of households are likely to be even stronger than in total populations of households - especially if the group comprises a large part of the population, and has little soope for inoreasing assets and status within its ranks as M1S rises. Henoe the impaot of MRS on the logarithm of poverty-risk, in large rural Indian saples in 1968-70, was much larger for the 900-odd oasual-labor households than for the 2350-odd cultivator households, and larger for the latter than for the 4100 households pooled from all groups: Gaiha, 1983, Tables 13, 18, 19. 9. For the seventeen villages where NHS and income-per-person oan be estimated from Table 3, allowing for ohanges in the price-index for agricultural labor in the two States between survey yearst and oonverting from Gujarat to Rajasthan prioes by using the data in Chatterjee and Bhattacharya, 1974, pp. 344-6, we obtained the correlation between village MRS and village average real inoome-per-person: r = -.2403, n 2 17, n.s. 10. Calculated by B. Longhurst from Norman et al., 1976. 11. Sarvekshana, Jan. 1979, pp. S293-8305 and S356-S369. The details are set out in Table 12. 12. This study indirectly suggests that marital fertility helps explain both parts of our paradox, in Kerala at least; for it decreases as NCP rises (Zachariah and Kurup, 1982, Table 3). 13. In both cases, MHS for craftsmen is now below village average MHSs. This contrasts with the opposite tendenoy in "proto-industrializing" NDCs (Andorka and Farago, 1983, p. 295; Wall, 1983, p. 448, and 1983a, p. 388). Perhaps the lower status and household size of artisans in many LICs today "pushes" bigger villages, with a greater artisan oomponent, towards lower MHS? 14. Jains are classified as 'high Hindus' in these surveys, though some would consider them a distinct religion. In Table 5, we have made minor ohanges to two surveys, to reconcile their classifioation of castes with that used by the other ten. 15. A tradition of arranged marriages - linked to the risk of being left as a single-parent family with no obvious means of support - imposed very low illegitimacy rates on many pre-industrial sooieties in the past (Laslett, 1971, pp. 142-5). It is reasonable to hypothesize that this applies even more strongly to most LIC societies today, so that lower marital fertility normally means lower oouple fertility, not high bastardy rates - but this requires research and evidence. 16. We use the classifioation of co-resident families used by Laslett, 1972a, p. 31. (A) Nuclear or simple a (1) sole (solitary), or (2) "no familY' (*resident unmarried siblings or other relatives, or unrelated persons), or (3) simDle (married couples or single parents with ohildren). (B) Complex = (4) extended (nuclear plus relative(s) of an earlier generation: extended uvwarde, e.g. by the presence of a nuclear spouse's parent or aunt of a later generation; or extended 112 downwards, e.g. by a nuclear spouse's grandchild; or extended laterally, e.g. by a nuclear spouse's sibling or cousin of the same generation); or (5) multiDle, with two or more related conjugal families, again up, or down, or on one level with the household head; a special case of the latter is the a "frerechem, with no "member of an earlier generation [but] married siblings connected entirely through the filial linkage of each to a conjugal unit no longer represented in the household" (ibid., p. 30). There remains a handful (6) Indeterminate. In this paper "household" means "co-resident family plus resident servants, lodgers, or other unrelated inmates"; "family" always means "co-resident family" unless otherwise stated. 17. On its own, this first effect does not alter the absolute gap between rich and poor HHS, but does make rich households' MHS - if initially larger than poor households' HHS, as in PI-NDCs - somewhat smaller (or less big) relatively to, i.e. as a ratio to, poorer households' HHS. Suppose that children either die in the first year of life, or survive through adolesoence. Now, let IMR fall due to the conquest of illnesses that had earlier killed 1 infant in 10, irrespeotive of poverty. Then IMR for the rich might fall from 200 to 100, and for the poor from 300 to 200. Assuming ten births per couple rich or poor, MHS, with all children and parents at home, would rise from 10 to 11 for the rich, and from 9 to 10 for the poor. The gap stays the same, but "rich" MHS falls from 111% to 1105 of "poor" MKS; other changes, e.g. decline of resident servants, could then easily bring it below 100%. (If we allow for the fact that subsequent mortality (between infancy and adolescence) probably was similarly reduoed by income-neutral health improvements, these would then shrink the rich/poor MHS ratio more - but still not affect the absolute MHS gap). 18. In 100 English oommunities, 1574-1821, community HHS was correlated with (1) proportion of households with servants (simple r = .599), (2) proportion of gentry (.528), (3) servant/population ratio (.437), (4) child/population ratio (-.335) and (5) proportion of households headed by married oouples (.296). In 382 oommunities in Suwa county, Japan, 1671-1870, comparable significant results were (1) .444; (2) n.a.; (3) n.a., but .402 for servant/household ratio (two other servant indicators also > .25); (4) n.a., but r for ratio of children under 10 to married women aged 21-40 was +.319 (and for unmarried children/household heads ratio, .529); (5) .600 (two other indioators of marital inoidence were p.83); (6) proportion of households with three or more generations, .529; (7) ditto without resident kin, -.684 (100 English, .0031); (8) proportion of females aged 12-40 married, -0.167(1). All significant at 1%, except (4) for UK (55). Laslett, 1972, p. 155; Hayami and Uchida, 1972, p. 492. 19. One can add to Schultz's reasons for implying "No". Windfall oil-based incomes reduce the oosts of child-oare (especially by immigrants); so does the relatively small likelihood of "modernizing away" complex families. Yet I would answer "Yes": higher inoome oan permit a Government to provide (a) social-seourity systems that reduce even the poorer Libyan's need to rely on his children for his old age, and (b) post-primary education to mothers, reducing fertility (p. 22) - and to children, raising the costs of rearing them. 113 20. Indeed, in the poorest countries, women with some primary sohooling have slightly higher fertility than women with no eduoation - oonsistently with the inverse U; see Birdsall, 1980, p. 49. 21. See fn. 16; higher adult death-rates would also reduce incidenoe of extended families, but the efrect on multiple families, up or down from the household head, is greater. 22. This is an aspeot of "Sanskritization": Srinivas, 1962. 23. Dirrerences significant at 5% and 10% respectively. Average holding per person was also somewhat bigger in complex than in nuclear oo-resident families. 24. The wrong assumptions that (a) most LIC cities are dominated by recent migrants or (b) migrants, even long-settled ones, have relatively small households, may well be to blame for the widespread belief that in LICs urban households are generally and signifioantly smaller (and less likely to be complex) than rural households. 25. For seven observations each on student migrants (as proportion of total population) and on working migrants (proportion of adults), the simple r with average household size was, respectively, -.5970 and -.6825: see Connell et al., 1976, p. 193, and Table 3 of this paper for MHSs. 26. In Villgraten village, Austria, in 1781, "households with more grown-up children, other things equal, had fewer servants ... Peasants ... (all owning 10-14 cattle and/or horses) who had no children above (10 averaged] 2.71 servants (28 cases); with one child, they had 2.45(11); with two, 1.64(14); with three, 1.06(18); with four, 0.72(18); and with five, 0.07(14)" (Schmidtbauer, 1983, p. 355). 27. It might be argued that this was a fault in the right direotion. Suoh an argument is not mere rhetorlc because risks of irreversible harm from poverty-induced shortages of dietary energy are much greater among under-fives than at later ages. 28. "Dependency ratio" usually means "persons not aged 15-59, as a proportion of persons aged 15-59". Sometimes "over 15" or "over 10" replaces "15-59"; where this happens, it is stated. It makes little difference to the overall relationships between poverty and the ratio. Nor does the distinction between "family" and "co-resident household"; few households at risk of poverty have servants or lodgers. 29. In 1973-4, children in rural Gujarat comprised 48.4% of rural households with below 34 Rs/person/month; 45.5% at 34-43 Rs; 41.7% at 43-150 Rs; and 43.4% above 150 RB. (This accords well with the 1972-3 data given by Visaria; only above the 6th deoile by inoome-per-person did the ratio fall below 40%.) For rural Haharashtra, the corresponding data were 50.0%, 46.7%, 39.2% and 31.0%, and for rural Rajasthan, 49.1%, 46.0%, 42,4% and 38.2%. In urban Gujarat, 52.1% of members of households spending below 34 Rs/person/aonth were ehildren; 46.0% at 34-43 Rs; 38.9% 43-150 Rs; and 24.1% at 150+. For Maharashtra 114 the corresponding figures were 50.9%, 44.6%, 36.6% and 19.2%, and for Rajasthan, 49.1%, 50.9%, 40.2% and 21.9%. Caloulation by K. Affan from NSS 28th Round (Oot. 1973-Sept. 1974, No. 240), pp. 11, 13, 81. 30. Sarvekshana, Jan. 1979. For example, we found similar "reversals" for rural Bihar (child/household ratio for households spending 0-13 Rs/person/month, 0.54 (2.0% of all 5739 sampled); 13-15, 0.48 (1.6%); 15-18, 0.50 (3.7%); and falls thereafter); Madhya Pradesh (at 0-13 Rs/person/month, a sample of 78 households oomprised 52% children; at 13-15 (sample of 115), 55%; and falls thereafter); and Punjab (0-13 Rs - 10 households - 37%; 1-18 - 8 - 46%; 18-21 - 21 - 50%; 21-24 - 43 - 53%; 24-28 - 108 - 56%; and thereafter falls). Urban series usually produced similar reversals. Cf. fn. 11. 31. In PI-NDCs, when poorer people usually married later than others (pp. 24-6), the age of marriage would tend to push up the average age of poor HHS, going against the other effects mentioned here and leaving the upshot indeterminate. 32. "A high proportion of babies aged 0-4 (in relation to adult women and older children)" was found in the low-participation, but not high-participation, villages in a 93-village Indian sample (from the 1960s), on the basis of principal component analysis (Dasgupta, 1977, p. 157; my emphasis). 33. (a) The coefficient of (age2) was negative (-.0052) but insignifioant even at 10%. The MEP and income-per-head specifioations are here preferred to those "per consumer unit" - oalled "standard inoome" by Gaiha (p. 7) - beoause Gaiha follows a rather unusual set of CUs, which weight children aged 0-1 at zero. (b) Significance levels oorreoted from text (thanks to Gaiha, pers. comm.). 34. Excluding under-fives in urban and rural Maharashtra only. 35. Table 16 (p. 60) of Visaria, 1980, seems to be transcribed wrongly, since it is not consistent wth the text (p. 58), the graphs (p. 61), or the known facts, all three of which are oonsistent with one another. Therefore, this paper uses pp. 58 and 61 and ignores p. 60. 36. There are mild falls in urban Gujarat in 1972-3 (51% of males of all ages in the poorest MEP decile, 47% in the richest); rural Maharashtra (51% to 45%, but all the fall is in the top two deciles); peninsular Malaysia (53% to 51%); and Taiwan (52% to 49%). 37. Corresponding figures for urban GuJarat were muoh less unbalanoed (respectively 109, and 117: Visaria, 1977, p. 8A). This suggests that a national business and political centre, like Bombay, is especially liable to the "exoessw presence, and influence, of wealthy and powerful males. This was confirmed in 1972-3 in urban West Bengal (with its national "centre" in Caloutta), a small female surplus in the poorest homes, balance among the moderately poor, and a growing male surplus in the top two-thirds; and also in urban Delhi (Sarvekshana, Jan. 1979, pp. S360-7). Also, urban Nepal showed a similar pattern to Maharashtra (Visaria, 1980, p. 61). 115 38. In most LICs, rural populations - normally defined as those In plaoes with below 5,000 inhabitants - still form 75% or more of national totals. Therefore, while sex-selective or age-seleotive migration greatly alters the demographio structures of urban places, and of HBP groups within them, the migrants are a muoh smaller proportion of rural residents, and affeot rural struotures (total and in HEP sub-groups) far less. 39. Possibly the effect of female-headedness on relative HEP, even holding civil status oonstant, is higher in middle-income aountries, where poverty is muoh less extreme. In Belo Horizonte, Brazil, in 1972, 90.1% of 1908 sampled male HHs were married, as against 15.8% of 379 female HHs (158 one-person households were excluded); but the inoidence of "poverty" was much higher among women of each civil status. 26 of 43 widowed women (2 of 6 men), 75 of 158 separated (9 of 32), 27 of 60 married (481 of 1720), and 42 of 118 single (30 of 150) were "poor". But "poor" was defined as "monthly inoome-per-CU below 100 Cr." or $16 U.S. - far above the poverty-levels oonsidered in this paper. See Merrik and Schmink, 1982. 40. Less "inequitable"; but perhaps more "harmful", if poverty ooncentrates in periods when a larger proportion of the household is vulnerable, e.g. comprises under-fives. This is a good example of the risks of applying diminishing-marginal-utility arguments to shifting populationsl 41. By timing, because surveys are harder to conduct when villages are less accessible, i.e. usually in the wet season when food stores are running low and the poor are poorer; by location, because administrative convenienoe looates study areas near to towns and roads; by procedure, because "household" surveys may miss many of the poorest, who are transient migrants (mainly rural) or homeless (mainly urban). Many relevant issues are summarized in Chambers (1980). 42. Some construotions of the egalitarian case (e.g. Lipton, 1968, pp. 92-8) are based on assumptions (easily put into ordinalist language) of diminishing marginal utility of the total value of commodities, plus statistioal independence of initial real inoome from both entitlement to, and enjoyment from, extra real income. "Everybody should be famous for fifteen minutes" (A. Warhol). On the same assumptions, given "total poverty", should everyone be poor for ten days? The answer has to depend on whether it is "the same misery" and especially on when It arrives in the life-span: see fn. 40. 43. Some studies show, for example, income per person disaggregated in different and non-oomparable ways from consumption per person, e.g. by caste groups and occupation groups respectively. Others measure both Income and consumption per household, so that (because householda with high total inoome tend to be larger) we oannot be sure that if, say, a low income-per-household group has a high excess of consumption over income, it is also a "poor" group (low income per person) with, therefore, low oapacity to borrow. 116 44. At rural outlay levels of Rs.34 per person per month in 1972-3, an NSS "borderline" between groups of households, 80% of outlay spent on food barely sufficed to meet 80% of the average Indian household member's caloric needs. Not all persons in such households, however, were ultra-poor: 48% of them (spending on average only Rs.29 per month in 1971-2) were children, with below-average needs (V. Patel, 1973, Appendix, p. ix, Table 4), a larger proportion than in other households (or. Table 2); and prices were then somewhat lower in the survey year, 1970-1, than in 1972-3. 45. Because of the notorious problem of concealment of income, and because consumption fluctuates less than income and is more likely to be reported correctly in a brief survey, outlay indicators are preferable. In poor households, however, concealment is normally much less, and is partly offset by the tendency to adjust reports towards the local average. It is, incidentally, arguable that welfare costs of work - or at least caloric costs - should be deducted from income or outlay. These questions are much more discussed, yet (in their impact on measures and rankings of poverty) much less important, than the choice-of-denominator problem on which we concentrate here. 46. Indian NSS (1971-2) weights for estimating equivalent adult consumers are: age 0-1, 0.43; 1-3, 0.54; 4-6, 0.72; 7-9, 0.87; 10-12, 1.03 male (0.93 female); 13-15, 0.97 (0.80); 16-19, 1.02 (0.75); 20-39, 1.00 (0.71); 40-49, 0.95 (0.68); 50-59, 0.90 (0.64); 60-69, 0.80 (0.51); 70+, 0.70 (0.50). Taiwanese Bureau of Statistics (1974) weights: age 0-1, 0.3; 2-4, 0.4; 5-7, 0.5; 8-10, 0.7; 11-14, 0.8; 15-20, 0.9; 21+, men 1.0, women 0.9 (Visaria, 1980, p. 200). Standard Lusk CUs (Schofield, 1979, p. 14) are: 0-1, 0.3; 1-3, 0.4; 4-6, 0.5; 7-9, 0.7; 10-12, 0.8; 13-19, 1.0; 20., men 1.0, women 0.7. 47. Often, we know the distributional data for a population's MEP and outlay-per-CU only by group (e.g. ranked decile) means. In such cases, given (i)-(iii) here, it is even less safe to infer, from (e.g.) similar Gini coefficients by per-person and per-CU indicators, to similar proportions in poverty by the two indicators. 48. For example, households with several underemployed adults, but no children, would be misclassified upwards by MEP (compared with outlay per CU). Conversely, households in the lowest HEP deciles with high child/adult ratios - but with children older than the nutritionally vulnerable 0-5 age-group - could be misclassified downwards by MEP (compared with outlay per CU). 49. Hence large parts of extra food, given to children in nutrition programs, are offset by reductions in their food at home - i.e., materialize mainly as extra income for adults in the same household (Beaton and Ghasseimi, 1982). 50. Similar arguments apply to analyses of poverty and undernutrition. Probably, MEP sufficient, under stated assumptions, to buy 80% (not 100%) of the average food needs for persons (by age, sex and activity groups) beat indioates absence of extreme, health-threatening ultra-poverty - if the extent of such poverty is being used as a 117 criterion for allooating resources among areas or projeots (Lipton, 1983). Medical experts are, however, fully justified in objeoting to any such criterion if the aim is to diagnose need at individual or household level. 51. Project or area ohoices, of course, depend also on severity of poverty - not just on incidence. The Sen index can be shown, on highly plausible assumptions, to be the only acceptable index to measure "severity" and "incidence" together (Sen, 1981). However, its seoond derivatives are completely counter-intuitive. Indication of proportions of persons in "poverty" and "ultra-poverty", corresponding to caloric risk (Lipton, 1983), may be a useful alternative approaoh. 52. It is almost universally agreed to be true that, when a rich person gives food to an ultra-poor person, the poor person's welfare gain is somehow "more" than the rich person's welfare loss. Any theory claiming that "problems", of measurement or otherwise, mean that we "cannot" utter statements not scientifically refutable, and agreed by almost all to be true, is a bad theory. 53. "If comparisons of expenditure are to be used to measure ohild oosts, then the households being compared must be equally well-off ... [Does this mean] the parents', the children's, or some oomposite 'household welfare'? Only the first olearly makes sense, since only the parents are present both before and after the arrival of the ohild [so that] it is possible to ask questions about the extra costs needed to maintain their previous [level] of living": Deaton and Muellbauer, 1981, p.7. 54. This prooedure would take care of economies of scale in oonsumption at the same time as household-composition effects. 55. Sir Dennis Robertson suggested this term, to exclude changes in levels of well-being not related to eoonomic activity. 56. The work in LICs on ESCs addresses itself to different questions. For example, Iyengar, Jain and Srinivasan (1969) enquire whether, for speoific groups of oonsumer outlay (e.g. food), the sum of the elasticities of expenditure with respeot to total outlay and to family size is below unity. If so - if, say, a household that increases in size and in total outlay by 25 peroent (so that outlay-per-person is oonstant) raises food outlay by only 20 percent - economies of scale in food oonsumption are inferred by the authors. This prooedure, due to Houthakker and Crockett, implicitly assumes that ESCs would be zero if all outlays rose by 25 percent under these circumstanoes. Thus zero ESCs in our sense are assumed. 57. Most US ECSs in transport are due to the oar as a "family good", and do not apply to poor LIC families - the expeoted dietary energy cost of walking is twice as high for two as for one. Reaoh-down olothing to younger siblings is probably oomoner among the lowest-MEP 25% in LICs than in the USA - but, sinoe clothing forms so small a part of the food-dominated budgets of the Third World's poor, that soale-eoonomy oan "bite" on only a small share of their outlay. 118 58. Many surveys estimate household data by (say) monthly outlay-per-household quintiles; divide each quintile's monthly outlay by MHS in that quintile; and present the result as "MIP quintiles". Sinoe (pp. 58-9) household-outlay rankings typioally assign 2 in 3 households - and persons - to different quintiles of households from HEP rankings, this prooedure is quite unacceptable. 59. This example is chosen so that both regions have the same numbers of poor, but one a larger number of ultra-poor. The case for using an 80 percent rigure as an ultra-poverty out-off is explained in Lipton, 1983. 60. The scale of the problem appears from some of the positive relationships of MHS to landholding reported above (pp. 11-12). Floors and ceilings should take account of total income per person (or per CU), not land per household. Land reform is hard enough to push through, without the needless handioap of widespread demographic resentment. 61. The two crises are to some extent alternatives. If domestio food growers are paid attractive prices for their food, which is then sold oheap to consumers, the crisis is likely to be mostly fisoal (though farm inputs could swell the import bill). If the cost to the State of consumers' food subsidies is kept down by squeezing farm-gate prices of food, then domestic production is hit, food imports rise, and the orisis is likely to be mostly in the balance of payments. 119 BIBLIOORAPHY S. Abmad, Inoome Distribution. Institutional Struoture and Eoonomio DeveloPment: a Case Study of Pakistan, D. Phil. (unpub.), Sussex University, 1981. R. Andorka and T. Farago', 'Pre-industrial household structure in Hungary', in Wall et al. (eds.), 1983. Anon., Rampura (Mehaana Dist., GuJarat) (surveyed - hereafter s. - Aug.-Dec.1962; referenoe (orop) year - hereafter r. - 1961-2), Indian Village Studies (hereafter IVS) No. 15, Agro-economic Centre for Gujarat and Rajasthan at Sardar Patel University (hereafter AERC), Vallabh Vidynanagar, 1970. Anon., Umedgur (Jalore Dist., Rajasthan): a Mahayan Village (a. Sept. 1965; r.1964-5), IVS No. 17, ABRC Vallabh Vidynanagar, 1971. Anon., Jambua (Panachamals Dist.. GuJarat): Change in a Tribal Village, Resurveys of Indian Villages: No. 3, AERC Vallabh Vidyanagar, 1980. R. Anthony and B. Johnston, Field Study of Agricultural Change: Northern Katsina, Nigeria, Food Research Institute, Preliminary Report No. 6, Stanford, 1968. 0. Beaton and H. Ghasseimi, 'Preschool feeding programs', Amerioan Journal of Clinical Nutrition (hereafter AJCN), 35, 4, April 1982. M. Bhat and K. Pichholia, Falna (Pali Dist.. Rajasthan): a Village in an IADP District (a. Nov. 1962-Apr. 1963; r. 1961-2), IVS No. 9, AERC Vallabh Vidyanagar, 1967. N. Birdsall, Policy Relevance of Recent Social Researoh on Fertility, ICP Staff Paper, Smithsonian Institution, Washington, D.C., 1974. , 'Analytical approaches to the relationship of population growth and development', Population and Develo2aent Review (hereafter PDR), I, 1-2, 1977. , Siblings and Schooling in Urban Colombia, D.Phil. (unpub.), Yale University, May 1979. , Population and Poverty in the DeveloPina World, World Bank Staff Working Paper (hereafter VBSWP) No. 404, World Bank, Washington, D.C., 1980. W. Borrne and M. Cameron (eds.), Population Change: Asia and Oceania, Prooeedings of the 1967 Sydney Conference of the International Union for the Scientifio Study of Population, Sydney, 1969. A. Bose, 'Migration streams in India', in Borrie and Cameron (1969). Botawana, Government of (CSO), Rural Income Distribution Survey 1974-5, Gaberone, 1976. 120 D. Brahmbhatt, Samaldevi (Panchmahals Dist., Gujarat): a Survey in a Baokward District (a. Aug.-Oct., 1965; r. 1964-5), IVS No. 21, AERC Vallabh Vidynanagar, 1977. R. Bromley (ed.), The Urban Informal Sector: Critical Perspectives (World Development, 6, 9-10, 1978). T. Burch, 'Some demographic determinants of average household size: an analytic approach', in Laslett and Wall (eds.), 1972. S. Butler, The Way of All Flesh, 1903 (Penguin reissue, 1947). H. Buvini', M. Lyeette and W. McGreevey (eds.), Women's Issues in Third World Poverty, mimeo, World Bank, 1982 (Johns Hopkins, forthcoming). M. Cain, 'Risk and insurance: perspectives on fertility and agrarian change in India and Bangladesh', PDR, 7, 3, 1981. J. Caldwell, P. Reddy and P. Caldwell, 'The determinants of fertility decline in India', mimeo, paper to British Society for Population Studies (hereafter BSPS) Conference on Indian Population, Oxford, December 1982. A. Carloni, 'Sex disparities in the distribution of food within rural households', Food and Nutrition, I, 1, 1981. P. Carrasco, Land and Policy in Tibet, Seattle, 1959. R. Cassen, India: Population, Economy, Society, Macmillan, 1978. R. Chambers, Rural Poverty Unperceived, WBSWP No. 400, Washington, D.C., 1980. 0. ChatterJee and N. Bhattacharya, 'Between-states variation in consumer prices and per-oapita household consumption in rural India', in Srinivasan and Bardhan (eds.), 1974. L. Chen et al., 'Anthropometric assessment of protein-energy malnutrition and subsequent risk of mortality', AJCN, _3, 1980. , E. Haq, and S. d'Souza, 'Sex bias in the family allocation of food and health care in rural Bangladesh', PDR, 79 1, 1981. K.M. Choudhary, Aidan-ka-Was (Jaipur District, Rajasthan): a Gramdan Village (a. Oot.-Nov. 1962; r. 1961-2), IVS No. 5, AERC Vallabh Vidyanagar, 1964. P. Collier and D. Lal, 'Poverty and Growth in Kenya' (revised), mimeo, Studies in Employment and Rural Development no. 55, DE/ERD (DPS), World Bank, Washington, D.C., July 1980. P. Czap, Jr., '"A large family: the peasant's greatest wealth": serf households in Hishino, Russia, 1814-1858', in Wall et al. (eds.), 1983. L. Danhieux, 'The evolving household: the oase of Lampernisse, West Flanders', 121 in Wall et al. (eds.), 1983. B. Dasgupta, Village Society and Labour Use, Oxford U.P. (Delhi), 1977. G. Datta and J. Meerman, Household inoome or household inoome ver oapita in welfare oomparisons, WVBSP No. 378, Washington, D.C., 1980. K. Davis and J. Blake, 'Sooial structure and fertility', Eoonomio Development and Cultural Change, 4, 3, 1956. A. Deaton and J. Muellbauer, 'On measuring child oosts in poor countries', mimeo, Living Standards Measurement Study, World Bank, Jan. 1983. V. Deijomeah and B. Anusionwu, Education and Inoome Distribution in Nigeria, mimeo, Income Distribution in Nigeria Project, World Bank, 1979. B. Desai, Moti Khadol (Kaira Dist., Gujarat): a Charotar Village (a. Nov. 1963-Feb. 1964; r. 1962-3), IVS No. 10, AERC Vallabh Vidyanagar, 1967. M. Desai, Studies in Rural Problems: 5 (Rajasthan): Hasteda (a. Oct. 1961-Mar. 1962; r. 1960-1), AERC Vallabh Vidyanagar, 1966. J. Dupaquier and L. Jadin, 'Structure of household and family in Corsica, 1769-71', in Laslett and Wall (eds.), 1972. A. Fauve-Chamoux, 'The importance of women in an urban environment', in Wall et al. (eds.), 1983. R. Frisch, 'Population, food intake and fertility', Science, 199, January 6, 1978. , 'Pubertal adipose tissue: is it necessary for normal maturation?' Federation Prooeedings, 39 7, May 15, 1980. 'Malnutrition and fertility', Science, 215, March 5, 1982. R. Gaiha, 'On measuring the risk of poverty in rural India', mimeo, Faculty of Management Studies, Delhi, 1982. , 'Poverty entitlements and development in rural India', mimeo, Faculty of Management Studies, University of Delhi, June 1983. and N. Kazmi, 'Aspects of poverty in rural India', mimeo, Faculty of Management Studies, Delhi, 1982. J. Goody, 'The evolution of the family', in Laslett and Wall (eds.), 1972. J. HaJnal, 'Two kinds of pre-industrial household formation system', PDR, t, 3, September 1982. S. Harbison, 'Family structure and family strategy in household deoision-making', in de Jong and Gardner (eds.), 1982. J. Harrington, 'Nutritional stress and economic responsibility: a study of Nigerian women', in Buvini6 et al. (eds.), 1982. A. Hayami and N. Uohida, 'Size of household in a Japanese oounty throughout 122 the Tokugawa era', in Laslett and Wall (eds.), 1972. A. Hill, The Demographic Situation in Sub-Saharan Afrioa, Discussion Paper no. 81-22, Population and Human Resource Division, World Bank, Washington, D.C., May, 1981. P. Hill, Rural Hausa: a Village and a Setting, Cambridge, 1972. -, Dry Grain Farming Families, Cambridge, 1982. B. Hoselitz, 'Economic Growth: Non-economic Factors', in Encyclopedia of the Social Sciences, MacMillan, 4, 1968. T. and V. Hull, 'Relation of economic class and fertility', mimeo, Population Institute, Gadjah Mada University, Yogyakarta, 1976. N. Iyengar, L. Jain and T. Srinivasan, 'Economies of scale in household oonsumption: a case study', Econometric Annual, XV, 4, 1969. a. de Jong and R. Gardner (eds.), Migration Decision Making, Pergamon, 1982. H. Joshi, 'Prospects and case for employment of women in Indian cities', Economic and Political Weekly (hereafter EFW), Annual Number, August 1976. A. Khan, R. Islam and M. Huq, Employment, Income and the Mobilization of Local Resources: a Study of Two Bangaldesh Villages, ILO(ARTEP), Bangkok, 1981. A. Kielmann et al., 'The Narangwal experiment in interactions of nutrition and infections: II. Morbidity and mortality effects', Indian Journal of Medical Research, 68 (Suppl.), December 1978. C. Klapisch, 'Household and family in Tuscany in 1427', in Laslett and Wall (eds.), 1972. J. Kochanowiaz, 'The peasant family as an economic unit in the Polish feudal economy of the eighteenth century', in Wall et al. (eds.), 1983. P. Kolenda, 'Region, caste and family structure: a comparative study of the Indian "Joint" family', in Singer and Cohn (eds.), 1968. S. Kossoudji and E. Mueller, 'Economic and demographic status of female-headed households in Botswana', mimeo, Population Studies Centre, Michigan University, Ann Arbor, 1980. S. Kumar, 'Role of the household economy in determining child nutrition at low income levels: a case study of Kerala', Occasional Paper no. 95, Division of Nutritional Sciences, Cornell, November 1977. K. Kurihara, Post-Keynesian Economics, Rutgers, 1954. 3. le Roy Ladurie, Carnival in Romans (tr. M. Feeney), Braziller, New York, 1979. 123 R. Lambert, Workers. Faotories and Sooial Change in India, Princeton, 1963. P. Laslett, The World we have Lost (2nd ed.), Scribners, 1971. , 'Mean household size in England since the sixteenth century', in Laslett and Wall, 1972. , 'The stem-ramily hypothesis and its privileged position', in Waohter et al., 1978. and R. Wall (eds.), Household and Family in Past Time, Cambridge, 1972. and 1. Clarke, 'Houseful and household in an eighteenth-century Balkan city', in Laslett and Wall (eds.), 1972. E. Lazear and R. Michael, 'Family size and the distribution of per capita income', American Economic Review, 70, 1, March 1980. 0. Lewis, Village Life in Northern India, Vintage Books (New York), 1958. M. Lipton, Assessing Economic Performance, Staples, 1968. , Poverty, Undernutrition and Hunger, WBSWP no. 597, Washington, D.C., 1983. , Labor and Poverty, WBSWP (forthcoming), World Bank, Washington, D.C., 1983a. , 'Rural development and the retention of the rural population in the countryside of developing countries', Canadian Journal of Development Studies, III, 1, 1982. P. Lloyd, Slums of Hope?, Penguin, 1979. R. Lucas, 'Migration amongst the Batswana', mimeo, Boston University, March 1982. S. Habmud and J. McIntosh, 'Returns to scale from family size - who gains from high fertility?', Population Studies, 1980. N. Hathew and W. Scott, A Develovment Monitoring Service at the Local Level. vol. 1: Socio-economic Observation Areas in Kerala, UNRISD, Geneva, 1980. P. Matlon, Income Distribution among Farmers in Northern Nigeria, Michigan State University, 1979. D. Mazuadar, Paradiws in the Study of Urban Labor Harkets in LDCs, WBSWP No. 336, Washington, D.C., 1979. , The Urban Labor Harket and Income Distribution: a Study of Malaysia, World Bank, 1981. T. MHKeown, The Role of Medicine, Blackwell, 1979. 0. Meesook, Income. Consumption and Poverty in Thailand, 1962/63 to 1975/76, WBSWP No. 364, World Bank, Washington, D.C., November 1979. T. Merrick and M. Schmink, 'Households headed by females and urban poverty in 124 Brazil', in Buvinic et al. (eds.), 1982. B. Miller, The Endangered Sex: Neglect of Female Children in Rural North India, Cornell, 1981. A. Mitra, India's PoDulation: Aspects of Quality and Control, vol. 1, Family Planning Association/ICSSR, Abhinav, New Delhi, 1978. F. Modigliani and R. Brumberg, 'Utility analysis and the consumption function: an interpretation of cross-section data', in Kurihara (1954). R. Mohan and R. Hartline, 'The poor of Bogota: who they are, what they do, where they live', mimeo, World Bank, 1980. C. Nakane, 'An interpretation of the size and structure of the household in Japan over three centuries', in Laslett and Wall (eds.), 1972. K. Natrajan, 'Child mortality: analysis of 1972 fertility survey', Indian Institute of Technology, Bombay (for Registrar-General's Office, Delhi), mimeo, n.d. (1980?). D. Norman et al., A socio-economic Survey of Three Villages in the Sokoto Close-settled Zone, Samaru Miscellaneous Paper no. 65, Zaria, 1976. , M. Newman and I. Oudraogo, 'Farm and Village Production Systems in the Semi-arid Tropics of West Africa', Research Bulletin, 4, 1, October 1981, ICRISAT, Patancheru. J. Nugent and R. Walther, 'Old-age security, household structure, marriage and fertility', mimeo, Department of Economics, University of Southern California, 1981. and F. Tarawneh, 'The anatomy of changes in income distribution and poverty among Mexico's economicallyactive population, 1950-70', mimeo, Department of Economics, University of Southern California, 1982. 0. Papanek, 'The poor of Jakarta', Economic DeveloPment and Cultural Change, 24, 1, 1975. T. Parsons and E. Shils, Towards a General Theory of Action, Harvard U.P., 1951. 0. Parthasarthy and G. Rama Rao, 'Employment and unemployment among rural labor households', EPW, VIII, 52, December 28, 1973. R. Patel, Afawa, (Surat District. GuJarat): an Advanced Village in a Traditional Setting (s. Sept. 1962-Jan. 1963; r. 1961-2), IVS no. 7, AERC Vallabh Vidyanagar, 1964. , Ankodia, Baroda District (s. Dec. 1961-Apr. 1962; r. 1961-2), IVS no. 3 (Gujarat), AERC Vallabh Vidyanagar, 1964a. V. Patel, ConsumDtion Patterns in Rural Gujarat: a Study of Four Villages in 125 Anand Taluka (1970-1), Research Study No. 31, AERC Vallabh Vidyanagar, 1973. J. Peach and T. Januazi, Bangladesh: a Profile of the Countryside, USAID, 1979. S. Purohit, Zawar (Udaipur Dist., Rajasthan): a Tribal Village near Zawar Mines (a. Jul.-Oct. 1964; r. 1963-4), IVS no. 19, AERC Vallabh Vidynanagar, 1974. E. Quizon and B. Evenson, 'Time allocation and home production in Philippine rural households', in Buvinic et al., 1982. M. Rao, 'Nutrition situation in India during 1971-2', Sarvekshana, II, 3, Jan. 1979. R. Repetto and A. Deolalikar, Dynamics of Demographio and Economic Change in Rural India: Matar Taluka. Gujarat State, 1965-75, mimeo, Harvard, 1980. S. Reutlinger and M. Selowsky, Malnutrition and Poverty: Magnitude and Policy Options, World Bank Occasional Paper No. 23, Johns Hopkins, 1976. H. Rouis, Nigeria: Income Distribution and Poverty Profiles, mimeo, World Bank, June 5, 1980. M. Rosenzweig and K. Wolpin, 'An economic analysis of the extended family in a developing country', Economic Growth Center Discussion Paper No. 317, Yale, 1979. and P. Schultz, 'Market opportunity, genetic endowments and the intrafamily distribution of resources: child survival in rural India', Economic Growth Center, Discussion Paper No. 347, Yale University, 1980. L. and S. Rudolph, The Modernity of Tradition, Chicago University Press, 1967. L. Ruzioka, 'Mortality in India: past trends and future prospects', mimeo, paper to BSPS Conference on Indian Population, Oxford, December 1982. Sarvekshana, National Sample Survey, Government of India, several issues. S. Saxena, Sahnusar (Sikar Dist.. Rajasthan) (s. Sept.-Oot. 1963; r. 1962-3), IVS no. 14, AERC Vallabh Vidynanagar, 1968. and A. Charan, Dingri (Udaipur Dist.. Rajasthan): a Village Near a New Rail Link (s. Jul.-Oct. 1964; r. 1963-4), IVS no. 18, AERC Vallabh Vidynanagar, 1973. S. Sohofield, Development and the Problems of Village Nutrition, Croom Helm, 1979. P. Schmidtbauer, 'The changing household: Austrian household struoture from the 17th to the early 20th century', in Wall et al. (eds.), 1983. 126 P. Schultz, Economios of Population, Addison Wesley, 1981. , 'Age of individuals and family composition as factors explaining the distribution of personal income', Beonomic Growth Center Discussion Paper no. 383, mimeo, Yale University, 1981a. A. Sen, Poverty and Famines, Oxford, 1981. R. Sieder and H. Mitterauer, 'The reconstruction of the family life course', in Wall et al. (eds.), 1983. A. Shah, 'Changes in the Indian family', EPW, 1, 1-2, Annual Number, Jan. 1968. , Household Dimensions of the Family in India, University of California, Berkeley, 1973. C. Shah, 'Food preferences and nutrition', Presidential Address, Indian Society of Agric. Economics, Bangalore, 1979. N. Shah, Oon (Surat Dist.. Gujarat): a Village on the Periphery of an Industrial Estate (s. Jul.-Sept. 1964; r. 1963-4), IVS no. 13, AERC Vallabh Vidynanagar, 1968. H. Shetty, Patadia (Sabarkantha Dist.. Gujarat): a Tribal Village (a. Oct. 1961-Jan. 1962; r. 1960-1), IVS no. 1, ABRC Vallabh Vidynanagar, 1963. E. Simmons, Rural Household Expenditures in Three Villages of Zaria Province, Samaru Miscellaneous Paper No. 56, Ahmadu Bello University, 1976. M. Singer and B. Cohn (eds.) Structure and Change in Indian Society, Wenner-Oren Foundation, New York, 1968. B. Singh, Agrarian Structure. Economic Change and Poverty: the experience of Central Gujarat, D. Phil. (unpub.), University of Sussex, August 1981. J. Smith (ed.), Female Labor Supply, Prineeton, 1980. M. Srinivas, Caste in Modern India, Asia, 1962. T. Srinivasan, Malnutrition: some Measurement and PolicY Issues, WBSWP No. 373, World Bank, 1980. and P. Bardhan (eds.), Poverty and Income Distribution in India, Statistical and Publishing Society, Calcutta, 1974. 0. Stark, Utility. Technological Change. Surplus and Risk: the Micro-econamics of Rural-to-Urban Migration of Labor in Less Developed Economies, D. Phil. (unpub.), University of Sussex, April 1975. J. Stycos, O. Seeley and R. Avery, Patterns of Fertility Decline in Costa Rica, Cornell, 1978, reported in People, 5, 3, July 1978, pp. 36-7. S. Vatuk, 'Cultural perspectives on social services for the aged in India', cited in Nugent and Walther, 1981. 127 J. daVanzo, 'Microeoonomio approaches to studying migration deoisions', in de Jong and Gardner (eds.), 1982. P. Visaria, Living Standards. Employment and Education in Western India, Working Paper No. 1, Joint ESCAP/IBRD Project on the Evaluation of Asian Inoome Distribution Data, World Bank, Washington, D.C., 1977. Size of Landholding. Living Standards and Employment in Rural Western India, 1972-3, Working Paper No. 3, joint ESCAP-IBED Project on the Evaluation of Asian Inoome Diatribution Data, World Bank, Washington, D.C., October 1978. , Poverty and Living Standards in Asia, Living Standards Measurement Survey, Working Paper No. 2, World Bank, Washington, D.C., 1980. , 'Demographio faotors in the distribution of income', mimeo, Development Researoh Center, World Bank, Washington, D.C., 1980a. K. Waohter and R. Laslett, 'The English evidence on household structure compared with the outoomes of micro-simulation', in Wachter et al., 1978. K. Wachter and E. Haamnell, 'Experimental results', in Waohter et al., 1978. K. Wachter, E. Hammell and P. Laslett, Statistical Studies of Historioal Sooial Struoture, Academio Press (New York), 1978. R. Wall, 'Mean household size in England from printed souroes', in Laslett and Wall (eds.), 1972. , 'The oomposition of households in a population of 6 men to 10 women: south-east Bruges in 1814' in Wall et al. (eds.), 1983. , 'Does owning real property influence the form of the household? An example from West Flanders' (Wall, 1983a), in Wall et al. (eds.), 1983. , J. Robin and P. Laslett (eds.), Family Forms in Historic EuroDe, Cambridge, 1983. World Bank, Focus on Poverty, revised edition, February 1983. A. van der Woude, 'Variations in the size and structure of the household in the United Provinces of the Netherlands in the 17th and 18th centuries', in Laslett and Wall (eds.), 1972. J. Wyon and J. Gordon, The Khanna Study: PoDulation Problems in the Rural PunJab, Harvard, 1971. K. Zaohariah and R. Kurup, 'Determinants of fertility deoline in Kerala', mimeo, paper to BSPS conferenoe on Indian population, Oxford, Deoember 1982. Wodld Bank The Deign of Development Economk Development Pro. PUblicationis Jan Tlnbergen Jects and Their Appraisal: rormulaktes a coherent govemment C 9Se and Prindples from of Reated polcy to further development objec- the Experience of the odves and ouUines m thods to Wori 8ak Interest sUmulate privat Investments. John A. King The Johns Hopkins Uniuersity Prss. The English-language edition is 1958; 6th printing. 1966.108 pages out of print. (including 4 annexes, Index). Fench: ProJets de dveloppement LC 58-9458. 15B1 0-8018-0633X. cconomique et leur evaluation. Dunod Accelerated Development .00 .00) paperback. Editeur. 24-26. bouleuard de rtHopital. In Sub-Saharan Africa: 7500 ParLs, Flrance. 1969. An Agenda for Action Development Stategies In 99 fancs. In the fall of 1979, the Afrfcan Gover- Semi-Industral Economies Spanish: La evaluaclon de prayectors de nors of the World Bank addressed a Bela Balassa desarrollo eco6iomico. Editorial Tecnos, memorndum to the Bank's president Provides an analysis of development 19 70. 545 pages (Including Indexes). expressing their alarm at the dim staeisI eiiiutil800 pesetas. econmlcpropecs fo th naion of econiomies that have established an sub-Saharan Africa and asking that Industrial base. Endeavors to quantify Ecnmc rotad the Bank prepare a special paper on the systems of Incentives that are Ecoomic Growth and the economk development problems applied In six semi-Industrial develop. l|uman Resources aof a hesecounties"and thm approp-i Ing economies-Argentina, Norman Hicks, assisted by -ate program for helping them. This Colombia. Israel, K~orea Singapome Jahangir Boroumand report, building on the Lagos Plan of and Tiiwan-and to indicate the Action, Is the response to that effects of these systems on the alloca- World Bank Staff Working Paper Noa. The report discusses the factors tion of resoues. International trae. 408. July 1980. Iu + 36 pages (Includ- that explain slow economic growth In and economic rowth. Ing 3 appendixes, bibliography, and Africa In the recent past, analyzes The Johns Hopkins Uniuersity Press. references). policy changes and program orlenta- 1982. 416 pages (including appendixes. Stock N1o. WP-0408. $3.00. t!ons needed to promote faster Index). growth, and concludes with a set of LC 81-15558. ISB1 0-8018-2569-5, recommendations to donors, Includ- $39.95 hardcouer. MEw Ing the recommendaton that aid to Africa should double In real terms to bring about renewed African develop- Etem and Southern The Extent of Poverty In ment and growth In the 1980s. The Latin America report's agenda for action Is general Africa: Past Trends and Oscar Altimir It Indicates broad policy and program Future P spects This work originated In a resah directions. overall priorities for Ravi Guihati rjc o h esrmn n acton, and key areas for donor atten- prJc for the me me and tion. Like the Lagos Rlan. the report World Bank Staff Working Paper (10. antalysIs of incomie distribution In the recognizes that Africa has enormous 413. August 1980. 24 pages. Latin American countries, undertaken economic potential. which awaits Stock (1o. WP-0413. $3.00. Jorntly by the raconomic Commrldon fullea development. Presents estimates of the extent of 1981; 2nd printing 1982. 198 pages absolute poverty for ten Latn (including statistical annex, biblogra- American countries and for the region phy). as a whole In the 19709. French: Ie developpement acceleri en World Bank Staff Working Paper Nlo. afrique au sud du Sahara: programme 522. 1982. 117 pages. Indicatif d'action. IS0tl 0-8213-0012-1. $5.00. Stock Nlos. SA-1981-E, SA-1981-F. free of charge. PFst Things PFrst: Meeting The Hungarian Economic Leveb ofo rbverty: Fbilcy Baic Human Needs in the Refonr, 1968-81 and Change Developing Countbes Bela Balassa Amartya Sen Paul Streeten, with Reviews the Hungarian experience World Bank Staff Working Paper 10. Shahid Javed BurkL with the economk reform inroduced 401. July 1980. 91 pages (including Mahbub ul Haq% In 1968 and provides a short descrip. ferencesJ Nornan Hicka, tion of the antecedents of the reform. and Frances Stewart ArAnayzes specfic mfo nmasurs Stock Nlo. WP-0401. $3.00. and Frances Stewart concerming agriculture. decislonmak- The basic needs approach to Ing by Industrial flms, price economic development is one way of determination, the exchange rate, Models of Growth and helping the poor emerge from their export subsidies, Import protection, Distribution for Brazil poverty. It enables them to eam or and Investment decisions and Indi- Lance Taylor, Edmar L. Bacha, obtain the necessities for life-nubi- cates their effects on the economy. laa Card don, housing, water and sanitation. Also examines the economic effects na oso, education, and health-and thus to of tendendes toward recentralization and Frank J. Lysy Increase their productivity. In the 1970s, as well as recent policy Expiores the Brazilian experience This book answers the critics of the measures aimed at reversing these from the point of view of political basic needs approach. views this tendencies. economy and computable general approach as a logical step in the World Bank 5141 Wki a 11 equilibrium Income distribution evolution of economic analysis and rl akS affWrkn Pper No. models. development pollcy, and presents a 506. February 1982. 31 pages (includ- clearsighted Interpretation of the Ing references). Oxford University Pres. 1980. issues. Based on the actual Stock lo. WP-0506. $3.00. 368 pages (including references, experience of various countries- appendixes, Index). their successes and failures-the LC 80-13786. ISBII 0-19-520206-6. book is a disillation of World Bank ImplemenDtng Programs of $27.50 hardcouer- 15B1N 0-19-520207-4. studies of the operational implica- HuaDe lomn$1.5perck tdons of meeting basic needs. It also Human Deveopment S14.95 paperback. discusses the presumed conflict be- Edited by Peter T. Knight- tween economic growth and basic prepared by Nat J. Colletta, needs, the reation between the Mew Jacob Meerman, and others. International Economic Order and basic needs, and the relation be- World Bank Staff Working Paper N1o. tween human rlghts and basic needs. 403. July 1980. Iu + 372 pages Oxford Uniuersity Pess. 1981; 2nd (including references). paperback printing, 1982. 224 pages Stock lo. WP-0403. $15.00. (including appendix, bibliography. Index). ~~~Intemnational Technology LC 81-16836, 15B11 0-19-520-368-2, Tmnsfeen Issues and $18.95 hardcouer ISBt1 0-19-520-369-0, PoHicy Options $7.95 paperback. Frances Stewart World Bank Staff Working Paper No. 344. July 1979. xii + 166 pages (including references). Stock N1o. WP-0344. $S.00. rour chapters provide an overview of Sructl Change and rCW alternative strategies: a detailed look D e CDa t POlicy at health. educatior. nutrition artd DauVs Chenet Piba tols n fertility lessons from existing pro. His Chenery Triba ftople and grams; and an examinaton of A retspective look at Chenerys Economi Deveopment broader issues In planning. thought and writing over the past two IIuma Ecologi Oxford UnJversity Pr. 19U2. 96 page ndecades and an extenson of his work ConsIdrtC ons Oxforudin sntaesticay appendix9).9 pgs In RedL*Lbution uith Qrowth and Robert Goodland (Including statistical appendLxP. FaUcm ofD n cevelops a LC 82-2153. ISBN 0-19-520389-5. $7,95 set of techniques for analyzing struc. At the curent tme. approxnteo paperback. tural changes and appiles them to 200 million tribal people [inv In all some mqjor problens of developing mions of the world and number countries today among thc poorst of the poor. This NEW Ox--ld UnluerSitij Ph= 19 79, 2 paper descrlbes the problems assoc - twIYEW OxJord Unluceslty PaS~ 1979: 2nd ated with the development process as paperback printing, 1982. 544 pages It affects tribal peoples It ouilnes the Reforming the 1lew (Including references, Index). requisites for meeting the human Economic Mechanim LC 79-18026. ISBN 0-19-520094-2, MOW* needs of triboi peoples; and In Hungary $34.50 hardcouer; IS( 0-l9-520095-0 pset general d r pe s toat and Bela Bala $12M paperback. ~~~~~~~~deflgned to assis the Bank staff and Bela Balassa $12.95 paperback. ~ prqect di0nes in Imror adng evaluates the reform measures taken rench: Changement des structures et apprrate procedures to ensure the in 1980 and 1981 (price setting, the politique de developpemenL survival of tribal peoples and to assist exchange rate and protection, wage Ecnomica. 1981. with their devdopmenL determination and personal Incomes, ISBN 2- 7178-0404-8 80 francs. May 1982. oil + 111 pages (Including 7 Investment decisions, and the organizational structure) that aim at Spanish: Camblo estructural y poiitica annees, bibilography). the further development of the de desarrollo. Editorial tecnos, 1980. ISBN 0-821J-0010-5. $5.00. Hungarian Mew economic Mecha. ISBNt 84-309-0845-5, 1000 pesetas. nism, Introduced on January 1 Me6.' World Bank Staff Working Paper No. Th_e Tropics And Economi World82 5pae.TourIsm-Passport to Deelopments A Provocative ISB4.1982.3600 pe Development? lNrspectves Iquliy into the Poverty ISBt 0-8213-0048-2. $3.00. on the socl and Cultu of Rations Effects of Tourism in Andrew M. Kamarck izw Developing Countries examines major characteristics of the Emanuel de Kadt editor tropical climates that are significant Socid Inmastiructure and The first serious effort at dealing wAth to economic development. Services In Z _mbabwe the effects of tourism development In The Johns Hopkins Untversity Press, Rashid Faruqee a broad senms concentrating on 1976:2nd prInting, 1979.128 pages T mjority oesmet of social and cultural questions. (including maps. bibliography. Index). Zimbabwe. coming to power after a A Joint World Bank-Unesco study. LC 76-17242. ISBN 0-8018-1891-5. long struggle for independene. has Ofbird UnIuersity Press. 1979. $11.00 (17.75) hardcouer, announced Its strong commitment to 378 pages (including maps, index). ISBN 0-8018-1903-2. $5.00 (f.0) social services to benefit the vast LC 79-1816. ISB1 0.19-520149-3. paperback. majority of the popultadon. This paper $245 ha,cow; ISBN 0-19-5201S0-7 nch: Les troplque tt le dheiopp- looks at Issues related to education. 09 00 mect tropmques un legdesans health housing and other tmportant paperback. ment etonomique: un rgard sans sectors and revtews specific plans and french: Le tourisme-passport pour le complaisance sur la pauvrete des resource requirenents to help deiloppemeft regards sur les effets nations. Economica, 1978. Improve the standard of lMving of the socIoculturels du tourisme dans les pays ISBN 2-7178-oI0.13. 25 rancs. population. en vole de dEwloppement economica, Spanish: Los tr6picos y desarrollo World Bank Staff Working Paper No. L980. econ6mico: reiexiones sobre la pobreza 493. October 1981. 11 pages (including 49 fiancs. de las naciones. Editorial Tecnox 1978. bibliography. map). ISBN 84-09-0740-8.350 peset. Stock No. WP-0495. $5.00. Twenty-fie Ym Of Etconomic Development, 1950 to 1975 David Monrwetz A broad assessment of development efort shows that although the deveopng countries hae been ratterns of Dlopment, Manow ed op Shelter 1950-1970 A Fbicy Vie. Anthony A. Churchill Hlollis Chenery Alan Berg DefiRnes the elements that constitute and Moises Syrquln DLscses the Importance of ade- shelter discusses the difficulties A comprehensive Interpretation of the quate nutrition as an objective as encountered In developing shelter structural changes that accompany well as a means of economk develop- programs for the poor, estniates or- the growth of developing countries, ment. OuUlines the many facets of Om ders of magnitude of shelter needs using cross-section and time-seles nutuition problem and shows how for the next twenty years, and pro- analysis to study the stablIty of efforts to Improve nutrition can help poses a strategy for meeting those observed pattems and the nature alleviate much of the human and needs. of time trends. economic waste In the developinrg September 1980. 39 pages. EngiLsh, Oxford University hess, 1975; 3rd orld. frnch, and Spanish. paperback prindng, 1980. 250 pages June 198. 108 pages (Including Stock M1os. BN-8002-E. 811-8002-f, (Including technical appendix, statisti- 6 appendLres. notes). EngiLsh. F17nch B-8002-S. $3.00 paperback. cal appendix, bibliography, Index). and Spanish (forthcoming). LC 74-29172. ISBN 0-19-920075-0, Stock Mos. B1-8104-E B1-8104-f, Watr Spply and $19.95 hardcouer; IsBn 0-19-920076-9, B(4104-S. SS.00. Wast Disposal $8.95 paperback. Wc of p l Spanish: ia estructura del crecimiento MctiUng Basic eds: meetins bass neizedosf ate problem of ec6nomico: un anallsis para el perlodo An OmeC and waste dLsposal and Its signiul- 1950-1970. Editorial Teconos, 1978. Mahbub ul Haq and cance to development In the context ISBN184-309-0741-6, 615 pesetas. Shahid Javed Burki of the Interrational Drinking Water Supply and Sanitation Decade. - Presents a summary of the main fiRnd- xamines the Bank's past role In Poverty and Basic Rees lngs of studies undertaken In the Inproving water supply and waste Poiver World Bank as part of a program for disposal facilities In developing Seies reducing absolute poverty and meet- countries and draws conclusions A series of booklets prpared by the Ing basic needs. for the future. staff of thc World Bank on the suen ect September 1980. 28 pages (including September 1980. 46 pages. EngiLsh. general studies that explore the con- 2 annexes). engish, french, Spanish, french, Spanish. and Arabic. cept of basic needs, country case Japanese. and Arabic. Stock Nlos. BN-8003-E, B1-8003-f, studies, and sectoral studies. SlDck tlos. B1N-8001-E, 81-8001-f, B1-8003-S, BN-8003-A. 8114001-S, B-80014, B11-8001-A. $3.00 paperback. Brazil $3.00 paperback. Peter T. Knight and Poverty and the Develop- Ricardo J. Moran ment of Human Resources: An edited andupdated edition of the Regional Perspective more detalled publication., Brzi Willem Bussink, David Davies, Human Resources Special Report (see Roger Grawe, Basil ICavalsky, description under Country StudIsanGu Pefrm n lIstlng, and uy i' Pfeffermann December 1981. 98 pages (including World Bank Staff Working Paper (o. statistical appendix, map). EngiLsh. 406. July 1980. 111 + 197 pages Stock (10. 8(1-8103. $5.00. (including 7 tables, 2 appendixes, S8ferences. footnotes). Stock (10. WP-0406. $5.00. MEW woverty and Human Development Paul isenman and others Since economic growth alone has not reduced absolute poverty, It has been necessary to consider other strategies. The strategy examined In this study - humnan development - epitomizes the Idea that poor people should be helped to help themselves. remarkably successful In achieving World Devlopment Report 1982 l m growth. the distribution of its (See Publications of Particular Interest beneflits among and within countries for description and sales Information.J) fack ieed The Case of Sd Laa has been less satisfactory. World Development Report 1981 P Isermn The Johns Hopkins University Pess. (Dlscusses adjustment-global and NbpH|eank Wat S p m 8r 197 1977; 3rd printing. 1981. 136 pages national-to promote sustainable gi9ni: 2n37-S (Including statistical appendizx gnrvth In the changing uorld Stock fv RP0197 c'charge. references). economy.) LC 77-17243. 15B1N 0-8018-2134-7, World Development Report 1980 naim 3oc Devooome. $16.50 (f8.00) hardcover; (Discusses adjustment and growth In for aneUp 15B1 0-8018-2092-8. $7.95 (f3.75) the 1980s and poverty and human Petr T. Knight paperback. development.) Wod Bank Reprint 51r1: humber 203. paperback. developmenL) ~~~~~~~~~~~~ftdie fiom World DewelopnienL moi. 9. no. french: Vingt-cinq annees de develop- World Development Report 1979 U/12 (1981).1063-82. pement economique: 1950 3 1975. (Discusses deuelopment prospects and Sbck ho. Rff420. ne al charge. Economica. 1978. International policy Lssues, structural ISB1 2-7178-0038-7, 26 francs. change, and country deuelopment In u_s Aa" d Spanish: Veinticinco aios de desarollo xperienc and Issucs.) iopint4, Ceme Kea l _ econ6mico: 1950 a 1975. Editorial World Development Report 1978 WOd Bank Rprint Serl5: Mumber 208 Tecnos, 1978. (Disusses the deuelopment experience, Neprint.om indigenow Anthmpolog in Mon- 1950-75, deuelopment priorities In the Wstemn Countries. edld by Mussein tahim 1SBIY 84-309-079240, 350 peset. middle-income deueloping countries. (Durham. Morth Carolina: Carolina Academic and prospects for alleviating pouerty.) Plre 1982).121-37. World Development Report Slck No. RP-O20 Ne of charye A large-format series of annual Latim AmerI and tbe Carbbeans studies of about 200 pages, the Ecomemk frtroince and PoUcles World Deuelopment Report, since Its Ouy l Pfeffenmann Inception, has been what The Cuad- Wridd Bank Reprint Serks. ttumnbe 228. Ian has called 'a most remarkable Rkpdnted fiom The Southetrn PAew of publication. It Is the nearest thing to Managemt and Economnics w. 2. no. I (Winter having an annual report on the 19821.129-72. present state of the planet and the Stock Mo. RP-0228. ee of charge. people who live on It. each Lssue brings not only an overview of the Pndemnizatlo and Devekoment state of development but also a tbedal of Traditkoaa Gras. Roots detailed analysis of such topics as ftaat O laautos structural change, the varying Mkhad M. Cermca experiences of low- and middle- World Bank Reprint Scrn e Humber 215. Income countries, the relation of *eprikned fem Dirctionst d Change: Modermiza- poverty and human resource ton Theoly. Rearch. and Realities.-Boulder. development, global and national Colorado: Westulew nws (1981): chapter 5. adJustment. and agriculture and food Sbck Mo. RP-0215. lee of charye. stability. Each contains a statistcal annex. World Development Indica- tors, that provides profiles of more than 120 countries In twenty-five multpage tables. The data cover such subjects as demography, Industry, trade, energy, Rnance. and develop- ment assistance and such measures of social conditions as education health. and nutriton. Confronting Utrba these studies Is pmsnted In The Design of MPaotion in West Af.ic fuUon PN Demogra Janes E. Austin World 8ank Staff Working Paper No. Describe a framc for 415. September 1960. ul + 365 pages s a carrying out urban (including statistcal annees. nutrition programs that examines bibliography). several key consieratons tn nutri- Stock lo. WP-0415. $15S.0. don educaon, on-site feeding, take- home feeding, nutrient-dense foods rton shops, food coupons fortifca- Economic Mothation ver s dor. dIrect nutrient dosagc, and food CitY Ugbts: Tesing pircesilng and didributlon. f about nte The Johns Hopkins Unluersitbj Pess. Ch t M ati in 1980. 136 pages. Thailad LC 79-3705. ISBif 0-8018-2261-0. Fred Amnold and $6.50 (14.50) paperback. Susan H. Cochrane IuW World Bank Staff Working Paper No. The Costs and Benefts of 416. September 1980. 41 pages Ana 8ng the Impact of FauPib Fmuing Frogrm (Including footnotes. references). Health Secem Prject George C. Zaldan Stock Mo. WP-0416. $3.00. p ie e from India, A technique for measurlng the Ghana and Thailand economic retun from Investing In Economics of Supplemental Rashid Farucqee population control. with an appraisal of inhement assumptions andi Feeding of Malnourished Revkws four categorl of health llmitatons. Children: Leakages, Cost, incatos (enviounent servies an B _ offmd servce rdved, and The Johns Hopkins Unluersily adnsse,its changes In mortailt, morbkd%& and 197. 62 pages (Including Odin rC Knudsen nutrtional status) In order to evaluate bibliography). Analyzes some of the economic the Impct of health pflbt In India, LC 70-15S166. ISBN104018-1317-4, feedIn of malnourihe hsudppleniental hrs, BankSff Woraiandg P400p(12.4papbck Demonstrates that supplemental World Bank Sta/f Working Paper feeding programs are economically 10. 546. 192. 44 pages. Aspects of Justifled if minimum Improvements ISBN1 0-8213-0117-9. $3.00. In mortality rates and more substan- Migrtion In West Af a tilal Increases In productivity take -Volume 1 place. K. C. Zachariah World Bank Staff Working Paper N1o. and others 451. Aprill981. Iu + 76 pages. Background data on migration In Stock No. WP-0451. $3.00. Benefis and osts o Food fourt er l-speaking countries: Rese_ft and Costs of Food Ghana. Sierra Leone, Liberla and _Itribution fli1clm The The Gambia. A regional analysis Expeients In Family na ae based on these studies is presented nring: Lessons from Pasquale L. Scandizzo and In MIWatdon In West Afrtca: teDvlpn ol OFlshri SwSal iy and Aspecs Roberto Cuca and Analy som of the dwaderists World Bank Staff Working Paper 11o. Catherine S Fierce and th main consequences of the 414. September 1980. ul + 363 pages ac hene reve fkod dstrlbudon poikcs folbwed by (Including statistical annexes. A compmhensive revlew of experi- the Indian govemnmt and provdes bibliography). mental efforts In the developing a quanticaton and a cost-benefit world to determine more effective anaysb of thdr effects on con- St1ck 1o. WP-0414. $15.00. ways of providing family planning sums producers, and the govern- services. ment budget Demogp pbh Aspects of t - Johns Hopkins Universily Press. World Bank Stf Working Paper Ilon in West Afhca 19,8. 276 pages (including bibilogra- o,. 509. August 1982.54 page -Volume 2 phy, index of experiments). ISB1 l 0-82-0011-3. $3.00. N. C. Zachaiah LC 77-16596. ISB 048018-2013-8. and others $19.50 (111.50) hardcover, ISBN0811018-2014-6, $895 (14.00) B nJord data on migratlon In perback. four lhmch-3peaking countrles: Ivory pp Coast Upper Volta, Senegal. and Togo. A regonal analysis based on Family Planng Progams: Health An Evaluation of Experience Fredrick Qolladay, Roberto Cuca coordinating author Integrating Famit Pla ing World Bank Staff Working Paper No. Draws on experience gained from with Headth Servie Does 345. July 1979. xil + 134 pages health components of seventy World It Help? (Including 2 annexes, refrences). Bank proJects in forty-four countries Rashid raruqee Stock Ho. WP-0345. $5.00.en 1975 and 1978. Stock Nto. WP-0345. ss.ao. Emphasizes the disproportionately Analyzes the findings of an experi- high expenditures Incurred on cura- ment carried out In iaramngwai a tive medicine, maintenance of expen- viliag In Funjab, India. between 1968 Fertility and Education: sive hospitals, and sophisticated and 1974 related to health care and What Do We Really Know? training of medical personnel at the family planning. The World Bank coi- Susan H. Cochrane cost of preventive care for the laborated with The Johns Hopkins A modl idetifyig themany han- majority of the people. Points out University In analyzing this data from A model IdenUfying the mnany chan- "tat low-cost health care systems are one of the best known and well-docu- ncis through which education might fansible and recommends that the mented fleld experiments in heath act to determine fertility and a review Bank begin reguiar and direct lend- L-are and family planning In the world. of the evidence of the relationinfohel,Inadtntoavg between education and the Interven- Ing for hcalth, In addition to having World Bank Staff Working Paper ing variables In the model that affect prohects in other sectors. No. S1S. September 1982. 47 pages. fertility. pSrQJe0-821n other2sectors. The Johns Hopkins University Press, Sector Policy Paper. February 1980. I 04213-0003.2. $3.00. 1979. 188 pages (including bibliogra- 90 pages (including 8 annexes, phy, Index). 4 figures, map). Englsh, rench, Kenya: Population LC 78-26070. ISBt 0-8018-2140-1, Japanese, Spanish, and Arabic. and Development $6.95 (4f475j paperback. Stock Mlos. PP-8001-E, PP-8001-F, (See descripton under Country PP-8001-J, PP-8001-S, PP-8001-A. Studis listing.) $5.00. Fertility and Its Regulation Manourished People In Bangladesh Health Issues and PoUdes A Poulcy View R. Amin and In the Developing Countries (See description under Development Rashid Faruqee Fredrick Goiladay iisdrig.) World Bank Staff Working Paper (1o. World Bank Staff Working Paper Nto. 38. April 1980. Iu + 50 pages 412. August 1980. 11 + 53 pages. MalnutrItlon and Povertys (including references). Stock (1o. WP-0412. $3.00. Magnitude and Policy Stock N1o. WP-038U. $3.00. Options MEW______ Shlomo Reutlinger and MEW Marcelo Selowsky lealth, Nutrtion, and The flrst large research effort In the Food Distribution and Family Planning In India: A World Dank to determine the global Rubition Intervention: The Surve of Experiments and dimension of malnutrition. Cae of ChlFe Spod Projcts The Johns Hopkins Uniuersity Prcss, Lloyd Harbert and Rashid Faruqee and 1976; 2nd printing, 1978. 94 pages Pasquale L. Scandizzo Ethna Johnson (including 5 appendLxes). The impact of Chile's Complementary Surveys fourteen experiments and LC 76-17240. ISBN10-8018-1868-0, Feeding Program (CFm), both on the special projects In health, nutrition, $4.75 (i2.85) paperback. direct and Indirect beneficiaries, is and family planning In India and pro- Spanish: Desnutricifn y pobreza: analyzed. Describes Chiies major poses guidelines for future Bank magnitudes y opciones de polftica. nutrition Intervention programs and projects on the basis of the survey. Editorial Tecnos, 1977. establishes the relative Importance of doraTens197 the CrP In terms of budgetary expen. World Bank Staff Working Paper ISB1 84-309-0726-2, 380 pesetas. dituwes and number of beneflciaries N1o. 507. February 1982. xi - 97 pages reached. Reviews briefly the pro- (Including references). Measuring Urban Mainutri- rfams patnd impotatial effectveness. Stock NIo. WP-0507. $5.00. tion and Poverts A Case Study of Bogota and CalL World Bank Staff Working Paper No. Colombia 512. May 1982. v + 45 pages (including Rakesh Mohan, bibliography, annex). M. Wilhelm Wagner, ISBt 0-8213-0001.6. $3.00. and Jorge Garcia Attempts to measure the extent of malnutrition and poverty In the dtis of Bogota and CalL Colombia. One of five papers resulting from a Population Plfcies and e ms rsearch program entlUed 'City Economic Development Heab Came in the Deloilng World: Stucdy a study of the workings of Timothy Klng and others (° f Sc y d Choke five maijor urban sectors In (btm 4CUC Colombia. The English-language edition is John R. evans Kamn Lashnman HaL and World Bank Staff Working Paper No. out prnt Word Bank Rprnt rs: umber 209. 447. April 1981. 80 pages (Including Spanish: Poifticas de poblaci6n y reprined hom ew CEngland Journal of Medicine, bibliography, appendixes). desarrollo econ6mico. Editorial vol. 305 IMouember 1981:U17-27. Stock 1o. WP-0447. $3.00. Tecnos. 1975. Stck Io. RP-0209. fre ofcharge. ISBt 84-309-0605-3, 440 pesetas. of Deprivaton and Migation In West Africa: frverty Based on the Proporton Demogaphl Aspects Populatlon Policy and speanto Food:An lastorY iL C. Zachariah and family Planning Programs: v.V. Bhanoji Rao Julien Conde Trends In Policy and World Bank Reprint Serics: Number 193. The first study of the large-scale Administration Reprinted 1mom World Deelopment uol. 9. no. 4 miovement of people In nine Wes Kandlah K~anagaratnam andStcNoRP,13 refchg. African countries. Dcisses the Cathtk KO. Ri-019e. Sr. of chage. volume and direction of Internal and external flows and the economlc and World Bank Staff Working Paper No. lutrition, ltealth. and education: The economic Significance of social characteristics of migrants. 411. August 1980. i1 + 22 pages Compiementarltles at Early Age A Joint World Bank-OECD study. Oxford (including footnotes). Marcelo Selowshy University Pfess. 1981. 166 pages Stock o1o. WP-0411. $3.00. World Bank Reprint Series: lumber 218. (including 22 maps. bibliography, Reprinted hom Journal of Development Index). Economics. uol. 9 11981.331-46. LC 80-21352. ISBN 0-19-520186-8 Regional Aspects of Family Stock lo. RP-0218. Fre of charge. LC9.9 80-2WS2 Ihardcover, Pianning and Fertility S19.9S 520187-6 $8.95 (f41SOJBehaior In Indonesia paperback. Dov Chemichovsky and Oey Astra Meesook Discusses the recent decline in Iutritlon and Food Reeds Indonesia's population growth rate In Developing Countries despite that country's relatively low Odin K. Knudsen and level of income and socioeconomic Pasquale L. Scandizzo development. Reviews the history and organization of the family plan- World Bank Staff Working Paper l1o. ning program and attempts to Iden- 328. May 1979. 73 pages (including tify those factQrs that have been 4 appendixes). responsible for Its success and Stock N1o. WP-0328. $3.00. assesses its prospects for the future. World Bank Staff Working Paper No. 462. Play 1981. 62 pages (including rFop alaUon and Pverq in appendL, references). the Devlioping World Stock No. WP-0462. $3.00. Mancy Birdsall World Bank Staff Working Paper No. 404. July 1980. 96 pages (including 2 appendixes, bibilography). Stock No. WP-0404. $3.00. WORLD BANK PUBUCATIONS ORDER FORM SEND TO: WORLD BANK PUBUCATIONS WORLD BANK PUBUCATIONS P.Q BOLX 37525 or 66, AVENUE D'ItNA WASHINGTON, D.C. 20013 75116 PARIS, FRANCE U.S.A. Name: Address: Stock or ISBN # Author, Title Oty. Price Total Sub-Total Cost: Postag & handling fee for more than two free Items ($1.00 each): Total copies: ..Air mall surcharge ($2.00 each): TAL PAYMENT ENCLOSED:__ Make checks payable: WORLD BANK PUBLICATIONS Preayment on orders from nDividuals Is rsquwJMd. Purcdae order are accepted fron booksellers library suppliers, IlIbaries, and Institutios. All prmicrsIncude oost of Wopote by fte leaM wMensive mearis. Th. prices and publicstion dates quoted In this Catalog are subject to chang wiftout notic. No refunds will be given for Iterns tha cwnno be fiNed. Credi will be applied tovrds future orders. No mom tha two free publiation wi# be p roide witout charge. Request for add Itwlonaopies wiU be filled at a charg of US 51.00 per copy to cover handling and postage costs. Airmail delivery will require a prhpayment of US m2.00 per copy. MaiIoder payment to the VId Tank need not be In U.S. dollar. but t amount rmttd must be at ft raft of mch on the day eth order is plae. The o rid Bank will ese ace Unesco W ounsan. HG 3881.5 .W57 W67 NO.623 c.3 LIPTON, IMICHAEL. DEMOGRAPHY AND POVERTY. DATE NAME AND EXTENSION NUMBER The World Bank Headquarters European Office Tokyo Office U 1818 H Street, N.W. 66, avenue d'1ena Kokusai Building Washington, D.C. 20433, U.S.A 75116 Paris, France 1-1 Marunouchi 3-chome Telephone: (202) 477-1234 Telephone: (1) 723-54.21 Chiyoda-ku, Tokyo 100, Japan Telex: WUI 64145 WORLDBANK Telex: 842-620628 Telephone: (03) 214-5001 RCA 248423 WORLDBK Telex: 781-26838 Cable Address: INTBAFRAD WASHINGTONDC ISSN 0253-2115/ISBN 0-8213-0286-8