Girls' Education is It ­ Nothing Else Matters (Much) By Surjit S. Bhalla, Suraj Saigal & Nabhojit Basu* February 27, 2003 *Oxus Research & Investments, New Delhi, India. Email: ssbhalla@vsnl.com This paper is prepared for the World Development Report 2003/04 on the delivery of social services 1 Executive Summary 1. This paper utilizes data from several sources to examine the levels, and trends, in living standards in different states of India (rural and urban areas considered as separate states). Two major results, supported by evidence at different levels of aggregation and different models, emerge. First, that girl and mother's education i.e. female education, is the single most important determinant of any improvement (change) in living standards in health and education. Today1, this result is part of the conventional wisdom. What is striking, though, is how all encompassing is the effect of female education. Knowledge of this variable explains practically all of the variation in changes in infant mortality 1983- 1999; knowledge of initial education of girls explains practically all of the variation in several variables (literacy, years of schooling, gender equality in schooling etc.) pertaining to education. 2. The second major result follows from this first result. If female education explains most of the variation, then it must mean that all of the other presumed determinants are not explaining much at all. What happened to state expenditures, growth in income, and the role of institutions and civil society? The paper examines the "contribution" of these other factors in as much detail as possible; unfortunately, the effects turn out to be insignificant or perverse i.e. state expenditures are negatively correlated with achievement. This negative finding could either correctly reflect the underlying reality, or be the outcome of mis-specification of the various models tried. The only additional variable that does seem to register a significant effect are private, household expenditures, on the respective outcomes. 3. There is one "surprising" conclusion that emerges in this study - it is that today (in 1999/2000) gender equality, in terms of schooling, has been achieved. Regardless of caste, religion, or income status, there is near convergence to equality. Defining schooling attainment for 5-18 year olds as the percentage of schooling years completed as a ratio of what they should have completed, given their ages, boys and girls fare equally. The aggregate all India ratio is 92 percent in terms of gender equality i.e. girls today have 92 percent of the education of boys aged 5-14, compared to three-fourths sixteen years ago in 1983. For the poor, the ratio was 65 percent in 1983; today it is 87 percent. 1See Bhalla-Gill(1990) for some early results for a cross-section of developing countries. 2 4. This result is universal. It applies across religion (Muslims, Hindus), across caste groups, across regions (urban/rural), across all states, and across income groups (poor, middle class and rich). The results are based on the large scale National Sample Surveys conducted in India in 1983, 1987-88, 1993-94 and 1999-2000. Cross checks of these data with other surveys and census data suggests that the results are accurate. 5. The average increase in schooling attainment over the years 1983-1999 is also explained mostly by one initial condition variable ­ the level of girl education in 1983. In other words, the lower the level of girl education in 1983, the greater the increase in the average education of the household, and the state. 6. Results pertaining to the decline in infant mortality between 1983-1999 point to only one initial condition variable; this variable is able to explain close to 80 percent of the decline. This is also a female education variable viz. the schooling achievement of adult females (18-40 years) in the household in 1983. 7. The three results together suggest both that nothing else matters and that the prognosis for future health and education improvements in India is very good. The equalizing nature in incomes due to equality in gender education should also not be under-estimated. 8. None of the popular determinants of living standards turn out to be significant. Private income growth ­ does not matter. Government expenditures ­ do not matter. Only initial conditions, outside of the purview of short-run policy, matters. Kerala has a low level of infant mortality because it had a high level of "adult female or mother's" education in 1980; it was lower in 1980 because such education was at a reasonably advanced level in 1960, and so on. 9. Given the disparate backgrounds which are demanding girls education, it is unlikely that civic society, panchayats, decentralization etc. are even minor determinants of this gender equality revolution in India. As mentioned above, income growth is also not even a minor determinant. 10. What appears to be happening is that parents are demanding more, and equal, education. If the school system is not providing education, parents, even poor parents, are substituting for such lack of governance by providing hard earned expenditures to educate their kids, especially girl kids. The poor are spending more than 5 percent of their expenditures on health and education, slightly less than half the average. This ratio has remained stable since 1983. 3 Section 1: Issues Explored This study aims to quantify the relationship between social service outputs and outcomes, examining, within a broader all-India context, the experience of five major states in particular: Himachal Pradesh, Kerala, Madhya Pradesh, Uttar Pradesh, and West Bengal. The period of study is 1980-2000, since reliable data in most cases is available only for the last two decades. In the Indian framework, it turns out that we lose only very little information by considering a relatively short time frame: over the 1960-80 period, GDP growth averaged less than one percent per annum; consequently few inroads were made in terms of poverty reduction and socio-economic development. Additionally, the 1980-2000 period is both long enough for real change (or its absence) to become apparent, and has been an era of relative dynamism in India. We concentrate largely on the fields of education and healthcare. The rest of this paper is organized as follows. Section 2 details the types of data and data sources used in this paper. The next section provides an overview of the existing literature on social services delivery in India, with an emphasis on works that concern our five study states. Section 4 looks at summary statistics on a range of income, expenditure, and human development indicators for our five states, providing a context for the two sections that follow. Section 5, which explores correlations between a range of dependent (infant mortality rates and several measures of educational achievement) and independent variables (income, expenditures, adult education, etc.); and Section 6, which develops "complete" (multi-linear) models to explain levels of, and changes in our chosen indicators of human development. Based on these findings, we look, in Section 7, at the issue of whether our five states will meet the much-cited Millennium Development Goals (MDGs) by the target year 2015. Section 8 concludes with a discussion of the policy implications of our findings. 1.1: Choosing the Comparator States Kerala has become almost a default choice for any Indian cross-state comparative study. On a range of socio-economic indicators, its achievements have been outstanding: comparable, in some respects, with the developed world, and frequently cited as a model for developing countries. However, even in 1960 Kerala was successful both relative to other states in India, and in comparison with considerably richer 4 countries in the west. It is useful, though, to examine Kerala's achievements both in absolute (level) terms, and in relative (change over time) terms, in order to better understand the context of its achievements. This helps answer some important questions: How did other states do during the same period of time, and, more importantly, which factors may have caused differences across states? What role, particularly, did initial conditions play? Are "successful" states able to maintain their "momentum" over time? West Bengal, like Kerala, has had an elected communist government for most of the last twenty years. This would presumably mean that social services in general, and targeted social services in particular, would receive greater-than-average attention in these two states. Assuming a strong correlation between outputs and outcomes, one should expect to find more rapid improvements in social outcomes in Kerala and West Bengal over the 1980-2000 period, especially compared with comparator states. Himachal Pradesh has, in recent years, been held up as India's new success story, with rapid improvements in education and health indicators. Again, it is important to see what lessons can be drawn from this small mountainous state. Madhya Pradesh and Uttar Pradesh lie on the other end of the spectrum, lagging, in many respects, behind the rest of India. Their massive size (in both geographic and population terms), wide gender-, caste- and religion-based disparities and low initial levels of achievement makes their inclusion in this study extremely useful for helping "correctly" derive the relationships between social sector achievements and policy-driven inputs. 1.2: Relating Social Outputs and Outcomes Which factors determine social sector outcomes? A straightforward (but misleading) answer might be: the quantity and quality of social services made available to a population over an extended time period. Both the "quantity" and "quality" of social services are, however, notoriously difficult to measure and few (if any) studies have convincingly addressed this issue; many, in fact, fail to distinguish between the two. Even if it were possible to arrive at accurate qualitative and quantitative measures of social services delivery, it would be difficult to sustain the argument that this single factor is solely (or even primarily) responsible for social sector outcomes. Instead, outcomes 5 are affected by a host of determinants, as well as by their interaction effects. Some important factors, which we attempt to quantify, include: 1. Economic growth at the aggregate (state) and disaggregated (by income or social group) levels. 2. Technological progress: this is particularly relevant to the field of healthcare (largely in the form of cheaper and more effective drugs), where improved technology has, over at least the last 50 years, enormously brought down costs, increased access, and greatly enhanced quality. It can be argued, convincingly, that technological progress is potent enough a factor to improve outcomes over time regardless of the impact of other factors; it becomes important, therefore, to try and separate the independent impact of technology on outcomes from impacts resulting from other factors. Historically, basic education has not gained as significantly from "technology effects" as has healthcare, but this is likely to change over the next decade or two with the continued downtrend in the costs of access to information technology. 3. Expenditure: this includes spending by governments, households/individuals, and non-governmental institutions. Government and non-governmental expenditure can be a useful proxy for measuring the quantity (though not the quality) of social service outputs, while private expenditure can be both a determinant of outcomes and an outcome of other factors. (For example, existing income, educational or health status can impact the composition of private expenditure, which can, in turn, affect future outcomes.) 4. Initial conditions: such factors as existing education levels (especially of the mother) and healthcare conditions, the degree of gender inequality, and a range of socio-economic and infrastructural conditions (e.g., achievements in land reform and the quality and reach of road networks) can have an enormous impact on social service outcomes in the medium- to long-term. The presence of certain initial conditions necessitates a change-based rather than a levels-based analysis of outcomes: it would not be very useful, for instance, to look at Kerala's current infant mortality rate in isolation of educational and health achievements that were in place two or three decades ago. Similarly, Himachal Pradesh's recent achievements on health and education cannot be fully understood in isolation of its past, and initial levels, of living standards. 6 5. Civil society institutions: the degree of decentralization (and, correspondingly, the strength of local self-governing institutions), the presence of non-governmental organizations, and the spread and depth of other democratic institutions. Since quantifying such variables is beyond the scope of this project (and known determinants are included in our models), it is assumed that the role of civil society institutions is captured by the regression residuals. If a positive residual emerges from a model, then it is likely that civil society institutions facilitated the process of improvement (i.e., helped make expenditures more efficient, or mother's education more effective). Correspondingly, if negative residuals are noted, then one of two scenarios are possible: (1) That, in the absence of civil society institutions, a state or region would have done worse than it did, and these institutions are having a compensating effect on outcomes; or (2) That available resources are not being optimally used, either by these institutions, or by the state. 1.3: Measuring Outcomes By arriving at a more complete understanding of the relative impact of different factors on outcomes, we can, it is hoped, design more effective policies. But measuring outcomes can be as tricky (and as contentious) as measuring outputs, since the quality of outcomes is somewhat intangible. Having noted this constraint, we use infant mortality as a proxy for healthcare achievement, and adult literacy, school attainment, school enrollment and school completion rates as proxies for educational achievement. Our choice of education indicators deserves some comment. Adult literacy rates are included, as they are in most studies, as an easy to interpret measure of basic educational achievement. Literacy, though, is not always an accurate indicator of change, and its impact on the economy is difficult to interpret. There is, importantly, a large "overhang" problem, caused by the fact that adult literacy data includes people nearing or past the retirement age. A more meaningful interpretation of literacy would require detailed distribution data, but this is not easily available. School enrollment rates are readily available but, again, are difficult to interpret since they usually refer to ever-enrollment rates (simply: did an individual ever, if even very briefly, attend school) rates within a particular age group; this figure that does not tell us 7 about the average level of schooling attainment since corresponding drop-out rates are often very high. The most interesting results on education in this study come from School Attainment rates that we construct from survey data. The availability of detailed household survey data at 4-6 year intervals across the 1980-2000 period allows us to construct a detailed estimate of school attainment at the all-India and state levels, as well as by socio- economic grouping. Various permutations are possible: we are, for instance, able to obtain data for an age group within a particular caste/religious group, subdivided by income levels, and either at an all-India level or at the state level. (One combination, out of many such, might be: school attainment for Muslim females from Uttar Pradesh aged 18-40.) School attainment in this context does not refer to absolute levels of attainment, but to average actual attainment relative to what attainment should have been. For instance, a seven year old child should have had one year of schooling (i.e., age-6 years), an eight year old two years, and so on. The household schooling attainment for each age-gender group (5-14, 5-18, 18-40, etc.) is thus a weighted average of individual achievements, and represents a percentage. Thus, if two female children in a household, aged 8 and 12, had, respectively, attained 1 and 4 years of schooling, the household schooling attainment for females aged 5-14 would be: (MaximumAtt n ActualAttainment 1 4 *100) ainment ( *100) + ( *100) 1 = 2 6 = 58.3% n 2 8 Section 2: Data Types and Sources Having established a basic framework for this study, it is useful to look at the types of data that are used in our analysis. Two basic "types" of data were collated: (1) Aggregate (state-level) data on government expenditures, health and education indicators, infrastructure, income growth, etc., obtained from censuses, national accounts, and similar sources; and (2) Micro (survey) data on household/per-capita expenditures, housing, education and health achievements and decisions, etc. Survey data was obtained from a range of sources, notably the National Sample Surveys (NSS) of 1983, 1987, 1993 and 1999, the National Family Health Surveys (NFHS) of 1992-93 and 1998-99, and from a number of specialized surveys.2 The NSS data were found to be especially useful since they were available at a highly disaggregated level, allowing for comparisons at the state and sub-state (urban and rural) levels, and on the basis of gender, religion/caste, and socio-economic status (including poor and non-poor). Since this study aims to look at trends over a fairly long time period (i.e., 1980 to 2000), and since it is essentially a "changes" analysis, we have attempted to collate data stretching back as far as possible. In some cases, the earliest available data dates to about 1983; for other types of data (certain health and educational indicators, income levels, and government expenditures), much longer time series are available. In all cases, the availability of a minimum of two data points (which are at least five years apart) was necessary for including a variable in our dataset. After gathering data from this diverse set of sources, we were able to pool the "macro" and "micro" data to form an exhaustive dataset. This allowed us to test a wide range of hypotheses concerning outputs, outcomes, and the possible correlations between the two, including lagged impacts. (For example, is government expenditure at the state level correlated with household level outcomes?) In addition, this allowed us to look at whether and how "macro" level outputs impacted sub-groups within states. 2The Government of India Planning Commission's National Human Development Report (2001) proved to be a particularly useful source for data compiled from a wide range of sources. 9 Section 3: Critical Literature Review A wide range of socio-economic literature relates social services outcomes to diverse, and often non-quantified (but not non-quantifiable) factors, such as political conditions, experiences with land reforms, and the spread of road networks. We examine below a sampling of this work. 3.1: Education and Politics Politics, or more specifically the highly-politicized position of teachers in India, is frequently cited as an important factor in determining educational outcomes. Nowhere is this more true than in the state of Uttar Pradesh (UP), where, as Kingdon and Muzammil (2001) argue, endemic teacher absenteeism and shirking have led to very poor educational outcomes. In turn, it is the strong political position of teachers in UP (and in several other Indian states) that explains absenteeism and a general lack of accountability. The Indian constitution provides for a special representation of teachers in the upper houses of the state legislatures; this has resulted in many teachers becoming deeply enmeshed in state politics. Over time, education has become highly politicized, and teachers' unions have grown in strength, leading to frequent, widespread, and astonishingly successful, teachers' agitations over pay and working conditions. 3.2: Land Reform, State Spending, Other Factors: The Case of Himachal Pradesh Himachal Pradesh (HP) has, in recent decades, dramatically improved educational outcomes. Here, as De et al (2000), illiteracy in 1961 was only slightly lower than in the four "BIMARU" States (Bihar, Madhya Pradesh, Rajasthan and Uttar Pradesh), and was worse than the all-India average; over the next forty years, though, it pulled well ahead of the Indian average in general, and the BIMARU states in particular. In 1991, in fact, it had a literacy rate of 61.9 percent, second only to Kerala's; by 1997, this had climbed up to 77 percent. (It must be noted, however, that HP shows very stark district level variations in literacy, ranging from 44.7 to 86.6 percent in 1997.) That Himachal Pradesh has managed such outcomes in spite of several large impediments ­ poor infrastructure, and the remote, mountainous location of many of its villages and towns ­ makes its achievement even more remarkable. 10 What explains Himachal Pradesh's success? In addition to a sustained and high level of expenditure by the state government, De et al find, the relatively egalitarian nature of Himachali society, a heightened sense of "unity" and common identity, and the limited role of caste barriers, have all allowed greater accessibility to education for all sections of society. (This sense of unity, argue De et al, has been bolstered by the early implementation of land reforms (beginning in the 1950s), which has made the distribution of power and status far more even than in many of India's states.) As importantly, the State has invested in public goods and other social services that have indirectly helped the spread of education. As early as 1951, it began building up its road network, allowing for easier accessibility in general. In addition, the government has ensured the provision of electricity to every HP village; it has a relatively wide-reaching public food distribution system, and has made significant progress in providing safe drinking water. 3.3: Healthcare in Kerala Sadanandan (2001) traces the historical underpinnings of Kerala's remarkable success story. He finds, importantly, that the erstwhile princely states that now comprise Kerala invested heavily in modern health services, especially when compared with the rest of British India. As a result, Kerala enjoyed a relatively wide and deep spread of hospitals and other health care facilities. This trend continued up to about 1970, when Kerala's fiscal problems caused a decline in budgetary allocations to healthcare, and a subsequent (relative) decline in the availability of healthcare, especially in rural areas. Although the private sector has filled some of the gaps (importantly, in rural areas) arising from the government's declining involvement, it has been unable or unwilling to extend the reach of services to historically under-served areas. It is important to note that in spite of Kerala's problems with health infrastructure in the recent past, the state has made remarkable progress in health care, particularly so with regard to infant mortality rates. 3.4: Decentralization and Social Service Delivery Outcomes Decentralization is the (not so) new buzzword in development planning, supposedly a panacea for the problems associated with top-down approaches to the delivery of social services. In India, this process received a major impetus from the 1992 constitutional amendments, which directed states to give a much larger role to panchayati raj (local 11 self governance) institutions. Prior to 1992, a few states, notably West Bengal, had taken important steps towards decentralization; since then, such states as Madhya Pradesh and Kerala have come to the forefront of decentralized planning and service delivery. Since decentralization per se is increasingly being looked at as a causal factor for improved social services delivery, it is useful to review the evidence of such a relationship. Mahal, Srivastava and Sanan (2000) find that decentralization, after controlling for socio-economic circumstances, the presence of civil society organizations, and the capture of local bodies by elite groups, is, indeed, associated with improved outcomes. A number of indicators of democratization and public participation ­ frequency of elections, presence of NGOs and parent-teacher associations, etc. ­ generally have positive effects; these effects, are, however, not always statistically significant. They caution, however, that it is too early as yet to comment on the sustainability of these efforts, and recommend further work on developing better measures of decentralization and social participation (e.g., data on candidate turnover from state-level elections). 3 Mehrotra (2001), on the basis of descriptive statistics, sees a stronger relationship between decentralization and improved social services delivery, particularly for the states of Rajasthan and Madhya Pradesh (MP). Mehrotra finds a large increase in school enrollment and in the number of schools opened in Madhya Pradesh following the introduction of the Education Guarantee Scheme (EGS) ­ taken to represent increased decentralization of education planning ­ in January 1997. While a total of 80,000 schools were opened in MP between 1947 and 1987, an additional 30,000 schools opened in the 1997-2000 period. Enrollment among girls (on an aggregate level), and among tribal children (regardless of gender) in particular, rose sharply during these three years. In Rajasthan, two initiatives in particular ­ the Shiksha Karmi Project (which began in 1987) and Lok Jumbish (1992) ­ are credited with enabling much of the State's relatively impressive literacy gains over the last decade. Here, the paper argues, the deepening of decentralization strengthened existing projects, whereas in MP, 3 A reasonable test for the efficacy of decentralization, which has not been adequately covered in the literature, is to relate decentralization to outcomes while keeping the social service inputs at a fixed level. Theoretically, the process of decentralization should ­ by enabling local self-governance and the increased role of NGOs and QGOs ­ on its own make the provision of social services more efficient. Hence one should expect improved outcomes even while keeping inputs fixed; in a regression analysis, this would take the form of an unexplained (and large) positive residual. 12 decentralization precluded the introduction of new initiatives. In both cases, Mehrotra enthusiastically finds that "deep democratic decentralization" is undoubtedly associated with improved educational outcomes ­ a finding that, the paper argues, holds equally for the provision of other types of social services. A few notes of caution are in order. A number of authors, particularly Behar and Kumar (2002) ­ who review the decentralization process in Madhya Pradesh ­ find stiff resistance to increased decentralization from a range of interest groups, particularly the bureaucracy and the political and socio-economic elite. Equally, they find a shortfall of capacities (such as the ability to keep accurate financial records, or to implement development programs) at the grassroots levels. Jha (2002) finds, after analyzing rural budget data for several states (including three of our study states), a notable slowdown in the fiscal and political devolution of authority in recent years; this is exacerbated by growing conflicts between local and state institutions. The Institute of Development Studies (2001) notes that even in Kerala, where a "People's Planning" campaign has been aggressively promoting the spread of decentralized planning, just 10 percent of all panchayats have been effectively incorporated in the planning process. Confirming this last finding, Nayar (2001) looks at the politics of decentralization of healthcare in Kerala. Despite achieving a large decline in infant mortality rates, Kerala currently has unduly high morbidity rates; the incidence of certain diseases, in fact, is on the rise. These complexities are likely to become more acute in the future, Nayar argues, unless certain crucial issues are addressed. First, Nayar finds, there is conflict between the professional and political leadership at the village level, where healthcare professionals are largely excluded from the planning process. Second, villages panchayats are frequently in conflict with the State government, particularly over such issues as drug supplies (which are controlled by the state), recruitment of staff and other management issues, and the allocation of funds for various programs. Third, Central and State government programs may conflict with each other, whereas panchayats are responsible for implementing both types of programs. Fourth, the devolution of financial and political powers to the village level has remained incomplete due to the opposition of State-level political leaders and bureaucrats. Finally, there is a great deal of confusion over the prioritization of preventive versus curative healthcare programs. 13 What lessons can be drawn from these diverse findings? Most importantly, there is a need for caution in finding causal links between greater decentralization and improved social services delivery outcomes. This is particularly true in states where decentralization has supposedly become deeply entrenched, or, conversely, where it may not have. In both cases, it is just as problematic to measure the true extent of decentralization as it is to measure its impact on social services delivery. 14 Section 4: Results: Summary Statistics 4.1: Income, and Expenditure on Social Services: The following two tables summarize trends in incomes over the 1980-2000 period, and in spending on two types of social services, education and healthcare. As the first table indicates, real per capita incomes in India rose significantly over the 1980-2000 period; this is in sharp contrast to the 1960-80 period, when average income growth averaged about one percent per annum. In terms of growth rates, Madhya Pradesh and West Bengal stand out as, respectively, the laggard and the leader among our group of five states; Uttar Pradesh remained the poorest of the five. Kerala and Himachal Pradesh, interestingly, had similar rates of growth and similar absolute levels of income at the two end points. Table 1: Per Capita Incomes (1993 Prices), 1980-2000 Average Income Per Capita Per Annum (Rs.) Annual (Log) 1980 2000 Growth Himachal Pradesh 5616 9614 2.69 Kerala 5561 9682 2.77 Madhya Pradesh 5175 7462 1.83 Uttar Pradesh 4196 6680 2.33 West Bengal 4983 9476 3.21 India 5071 9144 2.95 Turning to per capita annual expenditures on health and education, some interesting results emerge. Total spending on education4, which includes household and government spending, increased at a faster rate than the Indian average in just two of the sample states; West Bengal (92 percent growth) and Kerala (6 percent) represent, respectively, the maximum and minimum growth rates. Absolute expenditure on education was highest in Himachal Pradesh and Kerala, and lowest in Madhya Pradesh, in both periods. Himachal Pradesh, Kerala and West Bengal saw the largest increases in expenditure on healthcare, above the all-India average; the converse was true of Uttar Pradesh and Madhya Pradesh. 4 Total spending on education and healthcare (and therefore the shares of expenditure by states and households) are computed by combining survey data on household expenditures with budget data on state expenditures. 15 The state's share of per capita educational expenditure fell very sharply in every state barring West Bengal, where it rose by about 5 percentage points; the state's share of healthcare expenditure fell, usually sharply, in each sample state. Pooling both types of expenditure to obtain a proxy for total expenditure on "social services", we find that, except for West Bengal, every state has seen a sharp fall in the government's share of expenditure. Significantly, Uttar Pradesh in 1983 and Kerala in 1999 had the largest household shares in overall social services expenditures, while Himachal Pradesh had the lowest household shares in both years. Table 2: Spending on Education and Healthcare, 1983-1999 1983 1999 Share (%) Share (%) Total Total Change (Log) Annual Annual in Total Spending Spending Spending, (1993 Rs.) State HH (1993 Rs.) State HH 1983-99 Education Himachal Pradesh 392 79.7 20.3 695 73.2 26.8 57.1 Kerala 331 74.9 25.1 352 48.0 52.0 6.0 Madhya Pradesh 157 83.7 16.3 228 68.3 31.7 37.2 Uttar Pradesh 160 72.0 28.0 350 66.4 33.6 78.6 West Bengal 206 69.9 30.1 518 74.9 25.1 92.1 India 214 80.1 19.9 382 67.1 32.9 58.0 Healthcare Himachal Pradesh 289 57.1 42.9 505 41.2 58.8 55.8 Kerala 292 19.0 81.0 589 14.3 85.7 70.3 Madhya Pradesh 200 43.5 56.5 264 19.9 80.1 27.6 Uttar Pradesh 260 24.2 75.8 346 13.1 86.9 28.3 West Bengal 191 38.0 62.0 295 30.0 70.0 43.6 India 228 35.2 64.8 334 21.0 79.0 38.1 Education + Healthcare Himachal Pradesh 682 70.1 29.9 1200 59.7 40.3 56.6 Kerala 623 48.7 51.3 941 26.9 73.1 41.3 Madhya Pradesh 359 61.2 38.8 492 42.4 57.6 31.6 Uttar Pradesh 421 42.4 57.6 696 40.0 60.0 50.2 West Bengal 396 54.6 45.4 814 58.7 41.3 72.0 India 442 57.0 43.0 715 45.6 54.4 48.2 Notes: 1. Total spending on education and healthcare (and therefore the shares of expenditure by states and households) are computed by combining survey data on household expenditures with budget data on state expenditures. 16 4.2: Health Indicators: Infant Mortality Our five comparator states are a study in contrasts on the infant mortality scale, both in absolute (level) terms, and in relative (change) terms. Kerala is the clear outlier in both respects: not only has it brought down infant mortality from an already-low 54 in 1980 to an industrialized-country-standard 14 in 2000, but it also achieves a rare urban-rural parity in 2000, as well as the largest (log percent) improvement of all. Himachal Pradesh achieves an impressive 83 (log) percent decline overall, but this achievement must be seen in light of its high (compared with India's average) infant mortality rate of 143 in 1980. West Bengal, better than average in 1980, does well to achieve a 62 (log) percent decline overall. Madhya Pradesh and Uttar Pradesh both do poorly in comparison with the comparator states, as well as the Indian average: at all three levels (state, urban, rural), the infant mortality rates in both states were worse than the Indian average in 2000. Significantly, Uttar Pradesh achieves a small 22 percent decline in the urban infant mortality rate, while Madhya Pradesh in 2000 had the highest rural infant mortality rate, as it did in 1980. In each of the states barring Kerala, although urban-rural gaps have shrunk since 1980, they continue to remain high; in Madhya Pradesh, the 2000 urban- rural gap was a large 40 points. Table 3: Infant Mortality Rates Per 1000, 1980-20005 Total Urban Rural Period Period Period (Log) (Log) (Log) 1980 2000 Change 1980 2000 Change 1980 2000 Change Himachal Pradesh 143 63 -82.8 63 38 -51.6 146 64 -82.6 Kerala 54 14 -135.0 49 14 -125.3 56 14 -138.6 Madhya Pradesh 150 88 -53.3 105 54 -66.5 158 94 -51.9 Uttar Pradesh 130 83 -44.9 81 65 -22.0 139 87 -46.9 West Bengal 95 51 -62.2 59 37 -46.7 103 54 -64.6 India 115 68 -52.5 67 43 -44.3 123 74 -50.8 5The period of study for this paper is 1980-2000, but a vast amount of data, obtained from NSS household surveys, is available only for the years 1983, 1987, 1993 and 1999. Therefore, wherever a data point refers to 1980, 2000, or a non-survey year, the data has been obtained from non-survey sources (such as National accounts, censuses, or other government data sources). 17 4.3: Education Indicators 4.3.1: Adult Literacy6 Having noted earlier that adult literacy is not always the most accurate indicator of educational achievement, we still find some significant results from the available data. Kerala, as is frequently discussed, had achieved a very high level of literacy by 2000. Just as notably, the state is unusual in having a very low level of variation in terms of both gender and urban/rural residential status. Starting from a high base in 1980, however, the state has seen the lowest (log percent) increases in literacy over the last two decades; this is to be expected, though, since Kerala is getting closer to the "ceiling" level, i.e., universal literacy. Himachal Pradesh (HP) has made significant progress over the last twenty years, raising its overall literacy rate to 77 percent, its male literacy rate to 86 percent, and, most significantly, its female literacy rate to 68 percent in 2000. This is in sharp contrast to West Bengal, which although ahead of HP in 1980, witnessed slower growth in adult literacy than the Indian average, and thereby slipped well behind HP by 2000. Madhya Pradesh, and even more evidently Uttar Pradesh, witnessed the most impressive growth rates over the twenty-year period. These growth rates must be seen, however, in light of extremely low levels of literacy in 1980, and both states remained below the Indian average in 2000; this is particularly true of female literacy in Uttar Pradesh. Overall, female literacy rates increased much more rapidly (in many cases twice, or more, as fast) between 1980 and 2000 than did male literacy; once again, this is likely due to the "low-base" effect coupled with catch-up growth. We find similar trends, over the 1980-1995 period, for separate urban and rural data, with one very notable exception: male rural literacy in Madhya Pradesh grew faster (by 32 percent) than did female rural literacy (29 percent). 6 Adult literacy rates, and primary and middle school enrollment and completion rates are obtained from Government of India data obtained from a range of sources. In contrast, Schooling Attainment data (Section 4.4.4) are derived from NSS household survey data. 18 Table 4: Adult Literacy Rates, 1980-2000 Total Female Male Period Period Period (Log) (Log) (Log) 1980 2000 Change 1980 2000 Change 1980 2000 Change Himachal Pradesh 43.7 77.1 56.8 29.0 68.1 85.5 58.2 86.0 39.1 Kerala 78.1 90.9 15.2 70.8 87.9 21.6 85.9 94.2 9.2 Madhya Pradesh 35.6 64.1 58.7 22.7 50.3 79.4 47.7 76.8 47.6 Uttar Pradesh 30.8 57.4 62.3 14.1 43.0 111.7 45.6 70.2 43.1 West Bengal 48.1 69.2 36.4 33.2 60.2 59.7 61.2 77.6 23.7 India 40.8 65.4 47.1 25.7 54.3 74.7 54.9 76.0 32.4 Table 5: Adult Literacy Rates (Urban/Rural), 1980-1995 Total Female Male Period Period Period (Log) (Log) (Log) 1980 1995 Change 1980 1995 Change 1980 1995 Change Urban Himachal Pradesh 74.9 87.9 16.0 65.2 81.6 22.4 82.1 92.6 12.1 Kerala 84.5 93.2 9.8 78.4 90.7 14.6 90.9 95.8 5.2 Madhya Pradesh 62.4 74.1 17.1 45.8 61.5 29.5 76.4 85.1 10.7 Uttar Pradesh 53.6 68.7 24.8 39.1 57.9 39.3 65.3 78.2 18.1 West Bengal 69.8 80.7 14.5 60.1 73.9 20.7 77.2 86.2 11.0 India 65.1 77.0 16.7 51.9 67.4 26.1 76.3 85.7 11.6 Rural Himachal Pradesh 40.8 60.3 39.0 26.1 50.6 66.2 55.6 71.0 24.4 Kerala 76.6 88.3 14.2 69.0 84.2 20.0 84.7 92.8 9.1 Madhya Pradesh 28.4 39.1 31.8 16.9 22.7 29.3 39.5 54.4 32.0 Uttar Pradesh 25.5 39.5 43.6 8.7 20.7 86.6 40.9 57.0 33.2 West Bengal 39.0 54.9 34.2 23.0 41.4 58.9 53.9 67.5 22.5 India 32.8 46.4 34.7 17.6 31.7 58.7 47.4 60.6 24.6 19 4.3.2: Primary School Enrollment and Completion In terms of both primary school enrollment rates and completion rates, some very striking results emerge. Kerala and Uttar Pradesh witness a drop in completion and enrollment rates (with the exception of girls' completion rate in Kerala) during the twenty year period, but for (largely) opposite reasons: Kerala's decline (from a very high base) is largely due to a drop in ever-enrollment rates, while Uttar Pradesh's (from a very low base) seems to be driven by a large increase in drop-out rates. Uttar Pradesh, in fact, sees the largest declines in completion rates. In contrast, Madhya Pradesh (MP) witnesses a surprisingly large increase in completion rates, particularly for girls, pulling it up from average (state level) or below average (girls) to well above the all-India average. West Bengal, well ahead of MP in 1980, falls behind the all-India average with a languid growth in completion rates. Himachal Pradesh makes slow progress (although girls do much better than boys in terms of completion rates), but remains, as in 1980, ahead of the Indian average on most counts. Table 6: Primary School Enrollment & Completion Rates, 1980-2000 Total Girls Boys Period Period Period (Log) (Log) (Log) 1980 2000 Change 1980 2000 Change 1980 2000 Change Enrollment Rates Himachal Pradesh 97.0 88.6 -9.0 85.0 82.7 -2.7 108.6 95.2 -13.2 Kerala 107.2 87.1 -20.7 104.5 86.5 -18.9 109.9 87.7 -22.6 Madhya Pradesh 68.2 118.4 55.2 45.2 106.5 85.7 90.6 129.9 36.1 Uttar Pradesh 62.5 65.7 5.0 40.3 50.3 22.2 81.8 79.9 -2.4 West Bengal 93.9 107.2 13.2 81.9 103.3 23.2 105.4 110.9 5.1 India 80.5 92.6 13.9 64.1 83.2 26.0 95.8 101.5 5.8 Completion Rates* Himachal Pradesh 59.2 64.3 8.3 52.1 62.9 18.8 66.0 66.2 0.2 Kerala 99.7 94.0 -5.9 90.0 91.7 1.9 106.1 96.2 -9.8 Madhya Pradesh 42.4 99.2 84.9 22.1 84.7 134.6 62.7 113.0 59.0 Uttar Pradesh 32.3 28.6 -12.3 23.9 19.1 -22.5 57.4 37.6 -42.3 West Bengal 43.9 52.0 17.0 36.5 44.5 19.8 50.7 59.7 16.3 India 42.6 55.1 25.7 31.3 47.8 42.5 53.3 62.0 15.1 Notes: 1. * Completion Rates are estimated from enrollment and drop out rates using the following formula: CompletionRate = EnrollmentRate*(100 - DropoutRate) 20 4.3.3: Middle School Enrollment and Completion At the middle school level, enrollment and completion rates have followed slightly different patterns than those observed for primary schooling. Madhya Pradesh once again does exceedingly well, showing the largest improvement (albeit from a very low base) in both enrollment and completion. Himachal Pradesh, unlike at the primary school level, witnesses a very large increase in enrollment rates, while Kerala shows a small but positive increase from a high base. Completion rates in these two states go up significantly (to over 100 percent in Kerala), but girls in Himachal Pradesh see the most dramatic improvement in absolute as well as growth terms. Uttar Pradesh does very poorly on completion rates, experiencing a (log)16 percent drop in overall completion rates, and a 32 percent drop in boys' completion rates; girls experience a 37 percent increase, but only reach a shockingly-low 7.9 percent. West Bengal, as with primary schooling, experiences low growth (from a low base), taking it from about average in 1980, to well-below the Indian average in 2000. Table 7: Middle School Enrollment & Completion Rates, 1980-2000 Total Girls Boys Period Period Period (Log) (Log) (Log) 1980 2000 Change 1980 2000 Change 1980 2000 Change Enrollment Rates Himachal Pradesh 63.8 96.4 41.3 41.0 89.2 77.8 85.9 103.9 19.0 Kerala 89.4 97.3 8.5 84.4 94.8 11.6 94.4 99.8 5.6 Madhya Pradesh 33.4 61.1 60.4 17.4 50.1 105.7 48.2 71.2 39.0 Uttar Pradesh 38.7 37.4 -3.4 18.0 25.2 33.8 56.3 48.1 -15.8 West Bengal 45.3 52.2 14.2 37.0 44.4 18.1 53.3 59.7 11.3 India 41.9 56.6 30.1 28.6 48.4 52.5 54.3 64.1 16.5 Completion Rates Himachal Pradesh 41.2 80.1 66.4 20.9 71.0 122.4 63.6 89.8 34.4 Kerala 77.8 109.2 33.9 73.1 104.0 35.3 82.6 114.4 32.6 Madhya Pradesh 15.2 33.3 78.5 5.4 23.6 147.0 25.3 42.9 52.7 Uttar Pradesh 17.2 14.6 -16.2 5.5 7.9 37.0 28.9 21.0 -31.8 West Bengal 15.5 17.7 13.4 11.2 13.2 16.4 19.9 22.5 12.5 India 15.6 26.2 51.9 7.7 20.8 98.6 20.2 31.3 43.9 Notes: · Completion Rates are estimated from enrollment and drop out rates using the following formula: CompletionRate = EnrollmentRate*(100 - DropoutRate) 21 4.4.4: Schooling Attainment As discussed earlier in this paper (see Section 1 on "Issues Explored"), the availability of detailed survey data on a household level allows for a comprehensive analysis of school attainment. The results that come out of this analysis are highly significant, and indicative of a massive change in gender, religion/caste or income-based disparities in educational attainment, both across India, and in specific states.7 There is overwhelming evidence of convergence in educational attainment, with females, the poor, religious minorities, and "backward" castes witnessing disproportionately large increases. Almost all of these changes result from catch-up growth: if a group start from a low base, it will experience relatively larger increases. Some caveats are in order. Although there is strong evidence to show that convergence is taking place across socio-economic and gender lines, this analysis does not necessarily imply a huge increase in absolute levels of educational attainment, i.e., girls may be rapidly catching up with boys, but the average number of years of schooling for both boys and girls is still low. Nor does this analysis imply an improvement in the quality of education, which, as noted elsewhere, is almost impossible to measure. Finally (and this is most relevant to schooling in the 5-14 year age group), this analysis is based on the existing (and not the potential) sex structure of children within a household, i.e., there may be some self-selection bias since households that choose to "keep" their female children may be more inclined to treat their female children better.8 Even after noting these caveats, the results are striking and unambiguous. Looking at almost every possible combination of income/religion/gender/caste, there emerges a clear convergence of educational attainment in the 5-18 year age group between rich and poor, between males and females, across religions, and across castes. The table below shows just some of these results. Looking at the female-male ratios of educational attainment and the number of years of schooling, we find a sharp increase (to about 90 percent in some cases) between 1983 and 1999, in both the general population, and 7A detailed study on the effects of educational change by religion, caste, etc is beyond the scope of this paper. However, a few key statistics, presented in Bhalla (2003), are abstracted here; see this paper for details. 8India, as documented by several studies, has a very low female-male ratio in the overall population as well as in the sub-adult age group. This is the result of a high degree of female infanticide, and of higher infant mortality among girls than boys, primarily due to neglect of female children. 22 among the poor. A ratio of 90.6 (schooling attainment, ages 5-14), for example, implies virtual gender equality in attainment for that particular age group; the results are even more striking for urban India, where the ratio is 97 percent on average, and also 97 percent for the poor. There is convergence, too, in the 18-40 age group, although the relatively low ratios are the result of very long-term trends. It is very likely that these ratios will begin to approach 100 percent in the coming decades as the present generation of children ages. Table 8: Schooling Attainment and Schooling Years: India, 1983-1999 Poor All 1983 1999 1983-99 1983 1999 1983-99 Schooling Attainment (Ages 5-14) Male 69.4 97.7 34.2 84.5 105.0 21.7 Ratio (Female/Male) 64.5 86.9 29.8 75.0 90.6 18.9 Schooling Attainment (Ages 18-40) Male 23.1 27.0 15.6 37.0 40.2 8.3 Ratio (Female/Male) 34.4 47.9 33.1 48.0 61.6 24.9 Education Years (Ages 5-14) Male 2.5 3.4 30.7 3.2 4.0 22.3 Ratio (Female/Male) 60.1 84.0 33.5 71.0 88.5 22.0 Education Years (Ages 18-40) Male 3.6 4.2 15.4 5.8 6.3 8.3 Ratio (Female/Male) 34.2 47.4 32.6 47.8 61.3 24.9 Source: National Sample Survey data, 1983 and 1999. Looking at our sample states, the narrowing of educational gaps is evident across almost any "line" that one can draw. It is most telling (and meaningful), however, to look at data on female-male ratios of educational achievement for two age groups: 5-18 and 18-40. Tables 9a to 9d present data on this indicator for our five states, for both rural and urban areas, and for two population groups (Muslim and SC/ST). Over the 1983-1999 period, there has been a rapid convergence of educational achievement between males and females, regardless of urban-rural status, religion, or class. This is particularly true of the 5-18 age group, which responds faster to changes in household decisions and state policy. The fact that adult females too have made rapid strides is a strong indication of the depth of change in relative educational achievements across India. In 23 some cases, (e.g., adult Muslims in rural Himachal Pradesh) the ratio has more than doubled in the last sixteen years; in others, the ratio approached complete equality in 1999. Rural areas have, in general, witnessed a bigger change than have urban areas; this further confirms our general finding of rapid catch up in educational attainment. Table 9a: Female-Male Ratio of Educational Achievement, 5-18 Age Group, 1983 & 1999, Rural Areas State Muslim SC/ST 1983 1999 1983 1999 1983 1999 Himachal Pradesh 79.6 100.2 36.6 98.3 81.2 100.6 Kerala 97.6 100.2 93.7 94.0 98.1 100.0 Madhya Pradesh 53.2 83.6 42.0 88.8 54.7 83.3 Uttar Pradesh 49.1 78.4 37.1 67.7 53.0 79.2 West Bengal 78.3 91.2 79.3 93.7 83.4 91.9 Table 9b: Female-Male Ratio of Educational Achievement, 5-18 Age Group, 1983 & 1999, Urban Areas State Muslim SC/ST 1983 1999 1983 1999 1983 1999 Himachal Pradesh 95.9 100.2 58.3 97.7 95.7 99.7 Kerala 98.9 102.0 96.6 98.3 98.8 100.8 Madhya Pradesh 92.0 96.7 79.8 99.8 93.6 98.0 Uttar Pradesh 84.6 94.3 70.5 91.7 86.0 94.0 West Bengal 95.0 94.5 103.6 98.3 93.5 96.1 Table 9c: Adult (18-40 Age Group) Female-Male Ratio of Educational Achievement, 1983 & 1999, Rural Areas State Muslim SC/ST 1983 1999 1983 1999 1983 1999 Himachal Pradesh 46.2 73.1 21.5 68.2 48.7 73.9 Kerala 86.5 94.8 73.7 91.2 86.9 95.5 Madhya Pradesh 22.8 38.5 30.6 40.3 24.0 40.6 Uttar Pradesh 22.4 37.0 19.3 27.1 24.7 40.8 West Bengal 45.9 58.9 39.0 52.2 48.8 59.2 24 Table 9d: Adult (18-40 Age Group) Female-Male Ratio of Educational Achievement, 1983 & 1999, Urban Areas State Muslim SC/ST 1983 1999 1983 1999 1983 1999 Himachal Pradesh 73.6 95.8 81.3 N/A* 75.7 97.0 Kerala 93.4 99.8 80.4 94.2 93.5 99.5 Madhya Pradesh 58.1 75.1 46.2 72.1 61.0 76.5 Uttar Pradesh 59.1 79.0 49.3 69.3 62.4 81.5 West Bengal 79.1 85.1 39.5 68.0 80.8 86.6 Note: * Due to errors resulting from a small sample size, the female-male ratio of educational achievement among Muslims in Himachal Pradesh exceeds 100 percent. 25 Section 5: Results: Correlation Analysis 5.1: Methodology Three separate methods were used to study the determinants of levels and changes in infant mortality and educational achievement (four indicators). The first method involved simple correlation analysis. The independent variables used were the "standard" determinants (including measures of income, poverty, inequality, and expenditure), which were also used at the later stages of analysis. Second, through a series of regressions, we looked at the determinants of levels of infant mortality at each point in time, i.e., 1983 and 1999. It was found that certain factors that had a significant impact in 1983 were not important determinants in 1999, and vice versa. This suggests that structural, socio-economic, and technological changes strongly affect the relative importance of these presumed determinants. Third, we analyzed the living standards indicators in terms of changes between 1983 and 1999.The latter two methods (see Section 6) incorporated ­ in addition to regression analyses ­ an analysis of residuals (i.e., the gap between actual and predicted values) of each model. This allowed us to see how much above or below its expected "performance" each state (or urban and rural areas within a state) was at a point in time, and whether these gaps have tended to grow or diminish. 5.2: Correlation Statistics: Infant Mortality & Education Variables At the initial, exploratory stage of this research, we looked at a series of simple two-way correlations (in 1983 and 1999, separately) between indicators of education and infant mortality (the dependent variables) and various independent variables. The results help to critically re-examine certain fundamental relationships that have been expounded in the delivery of social services literature. The tables below summarize these correlation results. 26 Table 10: Correlations with Infant Mortality Rates Logistic Index of Infant Log Infant Mortality Mortality 1983 1999 1983 1999 Poverty Head Count Ratio (%) 0.11 0.33 0.11 0.24 Gini (Consumption) -0.55 -0.61 -0.56 -0.42 Adult (18-40) Years of Education -0.77 -0.74 -0.76 -0.56 Per Capita Expenditures: Private Consumption (Total) -0.32 -0.50 -0.28 -0.29 Private Spending on Education -0.56 -0.50 -0.55 -0.29 Private Spending on Health -0.25 -0.69 -0.24 -0.62 Private Spending on Edu+Health -0.45 -0.65 -0.44 -0.50 State Spending on Education -0.14 0.10 -0.17 0.22 State Spending on Health 0.28 -0.20 0.30 -0.10 State Spending on Edu+Health 0.02 0.03 0.00 0.15 Table 11: Correlations with Selected Education Indicators (1) Average Years of Average Years of Education Education Adult Literacy (5-18 Age Group) (18-40 Age Group) 1983 1999 1983 1999 1983 1999 Poverty Head Count Ratio (%) -0.20 -0.35 -0.30 -0.43 -0.25 -0.31 Gini (Consumption) 0.59 0.76 0.60 0.68 0.64 0.82 Adult (5-18) Years of Education 0.97 0.96 0.95 0.88 1.00 1.00 Per Capita Expenditures: Private Consumption (Total) 0.57 0.63 0.58 0.64 0.61 0.61 Private Spending on Education 0.75 0.73 0.73 0.64 0.83 0.81 Private Spending on Health 0.32 0.68 0.36 0.71 0.35 0.65 Private Spending on Edu+Health 0.58 0.77 0.60 0.74 0.64 0.80 State Spending on Education 0.34 0.17 0.36 0.08 0.25 0.17 State Spending on Health 0.04 0.26 0.16 0.38 0.09 0.27 State Spending on Edu+Health 0.30 0.24 0.44 0.18 0.29 0.23 27 Table 12: Correlations with Selected Education Indicators (2) Female-Male Ratio of Educational Attainment 5-18 Age Group 18-40 Age Group 1983 1999 1983 1999 Poverty Head Count Ratio (%) -0.24 -0.26 -0.25 -0.35 Gini (Consumption) 0.41 0.53 0.56 0.72 Adult (5-18) Years of Education 0.86 0.67 0.91 0.91 Per Capita Expenditures: Private Consumption (Total) 0.65 0.61 0.60 0.60 Private Spending on Education 0.62 0.57 0.71 0.72 Private Spending on Health 0.34 0.57 0.43 0.68 Private Spending on Edu+Health 0.53 0.63 0.64 0.77 State Spending on Education 0.37 0.04 0.35 0.10 State Spending on Health 0.10 0.29 -0.00 0.22 State Spending on Edu+Health 0.36 0.15 0.32 0.18 5.2.1: Poverty Are higher rates of poverty associated with lower educational achievements or higher rates of infant mortality? Much of the development literature suggests so, but simple correlation statistics suggest only a weak (albeit gradually strengthening) relationship. The correlation between infant mortality and poverty head count ratios (on a minus 1 to plus 1 scale) was an almost insignificant 0.11 in 1983, but increased to a mild 0.33 in 1999. (Correlating logistic indices of infant mortality with poverty, we find an even weaker relationship.) Similarly, five different indicators of educational attainment (adult literacy, and average years of education and female-male ratios of achievement in the 5- 18 and the 18-40 age groups) are mildly correlated with poverty; all of these correlations, however, have grown stronger between 1983 and 1999. The correlations between poverty and three educational indicators (adult literacy, average years of education (5-18 age group), and the female-male ratio (18-40 age group) of achievement) have increased very significantly over the last 16 years, while this has not been the case for the other two indicators. 5.2.2: Inequality Inequality in general, and income inequality in particular, has in the past been linked to lower educational and health achievements. We find the opposite: not only is greater inequality strongly correlated with lower infant mortality rates and higher schooling achievements, but this relationship has grown stronger over the years. In all cases but 28 one (logistic indices of infant mortality), there has been a significant increase in the correlation coefficient between 1983 and 1999. This is especially true of educational achievement and the female-male ratio of educational achievement within the 18-40 age group, and suggests the presence of a "catch-up" effect in levels of and disparities in education. 5.2.3: Public Spending on Education and Health One of the most important debates in development policy today concerns the efficacy of government spending on, or government provision of, education and healthcare services. A simple first test of the usefulness of government spending might compare state (per capita) expenditure with education and health achievements. Our results in this regard are both striking and counter-intuitive. Correlating state expenditure on health and education (combined) with infant mortality rates and educational achievement, we find, in most cases, a small-to-insignificant relationship. Moreover, most correlation coefficients have declined between 1983 and 1999, sometimes by more than half. This suggests, at first glance, that state spending is not having its desired (positive) effect, and that it is becoming less important over time. This preliminary conclusion will be looked at in greater detail in the next few sections. 5.2.4: Adult Education Higher levels of education among working and child-bearing age adult women (taken, in this case, to be the 18-40 age group), has been found in various studies to be positively associated with health and education indicators. Our results strongly support such conclusions. Several indicators of educational achievement, and (log) infant mortality levels, are strongly correlated with higher average adult years of education. Moreover, in most cases, correlations with adult education have increased over the years. Very strikingly, though, correlations with logistic indices of infant mortality, and with two indicators of education for the 5-18 age group (average years of education and the female-male ratio of education), have declined between 1983 and 1999. This suggests the impact of adult education on infant mortality has a "threshold effect", i.e., beyond a certain level of adult education, extra years of education have a non-linear and declining impact on health and education achievements. 29 5.2.5: Private Consumption and Health & Education Expenditure There are strong positive correlations between both average per capita consumption (total) and private spending on education and health, and lower infant mortality rates and higher educational achievements. In some cases, such as the correlation between infant mortality rates and private spending on health, there has been a dramatic strengthening between 1983 and 1999. Equally striking is the relationship of spending on health with a number of educational indicators. This suggests that improvements in educational achievement, and especially improvements in gender equality in educational achievement, are positively linked to private expenditure decisions. 5.3: Basic Variables: Graphical Relationships The results above are graphically illustrated in the charts below, which relate infant mortality and education to: (1) per capita consumption; (2) private expenditure on education and health; (3) public expenditure on education and health; (4) adult educational attainment of males and females. Per capita consumption, as illustrated in Graph 1, had a mild and perverse (negative) relationship with infant mortality in 1983 (i.e., higher incomes were associated with higher infant mortality rates), but not so in 1999 (except for urban areas). Public expenditure on education and health (Graph 2) had virtually no bearing on outcomes in either year, while private expenditures were only weakly related to infant mortality rates (Graph 2). The strongest results, though, concern the educational attainment of adult males and females (ages 18-40), which are taken as proxies for, respectively, father's and mother's education: as expected, there is a strong negative relationship between adult education and infant mortality in both 1983 and 1999 30 Graph 1: Per Capita Consumption vs. Infant Mortality/Education Years, 1983 and 1999 Infant Mortality Rate(log) Fitted Values Rural Infant Mortality Rate(log) Fitted Values Rural Fitted Values Urban Fitted Values Urban 5.5 xr MPr 4.5 UPr xr xr xr xr xr xr xuUPu xr xr HPr xu xrMPr xr xr xr xu 4 xu WBr MPu 5 HPr UPr xr xr xrxr xu xr xr HPu WBu xuxu xu xu xu WBr MPu xu 3.5 xu xr xrxu xu 4.5 xr xu UPu xu xu xu 3 xu xu HPu xu xu WBu KERALr 4 KERALrKERALu KERALu 2.5 5.5 6 6.5 7 5.5 6 6.5 7 Per Capita Consumption (log) Per Capita Consumption (log) 1983 1999 Edu Years(log),5-18 Group Fitted Values Rural Edu Years(log),5-18 Group Fitted Values Rural Fitted Values Urban Fitted Values Urban 2 2 KERALu HPu KERALu KERALr KERALr HPu xu xu xu xu xuxu WBu xuxuxu HPrxu xu xu xu xu xu xuMPu xuxr xu xu xu 1.5 HPr xr MPu WBu xr xu xr UPu xu xu xr xr xu xrxr xrxr UPu xr 1.5 xr xrxr xr WBr WBr xr xr 1 xr xr UPr UPr MPr xr xr xr MPr xr xr .5 1 5.5 6 6.5 7 5.5 6 6.5 7 Per Capita Consumption (log) Per Capita Consumption (log) 1983 1999 Note: Fitted values are estimated from a simple linear regression relating the dependent variable (y-axis) to a single independent variable (x-axis). 31 Graph 2: Infant Mortality vs. Private and Public Expenditure on Education and Health, 1983 and 1999 Infant Mortality Rate(log) Fitted Values Rural Infant Mortality Rate(log) Fitted Values Rural Fitted Values Urban Fitted Values Urban 5.5 5 xr MPr xr UPr xr xr xr MPr xr xr xu xr xr HPr UPuxr 5 HPr xr xr xu xr xu UPr 4 WBr xu MPu xr xr xr xr xu xr xr xu WBu xu HPu xu xuxu xu WBr MPu xu xr xr xu xu xu 4.5 xr xu UPu 3 xu xu KERALrKERALu xu xuxu xu HPu WBu KERALr 4 KERALu 2 1 2 3 4 2 3 4 5 Private Exp on Edu+Health(log) Private Exp on Edu+Health(log) 1983 1999 Infant Mortality Rate(log) Fitted Values Rural Infant Mortality Rate(log) Fitted Values Rural Fitted Values Urban Fitted Values Urban 5.5 xr MPr 4.5 UPr xr xr xr xr xu xr xr UPu HPr xr xr xu MPr xr xr xr xu xr 4 MPu xu WBr 5 HPr UPr xr xrxr xr xu xr xu xu HPu xu xu WBu xu WBr MPu xu 3.5 xu xr xu xr xu 4.5 xr xu UPu xu xu xu 3 xu xu xu HPu WBu KERALr 4 KERALu KERALr KERALu 2.5 2.5 3 3.5 4 2.5 3 3.5 4 State Exp on Edu+Health(log) State Exp on Edu+Health(log) 1983 1999 Note: Fitted values are estimated from a simple linear regression relating the dependent variable (y-axis) to a single independent variable (x-axis). 32 Graph 3: Private and Public Expenditure on Education and Health vs. Years of Education (Females, 18-40 Age Group) , 1983 and 1999 Edu Years(log),AdultFemales Fitted Values Rural Edu Years(log),AdultFemales Fitted Values Rural Fitted Values Urban Fitted Values Urban HPu KERALu 3 xu xu 2 KERALr xu WBu xu xu xu xu xu xu MPu xu UPu xu xu HPu xu xu KERALu xu xuxu KERALr xu xu xr HPr xr 2 xu WBu 1 WBr xu xr MPuxu UPuxu xuxu HPr xu xr xr xr xr xr xr xr xr xrxr xr xr xr WBr xr 0 UPr 1 xrxr MPr xr UPr xr MPr xr xr xr -1 0 1 2 3 4 2 3 4 5 Private Exp on Edu+Health(log) Private Exp on Edu+Health(log) 1983 1999 Edu Years(log),AdultFemales Fitted Values Rural Edu Years(log),AdultFemales Fitted Values Rural Fitted Values Urban Fitted Values Urban KERALu HPu 3 xu 2 WBu KERALrxu xu xu xu xu xu UPu xuMPu xu xu HPu KERALu xu xu KERALr xr HPr xr 2 xu xu xu xu xu WBu 1 WBr MPu xu xu xr xu UPu xu xu xu HPr xr xr xr xr xr xr xr xr xr xr xr xr xr WBr 0 UPr 1 xr xr xr MPr UPr MPr xr xr xr xr -1 0 2.5 3 3.5 4 2.5 3 3.5 4 State Exp on Edu+Health(log) State Exp on Edu+Health(log) 1983 1999 Note: Fitted values are estimated from a simple linear regression relating the dependent variable (y-axis) to a single independent variable (x-axis). 33 Graph 4: Infant Mortality versus Education of Adult Males and Females, 1983 & 1999 Infant Mortality Rate(log) Fitted Values Rural Infant Mortality Rate(log) Fitted Values Rural Fitted Values Urban Fitted Values Urban 5.5 6 xr MPr 5 xr 5 HPr UPr MPr xr xr xr xr xr xr UPr xr xr xr xr xr xr xr xr HPr UPuxu xu xr xr xr xu xu 4 WBr MPu xu WBr xu MPu xu xr xr xu xu xu WBu xu xu HPu xu 4.5 xu xr xu UPu xu xu xu 3 xu xuxu xu HPu KERALrKERALu WBu KERALr 4 KERALu 2 1 1.5 2 2.5 1 1.5 2 2.5 Edu Years(log), Adult Males Edu Years(log), Adult Males 1983 1999 Infant Mortality Rate(log) Fitted Values Rural Infant Mortality Rate(log) Fitted Values Rural Fitted Values Urban Fitted Values Urban 5.5 8 xr MPr xr 5 HPr UPr 6 xr xr xrxr xr xr xu WBr MPu xu xr xr xr xu MPr xu xr UPr xr xr 4.5 xrxr xr HPr UPuxu xu xu xr xr WBr xr xr xr xu xu MPu UPu 4 xu xu xu WBuxu xuxu xu HPu xu xu xu xu xu xu HPu WBu KERALrKERALu KERALr 4 KERALu 2 -1 0 1 2 0 1 2 3 Edu Years(log),AdultFemales Edu Years(log),AdultFemales 1983 1999 Note: Fitted values are estimated from a simple linear regression relating the dependent variable (y-axis) to a single independent variable (x-axis). 34 Section 6: Regression Results The previous section looked at partial relationships between our indicators of living standards and a number of possible determinants; this section develops "complete" models to explain both levels and changes in these indicators across our sample states. This is done in order to check the robustness of our earlier results, and to test which determinants continue to be significant in a multi-linear regression context. 6.1: Infant Mortality 6.1.1: Levels Analysis The following model was used to estimate the determinants of infant mortality at a point in time: Log(IMt ) = f (X1, X2,Z) (Equation 1) Where: (I) X1 is a vector of consumption variables: (1) Per capita consumption (log) at time t (2) Per capita private expenditure (log) on education and health (3) Per capita public expenditure (log) on education and health (I) X2 is a vector representing the average years of schooling of adult females (II) Z is a vector of exogenous variables: (1) An urban/rural dummy variable (2) The poverty head count ratio (3) The consumption Gini coefficient (4) The percentages of a population that are Muslims and scheduled castes / scheduled tribes. Additionally, in order to test the robustness of this model, a logistical index9 of infant mortality was estimated using the same parameters: 9A logistical index is based on the premise that, for certain human development indicators (and especially infant mortality), a natural "floor" or "ceiling" exists. These indices therefore attempt to measure how far a country/state/region is from the floor or ceiling level. For this analysis, we take the floor level for infant mortality to be 5 per 1000 live births. 35 LogisticIndex(IMt ) = f (X1, X2,Z) (Equation 2) Table 13 summarizes the regression results from this model for 1983 and 1999 Table 13: Level Regressions: Infant Mortality Rates in 1983 and 1999 Log Levels Logistic Index Levels (Model 2) (Model 1) Independent Variable 1983 1999 1983 1999 Constant Term 4.32 5.54 -7.05 -3.73 (5.13) (4.78) (-1.46) (-0.14) (Log) Consumption Per Capita -0.06 -0.09 -0.36 2.02 (-0.50) (-0.41) (-0.49) (0.40) (Log) Per Capita Private 0.15 -0.22 0.90 -9.01 Expenditure on Education & (1.77) (-0.96) (1.81) (-1.69) Health (Log) Per Capita Public 0.34 0.02 2.45 3.35 Expenditure of Education and (3.68) (0.09) (4.61) (0.62) Health Average Years of Education of -0.05 -0.14 -0.43 -3.14 Adult Females (-2.85) (-2.61) (-3.96) (-2.54) Urban/Rural Dummy 0.11 0.45 1.09 12.33 (1.24) (2.11) (2.20) (2.51) Poverty Head Count Ratio 0.004 -0.001 0.03 -0.07 (1.71) (-0.16) (1.95) (-0.44) Consumption Gini -0.02 -0.003 -0.16 0.26 (-2.78) (-0.10) (-3.29) (0.46) % of Population Muslim -0.004 -0.11 -0.02 -0.23 (-0.91) (-1.62) (-0.84) (-1.38) % of Population Scheduled 0.02 0.01 0.08 0.30 Caste/Scheduled Tribe (3.81) (1.36) (3.32) (1.31) N 30 32 30 32 F-Statistic 25.54 10.11 28.65 6.48 R2 0.92 0.81 0.93 0.73 Adjusted R2 0.88 0.73 0.90 0.61 Residuals from the regression results (Model 1) indicate how far above or below its "predicted" level each region is; the change in residuals (1983-99) is a "difference in difference" statistic, i.e., it indicates whether the gap between actual and predicted mortality has grown (either positively or negatively) or narrowed over the last two decades. 36 Table 14a: Rural (log) Infant Mortality Rates, 1983 & 1999: Select States 1983 1999 Change in Actual Predicted* Residual Actual Predicted* Residual Residuals, (%) (%) 1983-99 Himachal 4.98 4.90 -8.6 4.15 3.86 -29.8 -21.2 Pradesh Kerala 4.03 4.01 -1.4 2.64 3.00 36.6 38.0 Madhya 5.06 4.95 -11.3 4.54 4.44 -10.6 0.7 Pradesh Uttar 4.93 4.73 -21.1 4.47 4.26 -21.1 0.0 Pradesh West 4.63 4.66 2.7 3.99 4.30 30.6 27.9 Bengal Table 14b: Urban (log) Infant Mortality Rates, 1983 & 1999: Select States 1983 1999 Change Actual Predicted* Residual Actual Predicted* Residual in (%) (%) Residual, 1983-99 Himachal 4.14 4.26 11.4 3.63 3.31 -31.6 -43.0 Pradesh Kerala 3.89 3.84 -5.0 2.64 3.26 62.2 67.2 Madhya 4.65 4.47 -18.8 3.99 3.78 -20.7 -1.9 Pradesh Uttar 4.39 4.31 -8.9 4.17 3.81 -36.7 -27.8 Pradesh West 4.08 4.11 3.1 3.61 3.7 9.0 5.9 Bengal Note: * Predicted values (and therefore residuals) are estimated using Model 1 (Table 13). 37 6.1.1.1: Results (a) Regression Results (1) Education: Higher levels of educational achievement among adult females are consistently and strongly associated with lower infant mortality rates in both 1983 and 1999. Moreover, the effect of higher female adult education is more pronounced in 1999 than in 1983 ­ suggesting a change in the relative importance of various determinants of infant mortality, including "technology effects". (2) Expenditure Variables: Per capita consumption levels, and private and public per capita expenditure on education and health, are not important determinants of infant mortality rates. In fact, higher public expenditure is associated with higher infant mortality rates. (3) Exogenous Variables: Urban status in 1999 (after controlling for all other factors) was strongly associated with higher infant mortality rates, and, to a smaller degree, in 1983. The significance and importance of each of the other exogenous parameters changes a great deal over the 1983-1999 period. In 1983, the poverty head count ratio was significantly associated with higher infant mortality rates; by 1999, this effect had disappeared; conversely, higher consumption inequality in 1983 was associated with lower infant mortality rates, but not so in 1999. Finally, the population shares of Muslims and SC/STs show very different patterns and impacts. Infant mortality rates among the SC/ST group were higher than average in 1983, but statistically indistinguishable from the average in 1999. In contrast, infant mortality rates among Muslims have remained lower than average, but the magnitude of the "Muslim" effect has increased (while remaining statistically insignificant) between 1983 and 1999. (b) Residuals An analysis of residuals from Model 1 (but excluding the "exogenous parameters" discussed above; see Table 13) paints a more dynamic picture of these states' relative performance on infant mortality. Comparing rural areas in 1983, only West Bengal registered a slightly lower-than-expected infant mortality rate (i.e., a positive residual), and even Kerala did worse than expected. Sixteen years later, Kerala and West Bengal stood out as outstanding performers, with residuals in the range of 30-36 percent. Uttar Pradesh and Madhya Pradesh saw little change in their residuals over this period, and 38 both states remained well below potential. Himachal Pradesh registered a very large (- 30 percent) negative residual in 1999, and the largest (negative) "change in residuals" for the period. A similar if more dramatic story describes the experience of urban areas. Here Kerala, which was 5 percentage points below potential in 1983, leapfrogged over the other states in 1999 to register a 62 percentage point residual, and a 67 percentage point change in residuals. Mountainous Himachal Pradesh, which did better than expected in 1983, saw a massive negative gap emerging by 1999. Poor infrastructure ­ exacerbated by geographical factors that impact access to healthcare ­ may explain the state's unimpressive record. This is in sharp contrast to its recent record on education, and particularly on eradicating illiteracy. Madhya Pradesh and Uttar Pradesh do far worse than expected in both 1983 and 1999, but while MP sees little or no change in its residual, UP experiences a very large negative change. West Bengal, in comparison, does better than expected in both years. 39 6.1.2: Changes in Infant Mortality, 1983-99 If the parameters of Equation 1 remain stable, then broadly the same estimates (or elasticities) should be recovered from either a levels or a changes analysis. For example, private income "should" have a significant impact; public spending should continue to be insignificant; larger improvements in female adult education should be associated with larger declines in infant mortality over the period, etc. Unlike with indicators of education, changes in infant mortality rates, particularly at high initial levels, tend to be determined largely by changes in technology, increased access to medical facilities, and by parental education. A "class" variable was therefore created, placing states (and urban/rural areas within states) into three distinct categories based on initial levels of infant mortality. Class 1 includes those states/regions that were below the 20th percentile (i.e., the lowest, or best, levels) in terms of initial infant mortality (or logistic index of initial levels); Class 2 encompasses the middle levels, i.e., 20th to 60th percentiles, and Class 3 the highest (>60th percentile) levels. Our model of changes in infant mortality looks at interaction between initial conditions and three independent variables: changes in per capita private expenditure on health and education; changes in per capita public expenditures on health and education; and changes in educational attainment among adult females (i.e., ages 18-40), which is taken as a proxy for mothers' education. The model allows for differential impacts of these three variables on infant mortality reduction among each "Class" of states/regions, thereby allowing for a more detailed analysis of determinants. Its specifications are as follows: InfantMortality83-99 (Equation 3) ,Class= 1,Class( * X1) + 2,Class( * X 2) + 3,Class( * X 3) Where represents the initial "class" (percentile range) of infant mortality rates of a region/state, X1 is the change in (log) private per capita expenditure on health and education, X2 is the change in (log) public per capita expenditure on health and education, and X3 is the change in the average years of education of adult females. Table 15 contains the results of this regression. 40 Table 15: Determinants of Changes in Infant Mortality, 1983-99 Independent Variables Initial Class* Coefficient Change in Private Expenditure on 1 -1.75 Education and Health (-6.04) 2 -0.87 (-2.85) 3 -0.49 (-2.42) Change in Public Expenditure on 1 0.36 Education and Health (1.70) 2 0.22 (0.80) 3 -0.09 (-0.32) Change in Adult Female 1 39.52 Education (1.83) 2 1.94 (0.12) 3 -21.85 (-1.77) N 30 F-Statistic 24.23 R2 0.91 Adjusted R2 0.87 Notes: * "Class" represents the percentile range, in 1983, in which a region is placed in terms of infant mortality, where 1 represents percentiles 0-20, 2 represents percentiles 20-60, and 3 represents percentiles 60-100. Thus, a region falling within Class 1 has a lower infant mortality rate than one that is placed in Class 2. See text for details 6.1.2.1: Results (a) Regression Results Very broadly, the results of this model agree with those obtained from the levels analysis; this strongly suggests that the chosen model and its determinants are robust. We continue to find increases in public expenditure on education and health are not strongly correlated with improvements in infant mortality. In fact, increased expenditure is negatively related with infant mortality reduction in areas with lower initial levels of mortality. Increases in private expenditure on education and health, in contrast to the levels results, are positively correlated with improved outcomes; this effect is stronger in areas with lower initial infant mortality levels than it is in higher-mortality areas. Increases in per capita consumption/income (not included in the final model) do not appear to be important determinants regardless of initial infant mortality levels. 41 Very significantly, and confirming the earlier findings, an improvement in educational attainment among adult women is found, particularly for high infant mortality areas, to be a very strong predictor of improved outcomes; in lower initial level areas, though, this effect quickly tapers off or turns negative. This indicates the presence of a "threshold effect": beyond a threshold level of infant mortality, mother's education is vitally important; once this level is breached, though, the impact of education decreases. Putting these findings together, we conclude that: (1) public expenditure is largely ineffective in bringing about improvements in infant mortality outcomes; (2) mothers' education is an important factor at high levels of infant mortality, where even small increases in educational attainment can have a large effect; and (3) private expenditures on education and health are strongly and positively associated with improvements only at lower levels of mortality. (b) Residuals Analysis How did our study states perform in terms of changes in infant mortality? In general, the results of a residuals analysis (Table 16) closely match the results obtained from the levels analysis above. Three of the five states (Himachal Pradesh, UP and MP) display similar trends; however, a changes analysis for Kerala and West Bengal reveals somewhat different trends. According to the levels analysis, urban Kerala (relatively) outperformed rural Kerala in terms of "change in residuals" statistics. The multi-linear regression, though, indicates the opposite. Similarly urban West Bengal does better than expected in a levels analysis, but slightly worse than expected when residuals are computed from the changes regression. 42 Table 16: Change in Infant Mortality Rates (log %), 1983-99: Select States Rural Urban Actual Predicted* Residua Actual Predicted* Residual l (%) (%) Himacha -82.6 -99.7 -17.0 -51.6 -73.5 -21.9 l Pradesh Kerala -138.6 -106.8 31.9 -125.3 -117.5 7.8 Madhya -51.9 -50.5 1.4 66.5 66.2 0.3 Pradesh Uttar -46.9 -52.4 -5.5 -22.0 -42.3 -20.3 Pradesh West -64.6 -43.7 20.9 -46.7 -47.3 -0.6 Bengal Note: * Predicted values (and therefore residuals) are estimated using the Model in Table 15. 43 Tables 17a and 17b explains the importance of female education to the achievement, and reduction, in infant mortality in different states of India. Several of the expected determinants are introduced into a simple model relating the (log) level of infant mortality to its assumed determinants. Mother's education (adult females in household aged 18- 40) by itself explains about 60 percent of the variation in (log) infant mortality in both 1983 and 1999 for the 15 different states in India for which data are available.10 Other presumed determinants are introduced on an individual basis into this simple model. Very few of these variables are significant, and when they are e.g. %SC/STs , the impact is very small, as well as the additional explanatory power obtained by inclusion of the model. One noteworthy result (and confirming the consistent result documented throughout this paper) is that state expenditures on health and education have a "perverse" effect on levels of infant mortality i.e. higher expenditures in states which have higher mortality levels. One explanation for this perversity could be that higher state expenditures are allocated to areas of worse health conditions in order to help improve the well being of the poor. While appealing, this hypothesis is not supported by the data relating the change in infant mortality to the change in expenditures. 10Note that urban and rural areas within a state are considered to be "separate" states. 44 Table 17a: Two Variable Level Regressions for Infant Mortality 1983 1999 Second Variable Mother's Second Adjusted Mother's Second Adjusted Education Variable R2 Education Variable R2 Mother's -0.11 - 0.62 -0.15 - 0.60 Education (-7.13) (-6.92) Consumption -0.13 0.24 0.64 -0.15 -0.06 0.59 (log per capita) (-6.77) (1.62) (-5.13) (-0.34) Private Exp on -0.12 0.07 0.61 -0.15 -0.04 0.59 Edu + Health (-5.74) (0.60) (-3.92) (-0.21) (log per capita) State Exp on -0.12 0.36 0.68 -0.16 0.22 0.62 Edu+Health (log (-7.99) (2.64) (-7.20) (1.48) per capita) Poverty Head -0.11 -0.002 0.61 -0.15 0.002 0.59 Count Ratio (%) (-7.06) (-0.71) (-6.23) (0.56) Gini -0.11 0.003 0.60 -0.16 0.005 0.59 (Consumption) (-5.73) (0.32) (-4.33) (0.26) % Muslim -0.11 -0.007 0.70 -0.14 -0.01 0.65 (-7.71) (-3.06) (-6.62) (-2.29) % SC/ST -0.07 0.02 0.77 -0.09 0.02 0.66 (-4.32) (4.67) (-2.99) (2.42) Note: Each row represents a distinct model, with a constant idependent variable (infant mortality) and one fixed independent variable (years of education of adult (age 18-40) females, i.e., "Mother's Education"). Each successive model (except for the first) adds a second independent variable to the equation, i.e., any one equation contains just two independent variables. 45 Table 17b: Change in Infant Mortality: Model with Only Initial Conditions Class Coefficient (t-Statistic) Initial (1983) Years of 8.49 Education of Female Adults (4.98) (18-40) Change (1983-99) in Years of 1 25.53 Education of Female Adults (1.80) 2 21.31 (2.64) 3 25.32 (3.35) N 30 F-Statistic 34.9 R2 0.84 Adjusted R2 0.82 46 6.2: Education Achievements A very significant result that emerges from our study of educational achievement is the presence of huge "catch-up" effects across states, within states, and across gender-, religion-, and class-lines. We look at a number of indicators of education but concentrate primarily on two: relative educational achievement (as a percent of potential), and years of education. In both cases, we look at two age groups, 5-18 years and 18-40 years. Available data on literacy and school attendance rates, as discussed elsewhere in this paper, are not very accurate indicators of educational achievement, and we therefore construct our two indicators from raw household survey (NSS 1983 and 1999) data. (See Section 4 for summary statistics on educational attainment.) 6.2.1: Determinants of Changes in Educational Achievement Unlike with the infant mortality models, we concentrate on the determinants of changes in educational attainment for the 5-18 age group. Several models, which progressively add a number of exogenous factors to the base model, are presented. The first model (Table 18) attempted to explain variations purely on the basis of urban-rural status. The results indicate the existence of a strong catch-up effect across rural-urban lines. Model 2 adds two "initial conditions" to the equation: initial (1983) years of education among adult females, and the initial female-male ratio of educational achievement for the 5-18 age group. Both variables are significant (and negative), again indicating the presence of strong catch-up effects. Note, however, that the urban dummy becomes insignificant in this model ­ indicating that the other two initial conditions dominate the urban-rural status. The final model leaves out this variable and adds three additional (expenditure- based) determinants. It is specified as follows: EducationYears83 -99= + 1X1 + 2X2 + 3X3 + 4X4 + 5X5 + (Equation 4) Where X1 represents the initial (1983) average years of education for adult females, X2 is the initial female-male ratio of educational achievement, X3 is the (log) growth in per- capita consumption, and X4 and X5 are, respectively, the (log) changes in per capita private and per capita state expenditures on education and health. Even with the addition of the three expenditure variables, initial education conditions continue to explain most of the variation (as high as 80 percent) in educational outcomes. In addition, and in sharp contrast to the results for infant mortality, private 47 growth rates are an important determinant of changes in educational attainment. State expenditure, once again, is shown to have a neutral-to-negative impact on outcomes; its coefficient in the regression tells us that states with higher public expenditures on health and education have, in general, shown a smaller improvement in educational outcomes. Private expenditure, in contrast to the infant mortality model, does not significantly affect outcomes. An analysis of residuals from this model (Table 19) is instructive. Himachal Pradesh and Kerala, both ahead (except for rural Himachal Pradesh) of the Indian average in 1983, see relatively modest improvements over the 1983-1999 period. Rural Kerala and urban Himachal Pradesh, in fact, do worse than expected. While rural Madhya Pradesh does relatively well, urban MP sees only limited growth, and West Bengal's performance is fairly close to expectations. Uttar Pradesh, as with infant mortality, does very poorly in comparison to its potential. In general, then, with the important exception of Uttar Pradesh, we find strong evidence of catch-up growth in educational achievement. This is best illustrated by slow growth in urban Kerala and relatively rapid growth in rural Madhya Pradesh. 48 Table 18: Determinants of (Percent) Changes in Education Years (5-18 Age Group), 1983-99 Coefficient (T-Statistic) Independent Variable Model 1 Model 2 Model 3 (Final) Constant Term 36.96 71.62 71.73 (12.84) (9.27) (8.90) Urban/Rural Dummy -24.94 -4.05 - (-6.13) (-0.89) Initial (1983) Years of Education, - -2.76 -4.18 Adult Females (-2.04) (-4.60) Initial (1983) Female-Male Ratio - -0.43 -0.36 of Educational Achievement, 5-18 (-3.10) (-2.80) Age Group Change in (log) Per Capita - - 0.17 Consumption (2.31) Change in (log) Per Capita Private - - -0.05 Spending on Education & Health (-0.87) Change in (log) Per Capita State - - -0.08 Spending on Education & Health (-1.84) N 36 36 32 F-Statistic 37.54 48.17 36.05 R2 0.52 0.82 0.87 Adjusted R2 0.51 0.80 0.85 Table 19: Change in Educational Achievements (5-18 Age Group), 1983-99: Select States Rural Urban Actual Predicted* Residual Actual Predicted* Residual (%) (%) Himachal 27.8 27.6 0.3 -2.7 1.9 -4.5 Pradesh Kerala 8.5 11.2 -2.8 5.7 2.5 3.2 Madhya Pradesh 52.4 45.5 6.9 10.5 15.9 -5.4 Uttar Pradesh 38.7 45.5 -6.7 14.1 18.3 -4.2 West Bengal 35.1 30.8 4.3 3.4 4.8 -1.4 Note: * Predicted values (and therefore residuals) are estimated using Model 3 (Table 18). 49 Section 7: Will the Millennium Development Goals be met? Having examined the determinants of health (infant mortality) and education outcomes, it is relevant to look at the "status" of each of our five states, in 1999/2000, with regard to the much-cited Millennium Development Goals (MDGs). We examine the feasibility, in particular, of meeting three of the published MDGs by the target year of 2015: infant mortality of one-third of the 1990 levels, universal primary education, and complete gender equality in educational attainment. 7.1: Infant Mortality Based on time trends in infant mortality decline over the 1980-2000 period, and extrapolating, we estimate the predicted levels of infant mortality in 2015 for each state and its sub-regions. This allows for a fair guess as to whether or not a state (or regions within states) will or will not achieve this particular goal or, alternatively, whether they will come close to meeting it. As Table 20 indicates, Kerala (and its sub-regions) will unambiguously reach its infant mortality targets, while HP as a whole, Madhya Pradesh as a whole, rural HP and urban MP are likely to get very close to this goal. All of the other regions are likely to fail in meeting their targets. 50 Table 20: Progress Towards Meeting Infant Mortality MDG Infant Mortality Rates Predicted Levels in 2015 1980 1990 2000 % of 1990 Level % of 1990 Will Level in Level MDG 2000 be Met? Himachal Rural 146 84 64 76.2 35 41.0 Close Pradesh Urban 63 42 38 90.5 26 61.9 No State 143 82 63 76.8 34 41.5 Close Kerala Rural 56 45 14 31.1 5 11.0 Yes Urban 49 42 14 33.3 6 13.0 Yes State 54 42 14 33.3 5 12.1 Yes Madhya Rural 158 142 94 66.2 64 44.8 No Pradesh Urban 105 84 54 64.3 33 39.0 Close State 150 133 88 66.2 59 44.4 Close Uttar Rural 139 104 87 83.7 61 58.9 No Pradesh Urban 81 76 65 85.5 55 72.5 No State 130 99 83 83.8 59 59.9 No West Rural 103 66 54 81.8 33 50.4 No Bengal Urban 59 41 37 90.2 26 63.6 No State 95 62 51 82.3 32 51.6 No 51 7.2: Universal Primary Education Himachal Pradesh and Kerala, and their rural and urban components, have already achieved a more ambitious target than universal primary education: on average, educational achievement in the 5-18 age group is more than 100 percent of what it should be11; the same is true for urban Madhya Pradesh. All of the other states and sub- state regions, and especially UP, MP, and rural UP and MP, are some distance away from meeting this goal (but are already close to achieving universal primary education). Looking at trends over the 1990-2000 period, however, these areas are likely to meet this important goal by 2015. Table 21: Progress Towards Universal Education Educational Attainment of Children (5-18), (%) 1990 1999 Shortfall in Will MDG be 1999 (%) Met? Himachal Rural 96.6 105.9 - Yes Pradesh Urban 106.9 105.9 - Yes State 97.4 105.9 - Yes Kerala Rural 114.2 111.9 - Yes Urban 111.8 112.4 - Yes State 113.8 112.0 - Yes Madhya Rural 61.5 77.6 22.4 Yes Pradesh Urban 97.4 100.4 - Yes State 68.3 82.6 17.4 Yes Uttar Rural 62.8 79.0 21.0 Yes Pradesh Urban 83.3 87.3 12.7 Yes State 66.1 80.7 19.3 Yes West Bengal Rural 70.6 87.7 12.3 Yes Urban 94.7 94.3 5.7 Yes State 75.2 88.9 11.1 Yes 11Please see a detailed description elsewhere in this paper on the construction of this measure of educational achievement. 52 7.3: Gender Equality in Education As discussed elsewhere in this paper, one of the most important results of this research project is India's (and our states') remarkable progress towards gender equality in educational achievements. Over the last twenty years, there has been a rapid convergence in male and female educational attainment. This is especially true of Himachal Pradesh and Kerala. Additionally, urban MP, UP and West Bengal are within 6 percentage points of a 100 percent gender equality on this measure. Uttar Pradesh and Madhya Pradesh as a whole, and their rural regions, lag behind other states and regions. Again, though, any extrapolation of past trends certainly suggests that gender equality will be achieved soon, and almost definitely so by 2015. Table 22: Progress Towards Gender Equality in Education Female-Male Ratio of Educational Attainment of Children (5-18), (%) 1990 1999 Shortfall in Will MDG be 1999 (%) Met? Himachal Rural 89.1 100.2 - Yes Pradesh Urban 96.7 100.2 - Yes State 89.5 100.2 - Yes Kerala Rural 95.0 100.2 - Yes Urban 101.3 102.0 - Yes State 95.9 100.7 - Yes Madhya Rural 65.8 83.6 16.4 Yes Pradesh Urban 95.9 96.7 3.3 Yes State 72.7 87.1 12.9 Yes Uttar Rural 61.3 78.4 21.6 Yes Pradesh Urban 92.2 94.3 5.7 Yes State 66.7 81.9 18.1 Yes West Bengal Rural 85.9 91.2 8.8 Yes Urban 96.1 94.5 5.5 Yes State 87.8 91.9 8.1 Yes 53 Section 8: Conclusions and Policy Implications This paper has examined several competing hypotheses of the determinants of living standards in different states of India. Achievements in two sectors ­ health and education ­ were examined for the twenty-year period 1980 to 2000. Some of the important findings are summarized below. The determinants of improvement in living standards can broadly be classified into the following four categories: initial conditions or history; inputs on the part of government; private inputs; and inputs by quasi-government organizations (QGOs). What has been attempted in this paper is a quantification of the contribution of each of these inputs. 8.1: Methodology All living standard indicators have a natural ceiling or floor, beyond which improvements are difficult, and only maintenance is possible. For example, primary school enrollment or completion is "censored" at 100 percent of the population, and infant mortality rates reach a floor close to 5 per 1000 births. It is also the case that where a state was initially has an impact on the rate of change that can be observed. For example, because of availability of technology (drugs) it is not that investment intensive a process to get infant mortality deaths to decline from levels like 140 or 160. Consequently, the same proportionate reduction from 40 to 20 is likely to involve a lot more expenditures and effort than a decline from 160 to 80. To partially account for such estimation problems, it was suggested that a logistic formulation would be better able to account for the underlying non-linear reality. Consequently, all regression results for infant mortality were estimated both with a log level and log change formulation. None of the results, however, are affected by the change in specification. 8.2: Initial Conditions As expected, initial conditions play a strong explanatory role in explaining changes in living standards. For infant mortality changes, almost the entire change between 1983 and 1999 is explained by knowledge of what the IMR was in 1983. For education, the role of initial conditions is even greater ­ almost the entire magnitude of improvement between 1983 and 1999 is explained by knowledge of what the level of educational attainment was in 1983. This applies to all the four different indicators of education. 54 8.3: What matters for improvement in living standards Different states in India have followed different policies towards improvement in living standards and hence it is important to determine what worked and what did not. Our analysis was conducted with both a partial (two variable correlations or regression) and complete (several determinants) model. Some conclusions follow. 1. Does Initial Inequality matter? No The conventional wisdom states that initial inequality is harmful for improvement of living standards, i.e. a more equal economy is likely to also lead to "easier" progress in living standards. The opposite conclusion is suggested by both the partial and complete model. Correlation between inequality (Gini) and (log) infant mortality in 1983 is ­0.55 which marginally increases to ­0.6 in 1999. The two most unequal urban areas in the country in 1983 were those located in Kerala and Himachal Pradesh, with (per capita expenditure) Gini coefficients of 40.5 and 44.7 respectively. These two urban areas are the best performers in terms of infant mortality improvement (Kerala) and schooling (Himachal Pradesh). There are some intuitive reasons to believe that there is a negative association between initial inequality and lower levels and higher improvement in infant mortality. The answer lies more in the realm of aggregation and statistics than economics. If incomes are unequally distributed, then a "smaller" population can achieve bigger gains in health care and hence, or may, show lower average levels of infant mortality or larger gains. This phenomenon is even more pronounced if there are "threshold" effects, as is likely the case with infant mortality. Inequality ceases to have any partial effect in 1999, as compared to a negative (high inequality leading to low mortality) and significant effect in 1983. 2. Do Regions with Higher Poverty have Lower Living Standards? Earlier Yes, Not Now The analysis suggests that higher poverty does not automatically mean lower living standards. There is only a weak, but positive, partial association between poverty (head count ratio) and infant mortality in 1983, but this correlation increases to 0.32 in 1999 from only 0.11 in 1983. (There is also a very weak negative correlation between 55 inequality and poverty ­ a minus 0.23 in 1983 declining to -.09 in 1999). The correlation with education is stronger, and this has increased over time i.e. today, being poor does mean less educational achievement. The complete regression model, (possibly because it contains the level of average per capita expenditure as an explanatory variable), shows that in 1999, unlike 1983, poverty levels have no independent effect on infant mortality. 3. Do Regions with Higher SC/STs have Lower Living Standards? Earlier Yes, Not Now The proportion of SC/STs in a state is positively correlated with higher poverty, so the effects of the proportion of SC/STs in a population are similar to the effects of higher poverty. However, it is revealing to note that the partial effect of SC/STs on infant mortality decline becomes insignificant in 1999, compared to a significantly positive effect in 1983. 4. The Role of State Expenditures in Achieving Improvements in Living Standards The level of state expenditures may have an insignificant and/or differing effect with living standards, but it is generally expected that increases in such real per capita expenditures will have a positive effect, i.e. such expenditures are productive if they lead to improvements. One of the more consistent findings reached by estimation of several models, several specifications, and several dependent variables is that state spending has often a statistically significant and negative effect on improvement in living standards. This is especially noteworthy because expenditures incurred by the households usually have the opposite, positive effect. The juxtaposition of these two results means a strong conclusion that state spending is particularly ineffective in improving living standards in either health or education. It is also observed that the negative effect of state spending on improvement in infant mortality is particularly acute in the states that already have relatively low levels of infant mortality. 56 5. The Role of Private Expenditures in Achieving Improvements in Living Standards Private expenditures generally have a positive effect on improvement in living standards. This is particularly true for improvements in health (taking infant mortality rates as proxy), and especially so at lower initial levels of infant mortality. In other words, private expenditure on health and education is especially beneficial in those states (or areas within states) where infant mortality rates have already crossed a "threshold" level. Our results indicate that private spending is not significant in explaining changes in educational achievement within the 5-18 year age group. This, however, may be due to a host of other factors (including directives laid down in the Indian constitution calling for universal primary education) that are not related to expenditure. 6. What Explains Improvements in Infant Mortality? Our results indicate the presence of a significant "threshold effect" in determining changes in infant mortality rates. We find that above a certain initial infant mortality rate (this varies by urban and rural status), improvements in the education of adult females (i.e., females in the 18-40 age group, used as a proxy for "mothers") is very significantly related to improvements in infant mortality. Below the threshold level, a range of other factors, including private expenditures on education and health, become much more important. 7. What Explains Improvements in Education? We find a significant "catch-up" effect at work which tends to dominate changes in educational attainment, particularly in the 5-18 age group. In other words, those states or regions that were relatively worse off in this regard in 1983 have seen the biggest improvements over the 1983-1999 period. 8: On Achieving the Millennium Development Goals Our sample states are well on track to achieving two of the millennium development goals: on universal primary education, and on gender equality in educational achievement. In fact, Kerala and Himachal Pradesh have already achieved both of these goals, and the remaining states are likely to do so well before 2015. In terms of target infant mortality rates, Kerala, again, has already reached its target level, and is on track to reaching the natural floor level of 5 by 2015. Other states, however, lag well 57 behind in this respect. Four "regions" ­ Himachal Pradesh, Madhya Pradesh, rural Himachal Pradesh, and urban Madhya Pradesh ­ are likely to come close to meeting their targets by 2015. All other "regions" are unlikely to do so. 9: Inter-State Comparisons Himachal Pradesh has an average record on education, and a much below par record in terms of changes in infant mortality. Geographical and other factors have likely limited the spread of medical care facilities, resulting in only slow progress on health delivery. Kerala was already a "rich" state in terms of living standards in 1960 and 1980, but has made remarkable progress on infant mortality reduction. On education, its progress is "average" for variables like change in mean years of education of 5-14 year olds. Madhya Pradesh: Like UP, MP also does badly in terms of improvement in infant mortality. But on education it is one of the best performing states, especially for improvements in educational achievement in rural areas. Uttar Pradesh shows large negative residuals on both infant mortality and education i.e. in terms of performance, it consistently lags the average Indian state. This "performance" is particularly worse for the urban areas. Note that this finding has to do with achievement relative to "expectations" given the level of income, initial conditions, female education etc. West Bengal: This state posts an exceptional performance for improvement in infant mortality in rural areas. For urban areas, the performance is only slightly better than average. In education, West Bengal is an average performer, registering marginally better than average achievement in rural areas, and slightly worse than average in urban areas. 58 References: Behar, Amitabh and Yogesh Kumar (2002). Decentralization in Madhya Pradesh, India: From Panchayati Raj to Gram Swaraj (1995 to 2001). London: Overseas Development Institute, Working Paper 170. Bhalla, Surjit S. and Indermit S. Gill (1990). Social Expenditure Policies and Welfare Achievement in Developing Countries. Mimeo. Bhalla, Surjit S. (2003). Defying the Kuznets U: Inequality, Poverty and Growth in India. Prepared as a background paper for Human Development Report 2003/2004. De, A., C. Noronha and M. Samson (2000). Primary Education in Himachal Pradesh: Examining a Success Story. New Delhi: National Institute of Educational Planning and Administration and Ministry of Human Resource Development, Government of India: Year 2000 Assessment: Education for All. Institute of Development Studies, University of Sussex (2001) "Bringing Citizen Voice and Client Focus into Service Delivery" Case Study Series: People's Planning Campaign, Kerala, India. Jha, Shikha (2002). Strengthening Local Governments: Rural Fiscal Decentralization in India. Economic and Political Weekly, Vol. XXXVII, No.26, June 29, 2002, p.2611. Kingdon, Geeta Gandhi and Mohd Muzammil (2001). A Political Economy of Education in India ­ I: The Case of UP. Economic and Political Weekly, August 11 2001, p.3052. Mahal, Ajay, Vivek Srivastava and Deepak Sanan (2000). Decentralization and Public Sector Delivery of Health and Education Services: The Indian Experience. Bonn: University of Bonn ­ Center for Development Research (ZEF), Discussion Papers on Development Policy No.20. Mehrortra, Santosh (2001). Basic Social Services for All? Ensuring Accountability through Deep Democratic Decentralization. United Nations Development Programme Human Development Report Office: Background Paper for Human Development Report 2002. Nayar, K.R. (2001). "Politics of Decentralization: Lessons from Kerala." In Imrana Qadeer, Kasturi Sen and K.R. Nayar (ed.), Public Health and the Poverty of Reforms (2001). New Delhi: Sage Publications. Sadanandan, Rajeev (2001). Government Health Services in Kerala: Who Benefits? Economic and Political Weekly, August 11 2001, p.3071. 59