76750

         Data and Dogma: The Great Indian
                  Poverty Debate


                                     Angus Deaton • Valerie Kozel

What happened to poverty in India in the 1990s has been fiercely debated, both politically
and statistically. The debate has run parallel to the wider debate about globalization and
poverty in the 1990s and is also an important part of that debate. The economic reforms of
the early 1990s in India were followed by rates of economic growth that were high by
historical standards. The effects on poverty remain controversial, however. The official
numbers published by the government of India, showing an acceleration in the rate of pov-
erty reduction from 36 percent of the population in 1993/94 to 26 percent in 1999/
2000, have been challenged for showing both too little and too much poverty reduction.
The various claims have often been frankly political, but there are also many important sta-
tistical issues. The debate, reviewed in this article, provides an excellent example of how
politics and statistics interact in an important, largely domestic debate. Although there is
no consensus on what happened to poverty in India in the 1990s, there is good evidence
both that poverty fell and that the official estimates of poverty reduction are too optimistic,
particularly for rural India. The issues covered in this article, although concerned with the
measurement of poverty in India, have wide international relevance—discrepancies
between surveys and national accounts, the effects of questionnaire design, reporting peri-
ods, survey nonresponse, repair of imperfect data, choice of poverty lines, and interplay
between statistics and politics.


Hundreds of millions of Indians are poor by national and international standards.
Indian politics are dominated by discussions of poverty, and measures of poverty
rightly attract a great deal of attention and debate. In the second half of the 1990s
India’s gross domestic product (GDP) grew rapidly by historical standards, and many
commentators have associated this acceleration with the economic reforms that
began in the 1990s. Yet the reforms themselves, and the limited opening of the
Indian economy that they involved, remain controversial, as does their effect on
poverty.

© The Author 2005. Published by Oxford University Press on behalf of the International Bank for Reconstruction and
Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oupjournals.org.
doi:10.1093/wbro/lki009                                                                                   20:177–199
   This debate is far from unique to India. The worldwide controversy about global-
ization and its effects on poverty and inequality has followed much the same lines as
the internal debate in India. India accounts for about 20 percent of the global count
of those living on less than $1 a day, so what happens in India is not only a reflection
of the worldwide trend but one of its major determinants.



A Brief Introduction to the Indian Poverty Monitoring System
India’s official poverty estimates are based on the regular consumer expenditure sur-
veys conducted by the National Sample Survey Organization (NSSO). These surveys,
pioneered by P. C. Mahalanobis in the 1940s and 1950s (Mahalanobis and Sen
1954), were the world’s first system of household surveys to apply the principles of
random sampling established in the 1920s and 1930s.
   The NSSO conducts large consumer expenditure surveys every five or six years and
smaller surveys annually. In recent years even small surveys have collected some
data on consumer expenditures. But to calculate official poverty statistics, the Plan-
ning Commission uses only the larger surveys that focus on consumer expenditures
on the grounds that the larger surveys are required to estimate poverty accurately
for each state because those estimates are the basis for various transfers from the
central government to the states. In recent years large surveys were conducted in
1983 (Round 38), 1987/88 (Round 43), 1993/94 (Round 50), and most recently,
1999/2000 (Round 55). (The next, Round 61, for 2004–2005, was in the field at
the time of writing.) For each of these surveys the Planning Commission has pub-
lished estimates of the number and proportion of people in poverty, by state and sector.
   These official poverty estimates count the number of people living in households
with monthly per capita total expenditure below a poverty line specific to the state
and sector (urban or rural). The original official state-specific poverty lines, based on
state differences in price levels, come from the report of an expert group (India,
Expert Group on Estimation of Proportion and Number of Poor 1993), which
derived them from earlier academic studies that had compared prices across states.
The poverty lines are updated periodically using a system of state-by-state price
indexes, which are estimated separately for rural households (the consumer price
index for agricultural laborers) and urban households (the consumer price index for
industrial workers). There are no predetermined all-India urban and rural poverty
lines. Instead, rural and urban poverty counts for each state are aggregated to get
urban and rural totals. All-India urban and rural poverty lines are then set so that
when they are applied to all urban or rural households without differentiation by
state, the total number of urban and rural poor people matches the sum of the state
counts. The poverty data from 1983 onward, which are based on these procedures,
are the subject of the current debate.

178                                        The World Bank Research Observer, vol. 20, no. 2 (Fall 2005)
Conflicts between National Accounts and Sample Surveys
Estimates of mean consumption are generated by both the national accounts statis-
tics and the NSSO from their regular surveys of consumer expenditures. The two sets
of estimates can be used to cross-check and validate one another, both for total con-
sumer expenditure and for individual commodities or groups of commodities, such
as grains, clothing, or services. There is a long tradition of such comparative work in
India. More controversially, before the 1990s, the Planning Commission used the
national accounts estimate of consumption as a “control� total for the surveys in
estimating poverty. Thus, for example, if the ratio of the national accounts estimate
to the survey estimate of mean consumption was greater than one, the Planning
Commission would multiply the total expenditure of each household by that ratio
before calculating the number of people living in households below the poverty line.
   This procedure is practiced in many countries, particularly in Latin America,
where income estimates from surveys are typically much smaller than those from
national accounts. India’s abandonment of such scaling up (for reasons discussed
later) has been the subject of considerable controversy. The proponents of such pro-
cedures believe that national accounts data are superior to survey data, so that it is
appropriate to use the additional information they contain to adjust the surveys.
Bhalla (2002), like Sala-i-Martin (2002), uses a variant of the method to estimate
global poverty (see Deaton 2005 for a discussion). Opponents of such procedures
question the automatic assumption of superiority for the national accounts data and
note that scaling up the surveys bypasses a central question in the debate: does the
growth measured in the national accounts show up in improvements in the living
standards of the poor?
   In India during the 1990s the national accounts estimates of mean consumption
grew much more rapidly than did the survey estimates. Scaling up would thus have
shown a more rapid reduction in poverty in the 1990s than shown by the survey-
based poverty estimates. In consequence, those who believe that the economic
growth following the reforms has been associated with large-scale poverty reduction
have tended to argue that the national accounts are right and the surveys wrong.
Antireformers and skeptics have argued for the surveys.
   Early comparisons of surveys and national accounts in India were carried out by
Mukherjee and Chatterjhee (1974) and by Srinivasan and others (1974). Using sur-
vey and national accounts data from the 1950s and 1960s, they examined the
match between the two estimates of total consumption and its distribution. For the
decade up to 1963/64 Mukherjee and Chatterjee write that agreement between the
revised series for national accounts consumption and the survey estimates is close,
although they note that the survey estimates are systematically and (on average)
increasingly below the national accounts estimates in the period up to the end of the
1960s. They also note discrepancies in the distribution of consumption over

Angus Deaton and Valerie Kozel                                                     179
commodities, with the surveys recording a higher share of food in the budget than
do the national accounts. Srinivasan and colleagues’ analysis is broadly consistent
with that of Mukherjee and Chatterjee, though they find that the surveys are lower
than the national accounts estimates from an earlier date. They also note that the
distribution of consumption is broadly similar in the two sources.
   If the early comparisons of national accounts and survey estimates of mean con-
sumption were relatively reassuring, more recent comparisons are anything but.
The gap between the two estimates has continued to widen, reaching levels that
would have been viewed with horror by the early researchers. Depending on the set
of adjustments made, the survey estimate of consumption is currently around two-
thirds of the national accounts estimate and has been falling steadily since the late
1960s, by 5–10 percentage points a decade.
   This differential rate of growth in consumption estimates is far from unique to
India. There appears to be a similar discrepancy between survey and national
accounts estimates of the growth rate of consumption for the world as a whole
(Deaton 2005). To take a specific example at a very different level of development,
the differential rate of growth in the United States is very similar to that in India
(Triplett 1997; Garner and others 2003). Although there are almost certainly errors
in both sets of estimates, the view of what is happening to poverty depends a good
deal on how much of the discrepancy is attributed to each.
   For many economists, who are well versed in the concepts of national income
accounting but much less so in survey practice, the automatic reaction is to trust the
national accounts over the surveys. That there is little basis for such a judgment was
carefully argued by Minhas (1988) in an article that should be read by anyone con-
cerned with the issue of the differences between national accounts and survey data.
Minhas’s work was central to the Planning Commission’s decision to abandon its
practice of scaling up the survey results to match the national accounts. Minhas lays
out the issues that have dominated the contemporary debate: the differential defini-
tion and coverage of consumption in the surveys and the national accounts, differ-
ences in timing, and the heavy reliance in national accounting practice on various
“rates and ratios� that link observable but irrelevant quantities to the unobservable
but relevant ones. These ratios are derived from surveys (for example, surveys that
link earnings in services sectors to value added in the sectors), but they are
frequently many years—often decades—out of date.
   The use of outdated rates and ratios in a growing economy experiencing struc-
tural development will typically lead to systematic trend errors in the accounts. Con-
sider the netting out of intermediate production from value added, which is
frequently done using a fixed ratio. Because the degree of intermediation tends to
grow as the economy becomes more complex and more monetized, the rate of
growth of GDP and of consumption will be systematically overstated in a growing
economy. Cooking oil, particularly vanaspati, provides a good example for India. The

180                                       The World Bank Research Observer, vol. 20, no. 2 (Fall 2005)
national accounts estimate consumption of vanaspati as total production less
imports plus exports, less consumption by government or business. In an economy
in which all vanaspati is used for household cooking, this gives the right answer. But
as the economy grows, consumers eat more meals out, so that an increasing fraction
of vanaspati is used by commercial food suppliers, restaurants, hotels, and street
vendors. Consumer spending on these services is derived from (fairly shaky) data on
the gross output of the services sector, adjusted to a value-added basis by deducting
the value of intermediate inputs, including vanaspati. At best, this adjustment is
done using one of the rates and ratios, which means progressive and increasing
overstatement if intermediation increases with income and if rates and ratios are
infrequently adjusted. In the case of vanaspati in India, no adjustment is made at all,
so that all vanaspati used in restaurants is counted twice, helping overstate the rate
of growth of consumption and GDP and to increase the ratio of national accounts to
survey consumption (see Kulshrestha and Kar 2005).
   Minhas (1988, p. 14) notes that “many discussions of sampling errors seem to
imply as if only the NSS estimates suffer from those errors. This is a gross misconcep-
tion.� He warns against adjustments that assume that only the survey estimates are
at fault:
      It is indeed hazardous to carry out pro-rata adjustment in the observed size
      distribution of consumer expenditure in a particular NSS [National Sample
      Survey] round by multiplying it with a scalar derived from the ratio
      between the NAS [National Accounts Statistics] estimates of aggregate private
      consumption for the nearest financial year and the total NSS consumer
      expenditure available from that particular round of household budget survey.
      This kind of mindless tinkering with the NSS size distribution of consumer
      expenditure, as practiced by the Planning Commission in the Seventh Five
      Year Plan documents, does not seem permissible either in theory or in light
      of known facts. (p. 37)
With consumption growing more than 1 percent a year faster in the national
accounts estimates than in the survey estimates, application of the pro rata adjust-
ment makes an enormous difference to the trend in measured Indian poverty. It is
unfortunate that so much of the current debate over this issue should have been so
little informed by what Minhas wrote 15 years ago.
    Kulshrestha and Kar (2005) and Sundaram and Tendulkar (2003b) provide a
more contemporary perspective on the discrepancy. Kulshrestha and Kar were pri-
marily responsible for the production of the national accounts in the 1990s, so their
views on the accuracy of the consumption estimates carry considerable weight.
They document the growing discrepancy between the two sources, from 5 percent
in 1957/58 to more than 38 percent in 1993/94, and note that the discrepancy is
larger and increasing faster for nonfood items than for food.

Angus Deaton and Valerie Kozel                                                        181
   These authors explore food items in more detail because there is often enough
additional information to make an informed judgment about the likely balance of
accuracy. With a few exceptions, their findings are similar to those of Minhas (who
comes at the issue from the survey side): when there is a discrepancy, the national
accounts estimates are typically less plausible and more likely to be in error. They
determine that the discrepancy for food and tobacco can be attributed to a few spe-
cific commodities (fruit, milk products, chicken, eggs, fish, minor cereals and their
products, vanaspati, oilseeds, and tobacco) and that for the major subgroups that
are important in poverty studies (major cereals, more commonly used pulses, edible
oils, liquid milk, and vegetables), the two estimates are relatively close. They con-
clude that nothing in their findings would make the survey data on household con-
sumption expenditure “unfit for measurement of poverty incidence� (p. 117).
Although this assessment may be too sanguine—nonfood items also play a role in
poverty measurement—this work establishes that there can be no automatic
presumption in favor of the national accounts.
   Sundaram and Tendulkar (2003b) report on the findings of a joint Central Statis-
tical Organization–NSSO exercise on cross-validation of the two sets of estimates.
They draw attention to the fluidity of the national accounts estimates, noting that
revisions for some categories are often so large as to cast doubt on the estimates in
general. This is closely related to Minhas’s concern with the outdated rates and
ratios. When the Central Statistical Office abandons a long-used ratio and new sur-
vey or other information is collected, information based on actual data paints a very
different picture from that based on the long-used approximation. Such revisions,
though welcome, do little to improve the large number of items still held hostage to
the inaccuracy of old and aging ratios. Sundaram and Tendulkar also argue that
survey data are to be preferred because they measure living standards directly, as
opposed to national accounts statistics, which derive consumption as a residual at
the end of a long chain of calculations.
   Sundaram and Tendulkar also draw attention to items included in the national
accounts estimates but not in the surveys, such as the imputed rents of owner-
occupiers and expenditures by nonprofit institutions serving households. Like
Kulshrestha and Kar, they demonstrate the increasing importance of a relatively
new item, financial intermediation services indirectly measured (FISIM), introduced
in accord with the recommendations of the 1993 revision of the UN System of
National Accounts. FISIM—the difference between interest paid to and interest paid
by banks and other financial intermediaries—has been added to national accounts
estimates of household consumption since 1993, with some backdating to the
1980s. The idea is that interest charged to borrowers contains, in addition to the
market rate of interest, a charge for intermediation services to lenders, while interest
paid to lenders is lower than market rates, with the difference attributed to financial
intermediation services to depositors. The difference between interest paid and interest

182                                        The World Bank Research Observer, vol. 20, no. 2 (Fall 2005)
received is therefore a measure of the value of financial intermediation to borrowers
and lenders. A similar item is included for risk-bearing services, measured as the
profits of insurance companies.
   In India the value of FISIM increased from close to zero in 1983/84 to 2.5 percent of
consumption in 1993/94, accounting for a quarter of a percentage point per year of
the difference in annual growth rates in consumption between national accounts
and surveys. In measuring the living standards of the poor, it can reasonably be
doubted whether any of the value of financial intermediation is relevant.



Survey Methodology: Reporting Periods
The design of the Indian surveys has evolved over time and is continually under dis-
cussion. An important design issue for poverty measurement is the length of the
reporting period. The NSSO had adopted a uniform 30-day recall period, based on
experiments carried out by Mahalanobis and Sen (1954) in the 1950s. The NSSO ran
a new series of experiments in the “thin� survey Rounds 51 (1994/95) through 54
(1998), which are run between the quinquennial rounds, randomly assigning
households to one of two questionnaires with different reporting periods. A ques-
tionnaire with a 7-day reporting period for high-frequency items (food, pan,
tobacco), 365 days for low-frequency items (durable goods, clothing, footwear, insti-
tutional [hospital] medical care, and educational expenses), and 30 days for every-
thing else gave poverty counts that were only half of those derived from the
questionnaire with a uniform 30-day reporting period.
   The reduction in measured poverty comes from two quite separate effects. The
first is that a higher rate of monthly expenditure is reported when people are asked
to report food, pan, and tobacco over the past 7 days rather than over the past
30 days. Higher reported expenditure, other things being equal, decreases mea-
sured poverty. The second effect comes from the low-frequency items. Although
the mean reported expenditure for this category decreases for the longer reporting
period, the lower tail of the distribution increases. With 30-day reporting periods,
most households report no purchase of low-frequency items, but in 365-day peri-
ods most households report at least some purchases. Thus despite the decrease in
the mean, the longer reporting period for the low-frequency items also acts to
reduce measured poverty.
   Measures of inequality are substantially reduced by moving from a 30-day to a
365-day reporting period for low-frequency items. Because the mean falls and the
bottom tail increases, measured dispersion in these purchases is much reduced, and
this carries through to total expenditure. This means that it is never legitimate to
compare measured inequality across surveys with different reporting periods with-
out some sort of correction.

Angus Deaton and Valerie Kozel                                                      183
    The questionnaire experiments showed that reporting periods make a difference
but did not settle the question of which is better. To find out, the NSSO launched
another set of experiments in the Round 55 survey in 1999/2000, updating and
extending the experiments of Mahalanobis and Sen (1954). Alternative question-
naires were randomized over the experimental households, using three different
reporting periods, 7 days, 30 days, and a gold standard of daily visits accompanied
by direct measurement. A pilot study was undertaken in five Indian states from January
through June 2000. According to initial results reported by the NSSO Expert Group on
Non-Sampling Errors (2003), for rural households the 7-day estimates were on
average 23 percent higher than the 30-day estimates, somewhat lower than the dis-
crepancy in the large-scale NSSO thin-round surveys. But comparisons with the daily
estimates show that the 30-day estimates were more accurate than the 7-day esti-
mates for many important commodities, including cereals and cereal products. Over
all the goods examined, there is no clear superiority of one reporting period over
another, and there is little evidence that the traditional 30-day reporting period is
seriously inadequate. This important study does not support the apparently sensible
hypothesis that high-frequency items in India are better measured with a 7-day
than a 30-day recall. It also does not support the idea that the discrepancy
between the national accounts and sample survey measures of consumption is due
largely to underestimation in the survey data associated with an overly long report-
ing period.
    How reporting periods affect estimates of consumption and poverty is relevant for
surveys in many countries. Nor is it the only such issue. In literate populations,
respondents can be asked to keep diaries as an alternative or supplement to inter-
views. Surveyors can also visit households several times, for example, to take
account of seasonality in expenditures or to gather data a day or two at a time when
it is believed that respondents cannot accurately recall expenditures over long refer-
ence periods. There is considerable international experience with these issues (see,
for example, Deaton and Grosh 2000), although there can be no presumption that a
design that is good for one country will be good for another.



What Happened to Poverty in India in the 1990s?
At the end of the 1990s there had not been a large-scale consumer expenditure sur-
vey in India since 1993/94, so there were no official estimates of national or state
poverty rates after that date. The Planning Commission does not endorse poverty
estimates based on the smaller consumer expenditure surveys that the NSSO runs
between the quinquennial rounds, even though the sample sizes are large enough to
support accurate poverty estimates at the national level. The last of these thin sur-
veys before the Round 55 survey of 1999/2000 was a half-year survey in 1998.

184                                       The World Bank Research Observer, vol. 20, no. 2 (Fall 2005)
Neither it nor the immediately preceding small surveys showed progress in reducing
poverty (see, for example, Datt 1999). It is widely believed that there were problems
with the sampling in the rounds from 1994 through 1998, but there has been no
official confirmation of any such difficulty.


The Design of Round 55, 1999/2000
In planning for Round 55, the NSSO faced the immediate problem of deciding what
design to use. The experimental questionnaires tried in Rounds 51 through 54 had
showed more consumption and less poverty. The results of the expert group analysis
of the experimental questionnaires were not yet available, so there was little solid
scientific guidance. By contrast, the consequences for poverty estimation of adopting
the new questionnaire were well understood, so that a decision that would normally
be left to statistical experts became politicized.
   After considerable internal debate, a compromise solution was adopted. For food,
pan, and tobacco each household was asked to report all items over both a 7-day
and a 30-day period. The traditional 30-day reporting period for durables, clothing,
education, and institutional medical expenses was replaced by a 365-day period
only.
   Although this new, compound design might well be defensible on its own terms, it
was incompatible with previous surveys. Consumption and poverty estimates based
on it thus cannot reliably be used to assess trends or even changes from the previous
large survey, Round 50. For high-frequency items, having both the 7-day and the
30-day reporting periods is likely to prompt respondents to reconcile the two reports.
Thus, for example, the reported consumption of milk might be expected to be quite
similar for the 30-day period, on the one hand, and for the 7-day period, on the
other, once the 7-day results have been scaled up by a factor of 30 over 7, something
that might not happen if the same respondent were asked to report over one period
or the other but not both. Indeed, the means of total estimated consumption for the
two reporting periods are much more similar in Round 55 than was the case in the
experimental thin rounds, which randomly assigned each household to one or the
other reporting period, not both. It remained unclear whether this meant that con-
sumption reported with the 30-day recall was pulled up to meet the 7-day reports or
whether consumption reported for the 7-day recall periods was cut back to accord more
closely with the 30-day reports. Most likely, some combination of both was at work.
   The presence of both questionnaires on the survey increased the interviewing
time and forced a number of other changes in the survey. The employment and
unemployment survey, usually given to the same households that answer the con-
sumer expenditure schedules, was given to a separate sample of households in the
Round 55 survey. But even within the consumer expenditure schedule, there were
important changes, nearly all in the interests of compression and time saving.

Angus Deaton and Valerie Kozel                                                    185
   Despite these difficulties, the 30-day responses were adopted as the basis of the
new official poverty totals, although Planning Commission press releases also pro-
vided (lower) estimates of poverty using the 7-day reporting period results. Estimates
based on 30-day recall, which were the only ones even nominally comparable with
the previous poverty estimates from 1993/94, showed a marked reduction in pov-
erty rates between 1993/94 and 1999/2000. Estimated poverty fell from 37 percent
to 27 percent among rural households and from 33 percent to 24 percent among
urban households. All-India poverty fell 10 percentage points over the six-year
period, from 36 percent to 26 percent.
   Although these estimates were accepted by the government of India and vigor-
ously defended by at least one government minister, there was widespread skepti-
cism about their validity. Estimated poverty was believed to be too low because
reported consumption over the 30-day reporting period had been upwardly biased
by the simultaneous presence of the 7-day reporting period. But no one knew how
far off the official estimates were. One of the first researchers to see the difficulties
with Round 55 was Abhijit Sen (2000). Even before the results were published, he
delineated the contamination problems that were to dominate the interpretation of
Round 55, which he refers to as a “failed experiment.�
   Such problems are far from unique to India. There is always a conflict between
updating and improving a survey instrument on the one hand and maintaining con-
sistency in estimation on the other. Yet there have been few cases as dramatic or
where the consequences of the change were so little anticipated in advance.


Making Adjustments
Various researchers have developed ways of adjusting the Round 55 expenditure
data to correct the official estimates. These adjustments are based on assumptions
that allow imputation of missing data or adjustments to contaminated data and, by
the nature of the problem, at least some components of the assumptions are untest-
able. Assumptions differ, none is uncontroversial, and all have been debated. As will
be discussed later, the official counts have also been challenged based on the poverty
lines used and how they have been updated over time.
   Deaton (2003a) and Tarozzi (2003) base their corrections on the fact that an
important section of the questionnaire was unchanged between Rounds 50 and 55
and can therefore be compared. If the probability of being poor conditional on expen-
ditures on the consistently measured goods does not change from 1993/94 to 1999/
2000, then that probability can be calibrated from Round 50 and used together with
expenditure on the consistent items in Round 55 to estimate poverty in that round.
   According to Deaton’s and Tarozzi’s calculations, most of the “official� decline in
poverty is real. For rural households, for which official calculations show the poverty
headcount ratio falling from 37.3 percent in 1993/94 to 27.0 percent in 1999/2000,

186                                        The World Bank Research Observer, vol. 20, no. 2 (Fall 2005)
Deaton finds a fall from 37.3 percent to 30.2 percent, or 70 percent of the official
estimate. In the urban sector, where the official headcount ratio is 23.6, his estimate
is 24.7 percent, which reduces the decline from 8.8 percentage points to 7.5. Driv-
ing these results is the substantial increase in consumer expenditures on items that
were consistently surveyed using the 30-day reporting period and the difficulty of
explaining the increase without a substantial increase in total expenditure and thus
in the fraction of the population that is poor.
   Sundaram and Tendulkar (2003a, c, d) argue that the 30-day responses were not
much contaminated by the presence of the 7-day responses, based on an examina-
tion of the results of an abbreviated consumption questionnaire given to the employ-
ment and unemployment subsample of Round 55. If that is the case, the only source
of inconsistency between Round 50 and Round 55 questionnaires is the treatment
of the low-frequency items (clothing, durables, educational expenses, and institu-
tional medical expenditures), which were surveyed at 30-day reporting periods in
Round 50 and at 365-day periods in Round 55. Because Round 50 solicited expen-
ditures on these goods for both 30-day and 365-day reporting periods, however,
reconstructing total expenditure for Round 50 using the 365-day data permits con-
struction of a notionally consistent “mixed reference period� measure of per capita
expenditure for both Rounds 50 and 55.
   Sundaram and Tendulkar (2003d) estimate a decline in rural poverty from
34 percent in 1993/94 to 29 percent in 1999/2000, or about half the official estimate
and less than Deaton’s estimate, which was about 70 percent of the official estimate.
For urban households, Sundaram and Tendulkar estimate a poverty decline from
26 percent in 1993/94 to 23 percent in 1999/2000, confirming only about a third
of the official decline compared with Deaton’s confirmation of 85 percent.
   Sundaram and Tendulkar’s estimates are not comparable with official estimates
because of their use of the mixed reference period measure for Round 50 and because
they use different poverty lines than the state- and sector-specific poverty lines of the
Planning Commission. Rather than work with the Planning Commission’s poverty
lines, which have been called into doubt (as discussed later), Sundaram and Tendulkar
use the all-India lines for 1973/74, updated only for the general rate of price inflation.
Sundaram and Tendulkar (2003c, d, e) have also extended their results to the major
states and have used the same corrections to investigate what happened to the poverty
rates of different social and economic groups. In line with other work, particularly that
of Deaton and Drèze, discussed shortly, they find that some groups have done much
better than others. They conclude that although poverty reduction among some of the
most vulnerable groups (scheduled castes, agricultural laborers, and urban casual
laborers) has been in line with that of the general population, groups such as the
scheduled tribes have been left behind.
   Sen and Himanshu (2004) have recently challenged both Deaton’s and
Sundaram and Tendulkar’s conclusions. With respect to Deaton, they apply the

Angus Deaton and Valerie Kozel                                                        187
“consistent goods� methodology to calculate the probability of being poor according
to the mixed reference period definition of consumption used by Sundaram and
Tendulkar. Given the validity of the mixed reference period correction, the incom-
patibility of Rounds 50 and 55 comes from the responses on food, pan, and tobacco,
which were most likely biased upward in Round 55 by the presence of the seven-day
questions. In consequence, applying Deaton’s method with the mixed reference
period adjustment should raise the poverty count over the unadjusted counts. But
Sen and Himanshu show that exactly the reverse is true. Since it is implausible that
the presence of the 7-day questions resulted in a decrease in the consumption
responses in the 30-day reports, something had obviously gone wrong.
   Although exactly what went wrong is not yet entirely clear, the most likely possi-
bility is that Indian households have been shifting their consumption away from
food toward an assortment of nonfood goods and services, not just by moving along
the Engel curves (which would not be a problem for Deaton’s method) but by shift-
ing the Engel curves. As a result, the increase in consumption of miscellaneous
goods and services from 1993/94 to 1999/2000 cannot be used to measure the
extent of poverty reduction, as Deaton supposed.
   Sen and Himanshu also find unconvincing Sundaram and Tendulkar’s justifica-
tion for their use of the uncorrected 30-day expenditures for food, pan, and tobacco.
Sen and Himanshu produce new estimates using different, sensible, but largely ad
hoc corrections. Their estimates are in line with Sen’s (2000) original view that
there was little decline in headcount poverty in India in the 1990s. Using compara-
ble mixed reference periods for both rounds, they estimate that the rural headcount
ratio fell by only 2.7 percentage points between the two rounds, from 31.9 percent
to 29.1 percent, and the urban headcount ratio by 3.1 percentage points, from
29.2 percent to 26.1 percent. These estimates define the pessimistic pole in the
Indian poverty debate. Although there is no way to know with certainty what the
results of Round 55 would have been had the questionnaire been unchanged, this
small a drop in poverty during the 1990s seems implausible. There is considerable
evidence from sources other than the consumption surveys, such as information on
wage rates, ownership of durable goods, and incomes from other surveys, which,
though imperfect indicators on their own, taken together are extremely difficult to
reconcile with an India in which poverty rates are not declining.


Other Estimates of Poverty
The poverty estimates by Deaton (2003a, b), Sundaram and Tendulkar (2003d),
and Sen and Himanshu (2004) are all based on corrections to the Round 55 unit
record data. Several other researchers have followed alternative approaches, linking
poverty to external evidence. This is perhaps most explicit in the work by Datt and
others (2003), who use an econometric model that links poverty rates to their

188                                      The World Bank Research Observer, vol. 20, no. 2 (Fall 2005)
plausible determinants, including agricultural yields, nonfarm growth, development
spending, and inflation. Using the estimated model to project poverty for 1999/
2000 and ignoring the flawed data from Round 55, they find that the changes in the
explanatory variables would have warranted a decrease in the headcount ratio from
39 percent in 1993/94 to 34 percent in 1999/2000. This suggests that the pace of
poverty reduction was slightly lower in the 1990s than in the 1980s and lower than
might be expected given India’s high rate of economic growth in second half of the
1990s. The differences between official and predicted rates of progress are largely
due to slower progress in some of India’s largest and poorest states, particularly
Bihar, Uttar Pradesh, and Maharasthra. Bihar and Uttar Pradesh alone account for
over half the difference between official estimates and predicted poverty levels. Datt
and others’ projections are similar to Sundaram and Tendulkar’s (2003d) adjusted
calculations but show a good deal less poverty reduction than Deaton’s method does
and a good deal more than Sen and Himanshu’s.
   Kijima and Lanjouw (2003) combine forecasting with limited use of Round 55
data. Their forecasts are based on household characteristics that are among the ulti-
mate determinants of poverty. They use Round 50 data to regress household expen-
diture on household characteristics such as education of household members, land
holding, household size, caste, and a set of district dummy variables. The fitted
regression is then taken to the Round 55 data and used to predict total household
expenditure based on the same factors, and these predictions are then used to esti-
mate poverty. Kijima and Lanjouw’s poverty estimates are close to those of Sen and
Himanshu but show about half as much poverty reduction as those based on the
Deaton method.
   In assuming that returns to factors cannot change, Kijima and Lanjouw’s method
is inherently conservative. The major objection to these calculations is that if
poverty declined substantially over the 1990s and if, as might be expected (and
certainly hoped), the reductions were driven by increases in the rate of return to factors
such as labor, education, and land, then the calculations would not pick them up.
Those who argue that the reforms have reduced poverty in India do not claim that
the reforms have augmented India’s supply of factors, at least not in the short run,
but that they have raised the rate of return to those factors, for example, by allowing
people to participate in global markets previously closed to them. The Kijima and
Lanjouw poverty calculations rule out any such effects by construction and so
cannot address whether they exist.
   In another set of poverty estimates, Bery and Shukla (2003) update a similar
exercise by Lal and others (2001). These studies use information collected from the
Market Information Surveys of Households (MISH), large-scale annual surveys of
household expenditures on consumer durables and other consumption items run by
the National Council on Applied Economic Research. Because the design, sample
size, and survey methodology have been consistent over time, Bery and Shukla

Angus Deaton and Valerie Kozel                                                        189
argue that the MISH is useful for identifying trends in consumption patterns. The MISH
also includes a question on total household income, and these data are used to
estimate poverty. This analysis suggests that poverty has fallen sharply in India,
whether estimated from the income data or from the data showing increased owner-
ship of durable goods.
   A major concern with the MISH surveys is the adequacy of a single income ques-
tion: “What is your annual household income from all sources?� Household income
is a difficult concept to explain and to measure, especially for rural households,
many of whose members are self-employed in agriculture. For these households, cal-
culating incomes requires a great deal of imputation, as well as a careful separation
of business and personal expenses. For these reasons most creators and users of
household surveys do not regard such a question as useful. Unfortunately, it is diffi-
cult to explore these issues in detail, because the MISH data are proprietary and have
never been made available to independent researchers.
   Surjit Bhalla has been one of the most consistent advocates of the position that
poverty fell rapidly during the 1990s, not only in India but worldwide. Using
national accounts data to adjust the survey data, Bhalla (2003, p. 338) argues
that there was a sharp decline in poverty in India in the 1990s and that the official
estimate of 24 percent in 1999/2000 is too low: “It is almost incontrovertible that
poverty in India was less than 15 percent in 1999–2000.� Bhalla’s work defines
the opposite pole to that of Sen and Himanshu (2004) in the Indian poverty
debate.



The Choice of Poverty Lines
Although the recent debate on poverty in India has focused mainly on the measure-
ment of expenditures, poverty lines are equally important. How they are updated
and adjusted across regions or urban and rural households has a major effect on
poverty estimates. In India, as in many other countries, a base poverty line is
adjusted across time and space using price indexes, so the selection and construction
of these indexes become a key input into poverty measurement.
   The history of poverty lines in India is a case study in the interaction of science
and politics, with political decisions often claiming a scientific basis, sometimes with
justification, more often without. Although poverty lines are often linked to the
amount of money needed for a minimally adequate diet, the use and long-term sur-
vival of poverty lines depend on policymakers and others accepting them as useful.
For example, Rudra (1974), in discussing the history of Indian poverty lines up to
that time and the persistence of the “magic number� of 20 rupees per head in 1960/
61 prices shows that a food-based analysis would lead to a considerably higher number.
Yet the magic number persisted, as similar magic numbers have persisted in other

190                                        The World Bank Research Observer, vol. 20, no. 2 (Fall 2005)
countries, not because they are correct but because, once established as useful in
economic and political discussions, poverty lines are resistant to change.
   From the late 1970s to the mid-1990s the Planning Commission used only two
poverty lines for per capita household expenditure, 49 rupees for rural households
and 57 rupees for urban households at 1973/74 prices, which was close to the
15 percent urban price differential estimated by Bhattacharya and Chatterjee (1971)
using unit value data from the National Sample Survey. The poverty lines were held
constant in real terms and were converted to current rupees using the implicit price
deflator of consumption in the national accounts. This process ignored interstate dif-
ferences in price levels and in urban to rural price differentials. Furthermore, the
national accounts consumption deflator is probably not the best measure of inflation
for households near the poverty line. These problems and several others were dealt
with by an expert group in 1993 (India, Expert Group on Estimation of Proportion
and Number of Poor 1993). Their recommendations for new poverty lines were
adopted (in somewhat modified form) by the Planning Commission, and these pov-
erty lines have been used in official calculations since 1983 (back-casting the meth-
odology), as described in the first section.
   The expert group poverty lines have a serious flaw, however: the urban to rural
price differentials that they imply are too large to be credible. It is unclear how this
happened, whether because of an error in calculation or because the price indexes
used in the calculations produced the result through some unexpected cumulative
effect. The state by state urban and rural poverty lines were calculated indepen-
dently, without consideration of the implicit urban to rural price differentials. In any
case, the average ratio of urban to rural poverty lines is around 1.4 and varies
widely across states. In Round 50 (1993/94) it is more than 1.7 in Andhra Pradesh
and nearly as high in Maharashtra, Madhya Pradesh, and Karnataka, but less than
unity in Assam. As a result, official headcount measures of poverty are higher in
urban than in rural areas in some states, and the all-India headcount ratios differ lit-
tle for urban and rural areas. In Andhra Pradesh, which is the most dramatic exam-
ple, the 1999/2000 official estimates give a poverty rate of 27.2 percent for urban
areas and only 10.8 percent for rural areas. In addition, it is unclear whether there is
any good basis for the differences in poverty lines across states, given that the studies
used by the expert group were outdated even a decade ago.
   Another serious issue is the accuracy of the inflation rate used in the state-level
price indexes. Errors in the indexes will induce errors in the trend rate of poverty
reduction. These indexes are reweighted infrequently. For example, until 1995 the
consumer price index for agricultural laborers used weights based on a 1960/61
survey. And although this index and the index for industrial workers are almost cer-
tainly better than the price deflator of national accounts consumption, it is unclear
whether the prices or the weights that go into these indexes are the right ones for a
national poverty measure.

Angus Deaton and Valerie Kozel                                                       191
   Alternative price indexes can be calculated directly from the National Sample
Survey consumption surveys. Because respondents report both expenditures and
quantities for most foods in the survey as well as for tobacco, alcohol, and fuels, it is
possible to calculate unit values for each good for each household and to use these
unit values to form price indexes for urban and rural areas, states, and different
rounds of the survey. The details of this work were originally laid out in Deaton and
Tarozzi (2005) and were brought up to date by Deaton (2003b), who presents a set
of poverty lines based on the indexes. These price indexes are quite different from the
indexes implicit in the official poverty lines and much closer to the uniform 16 percent
urban to rural price differential originally calculated by Bhattacharya and Chatterjee
(1971) using similar procedures. As a result, headcount ratios calculated using
Deaton and Tarozzi’s lines show much higher rates of rural poverty relative to urban
poverty than do the traditional lines. Because their price indexes also rise somewhat
less rapidly than do the official indexes, the associated poverty rates for the country
as a whole decline somewhat more rapidly than do the official rates, at least until
Round 55 in 1999/2000, when contamination becomes an issue.
   Deaton and Tarozzi’s poverty lines and their procedures for correcting the Round
55 data form the basis for an analysis of poverty and inequality by Deaton and Drèze
(2002). Their estimates show a fairly steady rate of poverty decline in India in the
1980s and 1990s. Deaton and Drèze argue that their estimates are broadly consis-
tent with a range of other evidence, including growth rates from state GDP accounts
and growth in agricultural wages. So even if their estimates for 1999/2000 are
overoptimistic, as Sen and Himanshu (2004) argue, their general assessment of
trends is unlikely to be wildly incorrect.
   Deaton and Drèze show that the new poverty lines sometimes make a dramatic
difference to the estimates, particularly for urban and rural comparisons. For the
previously mentioned example of Andhra Pradesh, they estimate that 10.5 percent
of the urban population were poor in 1999/2000 and 26.2 percent of the rural
population, compared with the official estimate of 27.2 percent urban poverty and
10.5 percent rural poverty. They also note that there was a marked increase in consump-
tion inequality in the late 1990s between states, with the better-off states in the
south and west growing more rapidly than the poorer states in the north and east;
between rural and urban households, with growth more rapid in urban than in
rural areas; and within the urban sectors of many states, where consumption has
been growing more rapidly among the wealthiest.
   The questionnaire changes in Round 55 have obscured the trends in inequality.
The substitution of a 365-day reporting period for a 30-day period for low-frequency
items pulls up the bottom tail of the distribution of expenditures for those items
while simultaneously depressing its mean, so that expenditures on low-frequency
items are less unequally distributed than with a 30-day reporting period. This effect
carries through to per capita total household expenditure, artificially compressing

192                                        The World Bank Research Observer, vol. 20, no. 2 (Fall 2005)
its distribution in Round 55 compared with Round 50. This questionnaire-driven
reduction in measured inequality obscures the underlying increase in inequality,
which cannot be detected if the old and new questionnaires are compared without
explicit correction.
   A different perspective on inequality is provided by Banerjee and Piketty (2005).
Their data come from individual income tax returns, which, under some heroic
assumptions, can be used to document the very top end of the Indian income distri-
bution. As many observers have conjectured, Indians with the highest incomes did
extremely well during the boom of the 1990s. Banerjee and Piketty calculate that
average incomes increased by 70 percent in real terms for the top 1 percent and tri-
pled for the top 0.01 percent. This is the first evidence on the extent of increasing
inequality attributable to the rising income share of the wealthiest Indians. Their
findings are likely to trigger further analysis and debate. Their work is also relevant
for the debate over the use of national accounts and surveys. If inequality is increas-
ing, and if people with higher incomes are less likely than others to respond to sur-
veys, then the ratio of measured consumption in the surveys to true consumption
will fall. Indeed, Banerjee and Piketty show that around a quarter of the increase in
the gap between national accounts and survey estimates can plausibly be attributed
to the increase in inequality because of the rising income share of high-income
Indians.



Updates and Lessons
Despite an extensive body of empirical work by eminent researchers, the debate on
what happened to poverty in India in the 1990s continues. No doubt there will be
more reinterpretations of the 1999/2000 survey—though diminishing returns
have surely set in—and new data are continuously being collected. There have
been four thin rounds since the large survey, Rounds 56, 57, 58 (a half-year sur-
vey), and 59. But a preliminary examination suggests as much confusion as clarifi-
cation. Round 56, conducted in 2000/01, shows a further reduction in poverty,
very much along the trend line calculated by Deaton and Drèze (2002). The Round
57 data for 2001–2002 show a sharp increase in poverty over the previous year,
which appears to have been even more sharply reversed in the last six months of
2001–2002, according to Round 58. The Round 59 poverty numbers are close to
the level in Round 58. It is unclear why there should be such a pattern, although
the disposal of large public cereal stocks in 2000–2001 temporarily reduced the
rural consumer price index and certainly induced a fall in rural poverty. Yet these
sharp fluctuations from survey to survey also raise the suspicion that as was the
case from 1994 to 1998, the thin round surveys suffer from some little understood
sampling problem.

Angus Deaton and Valerie Kozel                                                     193
   It is impossible for a country to pursue a coherent antipoverty strategy without an
adequate poverty monitoring system. The statistical problems and debates discussed
here concern a country with one of the best developed survey capacities in the
world. The NSSO statisticians are highly skilled, and field staff are disciplined and well
trained. Matters are almost certainly worse elsewhere.
   Several conclusions and lessons can be drawn from the great Indian poverty
debate, for India and for other countries. Most lessons are especially relevant for sta-
tistical offices, but several are important for policymakers and researchers.


Lessons for Statistical Offices
India’s experience demonstrates the importance of annual consumption or income
surveys. With major poverty surveys occurring only every five or six years, the fail-
ure of even one survey means that there is no reliable estimate of the rate of poverty
decline for more than a decade. India switched from large annual surveys to less fre-
quent surveys because it was thought that poverty changed too slowly to make
annual measurement necessary. But the reduced frequency does not make adequate
allowance for variability in conditions and survey structure. In largely agricultural
societies, where harvests vary from year to year, a 5-year cycle runs the risk of hav-
ing surveys coincide with unusual harvests, so that it can take 10 or 15 years to
establish a trend. Or, as with Round 55 in India, if a particular survey is compro-
mised, and if the survey period coincides with important policy changes, there can
be a long gap between a policy change and measurement of the effect on poverty. In
India there will be no poverty measures that are comparable to the 1993/94 esti-
mates until the 2004/05 survey results are available in early 2006.
   Good annual surveys, even if not comprehensive, are an insurance against these
problems. In India the thin rounds could serve this function in principle, but there
have been persistent doubts about their validity based on sampling methods and on
the frame from which the samples are drawn, which is typically not the same as that
used for the large rounds.
   In balancing consistent survey design with changes to reflect economic develop-
ment, new areas of interest, and improved survey technologies, the key point is that
survey design should change slowly relative to the frequency of poverty monitoring.
Revisions once every five years may be acceptable if poverty surveys are run every
year, but catastrophic if poverty monitoring is done at five-year intervals. For mea-
suring changes in poverty, a single survey is of little value without an earlier, com-
parable survey. When major design changes (such as changes in reporting periods)
are implemented, it is desirable to randomly split the survey between the old and
new design, to provide a basis for both forward and backward comparability.
   The Indian NSSO has had a long history of experimentation in using what are
essentially randomized controlled trials to assess the consequences of changes in

194                                         The World Bank Research Observer, vol. 20, no. 2 (Fall 2005)
survey practice. That such experimentation precipitated the failure of the major
1999/2000 round should not be taken as a condemnation of experimentation.
Rather, the fault lies with an oversight mechanism that failed to stop the introduc-
tion of what was effectively a completely new and untested design into an important
and long-awaited survey. Statistical offices are often reluctant to undertake costly
experiments that have no immediate input into published statistics. But without
experimentation there can be no satisfactory progress in survey design. If properly
used, experimentation can help prevent bad designs from being adopted in regular
surveys.
    Another general issue concerns the relationship between the national accounts
and the surveys. These are usually the responsibility of different government agen-
cies, which complicates collaboration for resolving discrepancies and sharing data to
improve performance. Placing both functions within an overall national statistical
office or otherwise making collaboration easier is likely to improve the performance
of both. India’s experience shows that scaling up survey data to match the national
accounts is not a simple shortcut to better poverty estimates. The two measures of
consumption are not the same, and there is no straightforward way to adjust one to
the other. India’s experience also demonstrates that when surveys and national
accounts disagree, it should never be assumed that one or the other is correct. In
many countries growth as measured in the national accounts appears to be greater
than growth as measured through surveys of income and consumption. That makes
it difficult to use the national accounts to cross-check survey data or to forecast pov-
erty trends in the absence of good survey data. Once again, there is no shortcut to
poverty monitoring without a sound, regular survey.
    Finally, India’s experience shows how important it is for the survey agency to
develop a dialogue with outside users, particularly independent researchers and
journalists. Very little of the Indian debate would have taken place without public
access to the raw data, particularly by local analysts. This openness is a major
change from policy in the 1980s, when the raw data were not released to outside
researchers. That said, there still appears to be no effective mechanism through
which users can bring questions and suggestions to those who process and design
the surveys.


Lessons for Policymakers
Policymakers are accountable to the whole population, not just to the poor. But pov-
erty reduction is a key aim of policy in India and in most other poor countries. Thus
policymakers are accountable to the public for poverty reduction, and progress can-
not be monitored without an adequate statistical agency. Policymakers need to
establish and fund such an agency and ensure adequate financing for annual sur-
veys of consumption or income. Policymakers are also responsible for ensuring the

Angus Deaton and Valerie Kozel                                                      195
openness of statistical data, including unit record data, something that statistical
offices are sometimes reluctant to implement without orders from above. Without
domestic openness and debate, policymakers are unlikely to be held accountable for
their policies. Although many countries do not have the domestic statistical or ana-
lytical capacity to generate broad debate, as in India, moving in that direction would
be a worthy aim for many countries.
   There is no suggestion here that the statistical failures in India in the 1990s were
the result of undue interference by politicians or policymakers in data collection or
publication. Yet politics in the broad sense played a role. In evaluating the reforms,
the political right had an interest in showing low poverty, and the political left in
showing high poverty, and this undoubtedly intensified the debate on survey design
and led to the unfortunate compromise design that temporarily undermined the
poverty monitoring system.
   This politicization of data collection and interpretation is often bemoaned. Yet
political accountability is essential to poverty reduction, and policymakers have a
legitimate interest in monitoring the statistical system and asking for changes that
serve their interests. The principles of insulating statistical services from undue polit-
ical influence are well understood (see Martin and others 2002), but poverty moni-
toring systems are ultimately part of the political process. Most countries can only
envy India its statistical capacity and the central part that poverty and poverty mea-
surement play in Indian public life.


Lessons for Researchers
Mistakes are inevitable, and survey data can be compromised by internal and exter-
nal factors. Thus poverty assessments will often have to be made using imperfectly
comparable surveys. India’s experience illustrates the possibility of repairs to
enhance the credibility of estimates. But that experience also demonstrates that
repairs, however creative, are a poor substitute for the collection of clean, credible,
and comprehensive data. What are convincing assumptions to one person can be
unconvincing to another, and political positions inevitably influence the assump-
tions that people are prepared to make or accept.
   Nevertheless, the Indian debate shows that discussion and advance are possible, even
among those with very different preconceptions, and that the balance of opinion can be
changed by well-reasoned and transparent arguments. Some of the arguments and
methods used in the Indian debate were country specific (for example, the use of the sup-
plementary consumption questions in the employment and unemployment survey), but
some are more broadly applicable for dealing with partially contaminated data. This is
an area in which there are likely to be returns to good, technical statistical work.
   Another lesson is the importance of high-quality domestic researchers. Analysis
by outside researchers can often be helpful, but it has a quite different effect from a

196                                         The World Bank Research Observer, vol. 20, no. 2 (Fall 2005)
domestic debate among local researchers, policymakers, and the press. India is still
exceptional in having a large group of domestic researchers who can work with
basic data, but this is a relatively new phenomenon even in India. The Indian press
has also played a distinguished role in the debate. Daily newspapers regularly report
on new findings on poverty and regularly carry related op-ed pieces. A unique con-
tribution to the poverty debate was made by Economic and Political Weekly, in whose
pages much of the Indian debate took place. It is a cross between an academic jour-
nal and a popular magazine like The Economist. Published weekly, it provides rapid
dissemination of viewpoints and research findings and links researchers, the press,
and policymakers. Other countries, including many rich countries, would do well to
emulate its approach.



Notes
Angus Deaton is Dwight D. Eisenhower Professor of International Affairs and Professor of Economics in
the Woodrow Wilson School, Princeton University; his email address is deaton@princeton.edu. Valerie
Kozel is senior economist in the Social Protection Network at the World Bank; her email address is
vkozel@worldbank.org. This article is based on the introduction to the volume The Great Indian Poverty
Debate. The authors are grateful for comments and suggestions from the editorial board of the World
Bank Research Observer, as well as from T. N. Srinivasan. They would also like to acknowledge continu-
ing discussions with many of the people whose work is cited in this review.




References
Banerjee, Abhijit, and Thomas Piketty. 2005. “Top Indian Incomes: 1922–2000.� World Bank
  Economic Review 19(1):1–20.
Bery, Suman, and R. K. Shukla. 2003. “NCAER’s Market Information Survey of Households—Statistical
   Properties and Applications for Policy Analysis.� Economic and Political Weekly 37(4):350–54.
Bhalla, Surjit S. 2002. Imagine There Is No Country: Poverty, Inequality, and Growth in the Era of Globalization.
  Washington, D.C.: Institute for International Economics.
———. 2003. “Recounting the Poor: Poverty in India, 1983–1999.� Economic and Political Weekly
 37(4):338–49.
Bhattacharya, N., and Chatterjee, G. S. 1971. “On Rural-Urban Consumer Prices and per Capita
  Household Consumption in India by Levels of Living,� Sankhyà, Series B, 33(3,4).
Datt, Gaurav. 1999. “Has Poverty Declined since Economic Reforms? Statistical Data Analysis.� Eco-
   nomic and Political Weekly 34(50):3516–18.
Datt, Gaurav, Valerie Kozel, and Martin Ravallion. 2003. “A Model-Based Assessment of India’s
   Progress in Reducing Poverty in the 1990s.� Economic and Political Weekly 37(4):355–61.
Deaton, Angus. 2003a. “Adjusted Indian Poverty Estimates for 1999–2000.� Economic and Political
  Weekly 37(4):322–26.
———. 2003b. “Prices and Poverty in India, 1987–2000.� Economic and Political Weekly
 37(4):362–68.


Angus Deaton and Valerie Kozel                                                                              197
———. 2005. “Measuring Poverty in a Growing World: Or Measuring Growth in a Poor World.�
 Review of Economics and Statistics 87(1):1–19.
Deaton, Angus, and Jean Drèze. 2002. “Poverty and Inequality in India: A Reexamination.� Economic
  and Political Weekly, September 7, pp. 3729–48.
Deaton, Angus, and Margaret Grosh. 2000. “Consumption.� In Margaret Grosh and Paul Glewwe,
  eds., Designing Household Questionnaires for Developing Countries: Lessons from Fifteen Years of the Living
  Standard Measurement Study. Washington, D.C.: World Bank.
Deaton, Angus, and Alessandro Tarozzi. 2005. “Prices and Poverty in India.� In Angus Deaton and
  Valerie Kozel, eds., The Great Indian Poverty Debate. New Delhi: Macmillan.
Garner, Thesia I., George Janini, William Passero, Laura Paszkiewicz, and Mark Vendemia. 2003. “The
  Consumer Expenditure Survey in Comparison: Focus on Personal Consumption Expenditures.� U.S.
  Department of Labor, Bureau of Labor Statistics, Washington, D.C.
India, Expert Group on Estimation of Proportion and Number of Poor. 1993. Report of the Expert Group
   on Estimation of Proportion and Number of Poor. New Delhi: Planning Commission.
Kijima, Yoko, and Peter Lanjouw. 2003. “Poverty in India during the 1990s: A Regional Perspective.�
    Policy Research Working Paper 3141. World Bank, Washington, D.C.
Kulshrestha, A. C., and Aloke Kar. 2005. “Consumer Expenditure from the National Accounts and
  National Sample Survey.� In Angus Deaton and Valerie Kozel, The Great Indian Poverty Debate. New
  Delhi: Macmillan.
Lal, Deepak, Rakesh Mohan, and I. Natarajan. 2001. “Economic Reform and Poverty Alleviation: A
   Tale of Two Surveys.� Economic and Political Weekly 36(12):1017–28.
Mahalanobis, P. C., and S. B. Sen. 1954. “On Some Aspects of the Indian National Sample Survey.�
  Bulletin of the International Statistical Institute 34:5–14.
Martin, Margaret E., Miron E. Straf, and Constance F. Citro. 2002. Principles and Practices for a Federal
  Statistical Agency, 2nd ed. Washington, D.C.: National Research Council.
Minhas, B. S. 1988. “Validation of Large-Scale Sample Survey Data: Case of NSS Household Consump-
  tion Expenditure.� Sankhya, Series B 50(supp):1–63.
Mukherjee, M., and G. S. Chatterjhee. 1974. “On the Validity of NSS Estimates of Consumption Expendi-
  ture.� In Pranab Bardhan and T. N. Srinivasan, eds., Poverty and Income Distribution in India.
  Calcutta: Statistical Publishing Society.
NSSO(National Sample Survey Organization) Expert Group on Non-Sampling Errors. 2003. “Suitability
   of Different Reference Periods for Measuring Household Consumption: Results of a Pilot Study.�
   Economic and Political Weekly 37(4):307–21.
Rudra, Ashok. 1974. “Minimum Levels of Living: A Statistical Examination.� In Pranab Bardhan and
  T. N. Srinivasan, eds., Poverty and Income Distribution in India. Calcutta: Statistical Publishing Society.
Sala-i-Martin, Xavier. 2002. The Disturbing “Rise� in Global Income Inequality.          NBER   Working Paper
   8904. National Bureau of Economic Research, Cambridge, Mass.
Sen, Abhijit. 2000. “Estimates of Consumer Expenditure and Its Distribution: Statistical Priorities after
   the 55th Round.� Economic and Political Weekly 35(51):4499–518.
Sen, Abhijit, and Himanshu. 2004. “Poverty and Inequality in India: Getting Closer to the Truth.�
   Economic and Political Weekly 39(38):4247–63.
Srinivasan, T. N., P. N. Radhakrishnan, and A. Vaidyanathan. 1974. “On the Distribution of Expendi-
   tures in India: An Evaluation.� In Pranab Bardhan and T. N. Srinivasan, eds., Poverty and Income
   Distribution in India. Calcutta: Statistical Publishing Society.
Sundaram, K., and Suresh Tendulkar. 2003a. “Poverty Has Declined in the 1990s: A Resolution of the
  Comparability Problems in NSS Consumer Expenditure Data.� Economic and Political Weekly, January 25,
  pp. 327–37.


198                                                  The World Bank Research Observer, vol. 20, no. 2 (Fall 2005)
———. 2003b. “NAS-NSS Estimates of Private Consumption for Poverty Estimation: A Further Comparative
 Examination.� Economic and Political Weekly, January 25, pp. 376–84.
———. 2003c. “Poverty in India in the 1990s: An Analysis of Changes in 15 Major States.� Economic
 and Political Weekly, April 5, pp. 1385–93.
———. 2003d. “Poverty in India in the 1990s: Revised Results for All-India and 15 Major States for
 1993–94.� Economic and Political Weekly, November 15, pp. 4865–72.
———. 2003e. “Poverty among Social and Economic Groups in India in 1990s.� Economic and Political
 Weekly, December 13, pp. 5263–76.
Tarozzi, Alessandro. 2003. “Calculating Comparable Statistics from Incomparable Surveys, with an
   Application to Poverty in India.� Duke University, Department of Economics, Durham, N.C.
Triplett, Jack E. 1997. “Measuring Consumption: The Post-1973 Slowdown and the Research Issues.�
   Federal Reserve Bank of St. Louis Review May/June:9–42.




Angus Deaton and Valerie Kozel                                                                 199