deaton dreze poverty india

7/29/2019 Deaton Dreze Poverty India

1/20

Economic and Political Weekly September 7, 2002 3729

Special articles

Poverty trends in India in the ninetieshave been a matter of intense con-troversy.1 The debate has often

generated more heat than light, and con-

fusion still remains about the extent towhich poverty has declined during theperiod. In the absence of conclusive evi-dence, widely divergent claims have flour-ished. Some have argued that the ninetieshave been a period of unprecedented im-provement in living standards. Others haveclaimed that it has been a time of wide-spread impoverishment.2 Against thisbackground, this paper presents a reassess-ment of the evidence on poverty and in-equality in the nineties.

So far, the debate on poverty in the

nineties has focused overwhelmingly onchanges in the headcount ratio theproportion of the population below thepoverty line. Accordingly, we begin (inSection I) with a reassessment of theevidence on headcount ratios and relatedpoverty indexes, based on National SampleSurvey (NSS) data. In particular, we presenta new series of internally consistentpoverty indexes for the last three quin-quennial rounds (1987-88, 1993-94 and1999-2000). The broad picture emergingfrom these revised estimates is one of

sustained poverty decline in most states(and also in India as a whole) during thereference period. It is important to note,however, that the increase in per capita

expenditure associated with this decline inpoverty is quite modest, e g, 10 per centor so between 1993-94 and 1999-2000 atthe all-India level.

In Section II, we consider related evi-dence from three additional sources: theCentral Statistical Organisations nationalaccounts statistics, the employment-un-employment surveys of the NationalSample Survey, and data on agriculturalwages. We find that these independentsources are broadly consistent with therevised poverty estimates presented in

Section I. In particular, real agriculturalwages in different states (which are highlycorrelated with headcount ratios of ruralpoverty) have grown at much the same rateas the corresponding NSS-based estimatesof per capita expenditure in rural areas.While each of these sources of informa-tion, including the National Sample Sur-vey, has important limitations, they tendto corroborate each other as far as povertydecline is concerned, and the combinedevidence on this from different sources isquite strong.

The evidence on inequality is discussedin Section III, where we focus mainly onthe period between 1993-94 and 1999-2000. Based on further analysis of Na-

tional Sample Survey data and relatedsources, we argue that there has been amarked increase in inequality in the nine-ties, in several forms. First, there has beenstrong divergence of per capita expen-diture across states, with the already better-off states (particularly in the southern andwestern regions) growing more rapidlythan the poorer states. Second, rural-urbandisparities of per capita expenditure haverisen. Third, inequality has increased withinurban areas in most states. The combinedeffects of these different forms of rising

inequality are quite large. In the rural areasof some of the poorest states, there hasbeen virtually no increase in per capitaexpenditure between 1993-94 and 1999-2000. Meanwhile, the urban populationsof most of the better-off states have en-joyed increases of per capita expenditureof 20 to 30 per cent, with even largerincreases for high-income groups withinthese populations.

Section IV takes up some qualificationsand concerns. We pay special attention tothe apparent decline of cereal consumption

Poverty and Inequality in India

A Re-ExaminationThis paper presents a new set of integrated poverty and inequality estimates for India

and Indian states for 1987-88, 1993-94 and 1999-2000. The poverty estimates are broadlyconsistent with independent evidence on per capita expenditure, state domestic product

and real agricultural wages. They show that poverty decline in the 1990s proceeded more orless in line with earlier trends. Regional disparities increased in the 1990s, with the

southern and western regions doing much better than the northern and easternregions. Economic inequality also increased within states, especially within urban areas, and

between urban and rural areas. We briefly examine other development indicators,

relating for instance to health and education. Most indicators have continued to improve inthe nineties, but social progress has followed very diverse patterns, ranging from

accelerated progress in some fields to slow down and even regression in others. We findno support for sweeping claims that the nineties have been a period of

unprecedented improvement or widespread impoverishment.

ANGUS DEATON, JEAN DREZE


2/20

Economic and Political Weekly September 7, 20023730

in the nineties, which is not obviouslyconsistent with the notion that poverty hassteadily declined during that period. Wealso consider the possibility of impover-ishment among specific regions or socialgroups, in spite of the general improve-ment in living conditions. Finally, wecomment on the unresolved puzzle of thethin rounds.

In Section V, we argue for supplement-ing expenditure-based data with other in-dicators of living standards, focusing forinstance on literacy rates, health achieve-ments, nutritional levels, crime rates, andthe quality of the environment. This broaderapproach sheds a different light on povertytrends in the nineties. In particular, itprompts us to acknowledge that socialprogress has been uneven across the dif-ferent fields. For instance, the ninetieshave been a period of fairly rapid increasein literacy and school participation. On the

other hand, there has been a marked slow-down in the rate at which infant mortalityhas been declining, and a significant in-crease in economic inequality. An inte-grated assessment of changes in livingconditions has to be alive to these diver-sities. We also discuss other implicationsof this broader approach to the evaluationof living standards, going beyond thestandard poverty indexes.

The concluding section sums up theinsights of this enquiry.

IPoverty Indexes in the

Nineties

I.1 Official Estimates

We begin with an examination of house-hold per capita consumption and the as-sociated poverty estimates. Consumptionis only one element of well-being, but itis an important element, and much interestis rightly attached to the PlanningCommissions periodical estimates of

poverty-based on National Sample Surveydata. The most widely-used poverty indi-cator is the headcount ratio (hereafterHCR), i e, the proportion of the populationbelow the poverty line.

The latest year for which relativelyuncontroversial HCR estimates are avail-able is 1993-94, corresponding to the 50thRound of the National Sample Survey, aquinquennial round. This round wasfollowed by a series of so-called thinrounds, involving smaller samples andsomewhat different sampling designs;

indeed, in the last of these, the 54th Round,the survey was only in the field for sixmonths rather than the customary year andis therefore most unlikely to be compa-rable with any previous survey. These thinrounds suggested not only that povertyremained more or less unchanged between1993-94 and the first six months of 1998(the reference period for the 54th Round),

but also that average per capita expendi-ture stagnated during this period of rapideconomic growth. This is very difficult tosquare with independent evidence, e g,from national accounts statistics. As thingsstand, we do not have a good understand-ing of why the thin rounds give what appearto be anomalous results, and until thatpuzzle is resolved, our confidence in ourother results must remain qualified. Weshall return to this issue in Section IV.3,and ignore the thin rounds in the mean-time.

In contrast to the thin rounds, the officialcounts from the latest quinquennial round(the 55th Round, pertaining to 1999-2000)suggest considerable poverty decline be-tween 1993-94 and 1999-2000. Accordingto official estimates, widely relayed, theall-India headcount ratio declined from 36to 26 per cent over this short period. As iswell known, however, the 55th Round isnot directly comparable to the 50th Round,due to changes in questionnaire design.

Briefly, the problem is as follows. Afterthe 50th Round, the National SampleSurvey introduced an experimental ques-tionnaire with different recall periods fordifferent classes of goods, in addition tothe traditional 30-day recall question-naire. The experimental questionnaire useda seven-day recall period for food, pan, andtobacco, as well as a 365-day recall period

for less frequently purchased goods suchas durables, clothing, footwear, educationaland institutional medical expenditures.Prior to 1999-2000, the traditional 30-dayrecall questionnaire and the experimentalquestionnaire were administered to differ-ent(and independent) samples of house-holds. These alternative questionnairesproduced two independent series of ex-penditure estimates, with a fairly stableratio of the lower estimates based on thetraditional questionnaire to the higherestimates based on the experimental ques-

tionnaire. In 1999-2000, the 30-day recalland seven-day recall periods for food,pan and tobacco were used for the samehouseholds, in two adjacent columns onthe same pages of a single questionnaire.This effectively new questionnaire de-sign led to a sudden reconciliation of theresults obtained from the two differentrecall periods, perhaps reflecting effortsto achieve consistency on the part ofinvestigators and/or respondents. This

Table 1a: All-India Headcount Ratios(Per cent)

1987-88 1993-94 1999-00

Rural

Official estimates 39.4 37.1 26.8

Adjusted estimates:Step 1: Adjusting for changes in questionnaire design 39.4 37.1 30.0Step 2: Revising the poverty lines 39.4 33.0 26.3

Urban

Official estimates 39.1 32.9 24.1

Adjusted estimates:Step 1: Adjusting for changes in questionnaire design 39.1 32.9 24.7Step 2: Revising the poverty lines 22.5 17.8 12.0

Source: Planning Commission, Press Releases (March 11, 1997, and February 22, 2001), Deaton(2001a, b), and Table 2a below.

Table 1b: All-India Poverty-Gap Indexes

1987-88 1993-94 1999-00

Rural

Estimates from unadjusted data and official poverty lines 9.4 8.4 5.2Adjusted estimates:

Step 1: Adjusting for changes in questionnaire design 9.4 8.4 6.4

Step 2: Revising the poverty lines 9.4 7.0 5.2UrbanEstimates from unadjusted data and official poverty lines 10.4 8.3 5.2

Adjusted estimates:

Step 1: Adjusting for changes in questionnaire design 10.4 8.3 5.9

Step 2: Revising the poverty lines 4.8 3.7 2.3

Source:Authors calculations from unit record data from the 43rd, 50th, and 55th Rounds of the NSS.


3/20


reconciliation is likely to boost the expen-diture estimates based on 30-day data, andtherefore to pull down the official povertycounts, which are based on these 30-dayexpenditures. In addition, only the 365-day questionnaire was used for the lessfrequently purchased items, and this aban-donment of the traditional 30-day recallfor durables and other items also brings

down the poverty count. Indeed, mostpeople report no such purchases over 30days, but report something over 365 days.The bottom tail of the consumption distribu-tion is thereby pulled up, reducing bothpoverty and inequality compared with theprevious design. For this reason, as wellas because of possible reconciliation be-tween seven-day and 30-day reports, thelatest headcount ratios are biased downcompared with what would have beenobtained on the basis of the traditionalquestionnaire.

There is another, quite different problemwith the official estimates, which does notconcern the 55th Round specifically. Thisrelates to the state and sector specificpoverty lines that are used by the PlanningCommission to compute the poverty es-timates. In several cases the poverty linesare implausible, particularly the very muchhigher urban than rural lines in severalstates. The source of the problem lies inthe use of defective price indexes in ad-justments of the poverty line over time andbetween states. In the next section, we

discuss ways of overcoming this problemand other limitations of the official povertyestimates.

I.2 Proposed Adjustments

In this paper, we present a new seriesof consistent poverty estimates for the mostrecent quinquennial rounds (1987-88,1993-94 and 1999-2000).3 Essentially,these involve four major departures fromthe official estimates. First, an attempt ismade to adjust the 55th-Round estimates

to achieve comparability with the earlierrounds. Second, we use improved priceindexes to update the poverty line overtime, and to derive state-specific povertylines from the all-India poverty line. Third,a similar procedure is used to derive anexplicit estimate of the appropriate gapbetween rural and urban poverty lines (incontrast with the often implausible rural-urban gaps that are implicit in the officialestimates). Fourth, in addition to correctedheadcount ratios, we present estimatesof a potentially more informative poverty

indicator, the poverty-gap index. Eachof these departures calls for furtherdiscussion.

The possibility of adjusting the 1999-2000 poverty estimates arises from the factthat the 55th Round questionnaire retainedthe 30-day recall (and 30-day recall only)approach for a number of items such asfuel and light, non-institutional medical

care, and large categories of miscellaneousgoods and services. Further, it turns outthat expenditure on this intermediate groupof commodities is highly correlated withtotal expenditure.4 Expenditures on thesecomparably surveyed goods can thereforebe used to get an idea of trends in totalexpenditures, and hence, of trends inpoverty.

This procedure is valid if two assump-tions hold. The first is that reported ex-penditures on the intermediate goods, forwhich the recall period is unchanged, are

unaffected by the changes elsewhere in the

questionnaire. The second is that the re-lation between intermediate-goods expen-diture and total expenditure is much thesame in 1999-2000 as in 1993-94.5 Thesecond assumption would be underminedby a major change in relative prices of theintermediate goods relative to other goodsin the late 1990s. It can be checked to someextent by applying the proposed method

to the thin rounds instead of the 55thRound, and comparing the predicted dis-tribution of total expenditure with the actualdistribution. These checks suggest that thecorrection procedure works reasonably well[Deaton 2001a, Tarozzi 2001]. However,this should not be regarded as a definitivevalidation of the proposed method, giventhe ambiguities associated with the thinrounds (Section IV.3). There are otherpossible approaches to adjustment that havenot yet been explored, and further workmay lead to different conclusions. Mean-

while, we regard our adjusted figures as

Table 2a: State-Specific Headcount Ratios(Per cent)

Official Methodology Adjusted Estimates

198788 199394 199900 198788 199394 199900

RuralAndhra Pradesh

Assam

BiharGujaratHaryana

Himachal Pradesh

Jammu and Kashmir

KarnatakaKeralaMadhya Pradesh

Maharashtra

Orissa

PunjabRajasthan

Tamil NaduUttar Pradesh

West Bengal

All-India Rural 39.4 37.1 26.8 39.0 33.0 26.3Urban

Andhra Pradesh

AssamBiharGujarat

Haryana

Himachal PradeshJammu and Kashmir

KarnatakaKerala

Madhya Pradesh

Maharashtra

OrissaPunjab

RajasthanTamil Nadu

Uttar Pradesh

West BengalDelhi

All-India Urban 39.1 32.9 24.1 22.5 17.8 12.0

Source:Authors calculations based on NSS unit record data from 43rd, 50th and 55th Rounds.

21.039.4

53.9

28.6

15.416.725.9

32.629.542.0

41.0

58.712.8

33.346.3

41.9

48.8

15.9

45.2

58.0

22.2

28.330.430.4

30.125.440.7

37.9

49.811.7

26.435.9

42.3

41.2

10.5

40.3

44.0

12.4

7.47.54.7

16.89.4

37.2

23.2

47.86.0

13.520.0

31.1

31.7

35.0

36.1

54.6

39.413.6

13.315.3

40.823.843.7

44.3

50.46.6

35.3

49.0

34.9

36.3

29.2

35.4

48.6

32.517.0

17.110.1

37.919.536.6

42.9

43.56.2

23.0

38.5

28.6

25.1

26.2

35.5

41.120.0

5.7

9.86.1

30.710.0

31.3

31.9

43.02.4

17.3

24.3

21.521.9

41.1

11.351.9

38.518.4

7.215.049.2

39.8

47.340.342.6

13.7

37.9

40.2

44.933.7

15.1

38.8

7.934.8

28.316.5

9.39.3

39.9

24.3

48.135.040.6

10.9

31.0

39.935.122.9

16.1

27.2

7.533.5

14.810.0

4.62.0

24.6

19.8

38.526.743.5

5.5

19.4

22.530.814.7

9.2

23.4

13.638.1

16.411.8

1.73.8

26.0

21.0

20.721.220.8

6.6

19.8

26.229.322.3

4.7

17.8

13.026.7

14.710.5

3.63.1

21.4

13.9

18.518.215.2

7.8

18.3

20.821.715.5

8.8

10.8

11.824.7

6.44.6

1.21.3

10.8

9.6

13.912.015.6

3.4

10.8

11.317.311.3

2.4


4/20


the best currently available in terms ofdealing with the change in questionnairedesign, without pretending that they rep-resent the final word on the topic.

Turning to the price adjustments, onelimitation of the price indexes that havebeen traditionally used to update povertylines over time (e g, the Consumer PriceIndex for Agricultural Labourers) is that

they are based on fixed and frequentlyoutdated commodity weights. It is pos-sible to calculate alternative price indexesusing the information in the consumerexpenditure surveys themselves. For morethan 170 commodities, households reportboth quantities and expenditures, and theratio of the latter to the former providesan estimate of the price paid. These pricescan then be combined into consumer priceindex numbers that allow comparisonsacross states, and if we use data fromdifferent rounds, for states and the whole

country at different points in time. Onelimitation of these price indexes is thattheir coverage of commodities is onlypartial (a little more than half the budgetin the 55th Round, though more in earlierrounds), so that they cannot capture pricechanges in important items such as trans-portation, housing, most non-food goods,and services. However, CPIAL data sug-gest that the inflation rate for the uncov-ered items is not very different from thatapplying to the covered items.6 The priceindexes from the surveys have the advan-

tage of being based on several millionactual purchases in each round. They alsomake it possible to use formulas for su-perlative indexes, such as the Fisher idealindex or the Trnqvist index, that allowfor substitution behaviour as householdsadapt to relative price changes over time.

The calculated Trnqvist indexes for the43rd and 50th Rounds are reported inDeaton and Tarozzi (2000), and wereupdated to the 55th Round by Deaton(2001b).7 These price indexes differ fromthe official indexes in a number of ways.

In particular, they rise somewhat moreslowly over time than do the official priceindexes, especially in the rural sector. Forexample, the all-India rural Trnqvist indexrises by 69.8 per cent from 1987-88 to1993-94 and by a further 54.5 per centfrom 1993-94 to 1999-2000, comparedwith 78.7 per cent and 59.1 per cent forthe deflators implicit in the official all-India rural poverty line. For the urbansector over the two periods, the Trnqvistprice indexes rise by 73.8 and 57.7 per centversus 73.5 and 61.4 per cent for the implicit

deflator of the urban poverty line. Theprice indexes for each state show rathermodest differences from one state toanother. They also differ from those im-plicit in the official poverty lines, althoughthe two sets of deflators are correlated.This pattern is consistent with the factthat relative prices across states vary some-what over time, and that the interstate

prices used in the official deflators areoutdated.

The third departure concerns the gapbetween rural and urban poverty lines.From the mid-1970s until the early 1990s,there were only two poverty lines for India,one for rural and one for urban. The urbanline was around 15 per cent higher thanthe rural line, and both were held fixed in

Table 2b: State-Specific Poverty-Gap Indexes

Official Methodology Adjusted Estimates

1987-88 1993-94 1999-00 1987-88 1993-94 1999-00

RuralAndhra Pradesh 4.4 2.9 1.8 8.0 5.8 4.8Assam 7.4 8.3 8.5 6.5 5.7 6.1

Bihar 12.9 14.7 8.7 13.2 10.7 8.5

Gujarat 5.5 4.1 2.2 8.4 6.8 3.8

Haryana 3.6 5.6 1.3 2.8 3.0 0.7Himachal Pradesh 2.6 5.6 1.0 2.1 3.0 1.5

Jammu and Kashmir 4.5 5.6 0.6 2.4 1.6 0.7Karnataka 7.9 6.3 2.7 10.5 8.6 6.1

Kerala 6.4 5.6 1.5 4.8 3.9 1.7

Madhya Pradesh 10.6 9.5 7.7 11.2 8.2 6.6Maharashtra 9.6 9.3 4.4 10.8 11.2 7.6

Orissa 16.3 12.0 11.7 13.0 9.7 10.5

Punjab 2.0 1.9 0.8 1.0 1.0 0.3Rajasthan 8.6 5.2 2.1 9.2 4.4 3.0Tamil Nadu 12.6 7.3 3.8 13.7 9.1 4.6

Uttar Pradesh 9.9 10.4 5.8 7.5 5.8 3.9

West Bengal 11.6 8.3 6.5 7.7 4.2 3.5

All-India Rural 9.4 8.4 5.2 9.2 7.0 5.2UrbanAndhra Pradesh 10.6 9.3 5.6 4.9 3.4 1.9

Assam 1.5 0.9 1.5 2.0 2.0 1.9

Bihar 13.0 7.9 6.7 8.2 5.6 5.0

Gujarat 8.2 6.2 2.4 2.8 2.6 1.0Haryana 3.6 3.0 2.0 2.3 1.9 0.7

Himachal Pradesh 0.7 1.2 0.6 0.2 0.5 0.2Jammu and Kashmir 2.4 1.2 0.2 0.5 0.5 0.2

Karnataka 14.1 11.4 5.6 5.7 4.5 2.1Kerala 10.4 5.5 3.9 4.5 2.7 1.7

Madhya Pradesh 13.6 13.4 9.5 4.1 3.5 2.6

Maharashtra 12.3 10.1 6.7 5.3 4.6 2.8

Orissa 11.1 11.4 11.1 4.2 3.0 3.0Punjab 2.3 1.7 0.6 1.0 1.1 0.4Rajasthan 9.6 7.0 3.4 4.0 3.2 1.7

Tamil Nadu 11.5 10.2 4.8 6.2 4.5 2.0

Uttar Pradesh 12.2 9.0 6.6 6.3 4.6 3.3

West Bengal 7.4 4.5 2.5 4.2 2.9 1.9Delhi 2.8 3.9 1.5 0.7 1.7 0.4All-India Urban 10.4 8.3 5.2 4.8 3.7 2.3

Notesto Table 2:

Table 2a. The headcount ratios labelled official methodology are computed from the unit recorddata using the official poverty lines, as well as the official procedures for assigning poverty rates (or

poverty lines) to small states. We have also followed the official treatment of Jammu and Kashmir.

The all-India poverty rates are computed by adding-up the number of poor in each state and dividingby the total population. Because the Planning Commission uses interpolation rather than computationsfrom the unit record data, there are minor differences between these numbers and those published in

the official releases. The adjusted estimates are computed as described in the text (and more fully in

Deaton and Tarozzi, 2001, and Deaton, 2001b); they use price indexes computed from the unit record

data, and correct for the changes in questionnaire design in the 55th Round. The final column is asomewhat refined version of the corresponding column in Deaton (2001b). The estimates for Jammuand Kashmir are calculated directly, and not by assuming the poverty line or poverty rate for any

other state (as in the official methodology).

Table 2b. The poverty-gap indexes labelled official methodology are computed from the unit record

data using the official poverty lines, and using rules for assigning poverty-gap indexes to small states(and to J and K) that mirror the rules used by the Planning Commission for computing the official

headcount ratios. The adjusted indexes use the recomputed price indexes to update the poverty lines,and correct for the changes in questionnaire design in the 55th Round. All numbers are directly

computed from poverty lines and unit record data for each state, and the all-India estimates arecalculated as weighted averages of the state estimates.


5/20


real terms, with updating on the basis ofapproximate price indexes such as theWholesale Price Index or the CSOs pri-vate consumption deflator. The initial rural-urban gap of 15 per cent is anchored in1973-74 calorie consumption data, but itis essentially arbitrary since the urban andrural calorie norms themselves (2,100and 2,400 calories per person per day,

respectively) have a fragile basis.8 Morerecently, the Planning Commission hasadopted a modified version of the povertylines recommended by a 1993 Expert Group[Government of India 1993]. The ExpertGroup retained the original rural and urbanlines, but adjusted them for statewisedifferences in price levels, separately forurban and rural sectors, using estimates ofstatewise price differences calculated fromNSS data on expenditures and quantitiesusing similar methods to those adopted inthis paper. The Expert Group lines used

the then best-available information on pricedifferences across states, both urban andrural, but the information was outdated,especially for the rural sector.

Because the statewise adjustments weredone separately for urban and rural house-holds, the price differences between theurban and rural sectors of each state werederived only implicitly, and some are ratherimplausible, particularly the very muchhigher urban than rural lines in severalstates. For example, the most recent urbanpoverty lines for Andhra Pradesh and

Karnataka are around 70 per cent higherthan the corresponding rural lines, with theuncomfortable result that urban poverty ismuch higher than rural poverty in thesetwo states (see Table 2 in the next section).In Assam, by contrast, the rural povertyline is actually higherthan the urban line,and based on these odd poverty lines, Assamturns out to be one of Indias highest-poverty states for rural areas but lowest-poverty states for urban areas. It is hardto accept these and other implications ofthe Expert Group poverty lines.

There are grounds, of course, for ques-tioning whether it is even possible to derivecomparable rural and urban poverty lines.Comparisons of living standards in ruraland urban areas are inherently difficult,since there are large intersectoral differ-ences not only in the patterns of consump-tion but also in lifestyles, public amenities,epidemiological environments, and so on.One way forward is to avoid such com-parisons altogether, and to focus on sector-specific (rural or urban) poverty estimates.Yet there is a case for attempting to

compare private consumption levels acrosssectors, bearing in mind that this is at besta partial picture of the relevant differencesin living standards.9 These comparisonscan be made by anchoring poverty estima-tes in a single poverty line, adjusted whereappropriate to take into account rural-urbanprice differences, using the same methodas that described earlier for adjusting

poverty lines over time and between states.Based on this procedure, the urban povertyline tends be about 15 per cent higher thanthe rural poverty line, though there arevariations across states. As it turns out, thisrural-urban difference in poverty lines isbroadly consistent with the originalmethodology used before the adoption ofthe Expert Group recommendations.

To recapitulate, the revised poverty linesused in this paper, which are presented infull in Table 4 of Deaton (2001b), arederived as follows. Our starting point is

the official rural all-India poverty line forthe 43rd Round (1987-88): 115.70 rupeesper person per month.10 Rural povertylines for each state for the 43rd Round areobtained by multiplying this base povertyline by the rural price indexes for each staterelative to all-India. The urban povertylines for the 43rd Round, for each state aswell as for all-India, are calculated fromthe rural poverty lines by scaling up by therespective urban relative to rural priceindexes. In all cases, we use the relevantTrnqvist price indexes.11 To move to the

50th Round, the original all-India ruralline, 115.70 rupees, is scaled up by theTrnqvist index for all-India rural for the50th Round relative to the 43rd Round,1.698, to give an all-India rural povertyline for the 50th Round. This number isthen used to generate rural and then urbanpoverty lines for each state, followingexactly the same procedure as for the 43rdRound. Finally, poverty lines for the 55thRound are calculated in the same way froman all-India rural line, which is the 50thRound all-India rural line scaled up by the

value of the Trnqvist index between thetwo surveys.12

The motivation for the fourth departure(or rather extension) arises from the limi-tations of the headcount ratio as an indi-cator of poverty. The headcount ratio hasa straightforward interpretation and is easyto understand. In that sense it has muchcommunication value. Yet, the HCR hasserious limitations as a poverty index. Forone thing, it ignores the extentto whichdifferent households fall short of thepoverty line. This leads to some perverse

properties. For instance, an income trans-fer from a very poor person to someonewho is closer to the poverty line may leadto a decline in the headcount ratio, ifit lifts the recipient above the povertyline. Similarly, if some poor householdsget poorer, this has no effect on theheadcount ratio.

A related issue is that changes in HCRs

can be highly sensitive to the number ofpoor households near the poverty line (sincechanges in the HCR are entirely driven bycrossings of the poverty line). If poorhouseholds are heavily bunched near thepoverty line, a small increase in averageper capita income could lead to a mislead-ingly large decline in the headcount ratio.This density effect has to be kept firmlyin view in the context of comparisons ofpoverty change, involving questions suchas has there been more poverty declinein Bihar than in Punjab during the nine-

ties, or has poverty declined faster in thenineties than in the eighties? Often suchquestions are answered by looking at, say,the respective changes (absolute or pro-portionate) in headcount ratios. Thesechanges, however, are difficult to interpretin the absence of further information aboutthe initial density of poor households nearthe poverty line in each case.

One way forward is to use more sophis-ticated poverty indexes such as the Foster-Greer-Thorbecke (FGT) indexes or the Senindex. In this paper, we focus on the sim-

plest member of the FGT class (other thanthe headcount ratio itself), the poverty-gap index. Essentially, the poverty-gapindex (hereafter PGI) is the aggregateshortfall of poor peoples consumptionfrom the poverty line, suitably normal-ised.13 The PGI can also be interpreted asthe headcount ratio multiplied by the meanpercentage shortfall of consumption fromthe poverty line (among the poor). Thisindex avoids the main shortcomings of theheadcount ratio, is relatively simple tocalculate, and has a straightforward inter-

pretation.14

I.3 Adjusted Estimates

Table 1a presents official and adjustedestimates of the all-India headcount ratio.In each panel, the first row gives the officialestimates; the second row retains the of-ficial poverty lines but adjusts the 1999-2000 estimates for the change in question-naire design in the way described earlier;the third row gives fully-adjusted povertyestimates, which combine the adjustments


6/20


for questionnaire design and for price in-dexes. Table 1b gives the correspondingpoverty-gap indexes.

As the first two rows of each panelindicate, the official estimates are quitemisleading in their own terms: the 1999-2000 poverty estimates are biased down-ward by the changes in questionnairedesign. For headcount ratios, the estimates

adjusted for changes in questionnairedesign confirm about two-thirds of theofficial decline in rural poverty between1993-94 and 1999-2000, and about 90 percent of the decline in urban poverty. Forpoverty-gap indexes, the correspondingproportions (62 per cent and 77 per cent)are lower, especially for the urban sector.

The fully-adjusted estimates in the lastrow of each panel show somewhat lowerrural poverty estimates and much lowerurban poverty estimates for 1999-2000than even the official estimates. Note,

however, that because we are recalculatingthe poverty lines back to the 43rd Round,a good deal of the decrease took place inthe six years prior to 1993-94, not only inthe six years subsequent to 1993-94. Thefully-adjusted estimates for the headcountratios and poverty gap indexes suggest thatpoverty decline has been fairly evenlyspread between the two sub-periods (be-fore and after 1993-94), in contrast withthe pattern of acceleration in the secondsub-period associated with the officialestimates.

The rural-urban gaps in the povertyestimates are also of interest. Looking firstat the base year (1987-88), the rural-urbangap based on adjusted estimates is muchlarger than that based on official estimates.Indeed, the latter suggest no differencebetween rural and urban poverty in thatyear. This is hard to reconcile with inde-pendent evidence on living conditions inrural and urban areas, such as a life-expectancy gap of about seven years infavour or urban areas around that time.15

Our low estimate of the urban headcount

ratio relative to the official estimate (22.5per cent versus 39.1 per cent), and similardifferences in 1993-94 and 1999-2000,come from the fact that we take the ruralpoverty line in 1987-88 as our startingpoint, and peg the urban poverty linesabout 15 per cent higher than the ruralpoverty lines, in contrast to the much largerdifferentials embodied in the official lines.

Figure 1 shows the new estimates of theheadcount ratios together with the officialestimates going back to 1973-74. The fullyadjusted figures are lower throughout

because we treat the rural poverty line inthe 43rd Round as our baseline so that,with larger rural-urban gaps in the povertyestimates, we estimate lower povertyoverall. If instead, we had taken the urbanpoverty line as base, the adjusted figureswould have been higher than the officialfigures. From 1987-88 to 1993-94, theadjusted headcount ratio falls more rapidlythan the official headcount; this is becauseour price deflators are rising less rapidlythan the official ones. From 1993-94, the

adjusted figures fall more slowly becausethe effects of the price adjustment are morethan offset by the correction for question-naire design. The estimates for the thinrounds which look very different areincluded to remind us of the residualuncertainty about our conclusions, andwill be discussed further in Section IV.3below.

I.4 Regional Contrasts

State-specific headcount ratios are pre-

sented in Table 2a.16 The table has thesame basic structure as Table 1a, exceptthat we jump straight from official to fully-adjusted estimates. The latter suggest thatthe basic pattern of sustained povertydecline between 1987-88 and 1999-2000,discussed earlier at the all-India level, alsoapplies at the level of individual states inmost cases. The main exception is Assam,where poverty has stagnated in both ruraland urban areas. In Orissa, there has beenvery little poverty decline in the secondsub-period, with the result that Orissa now

has the highest level of rural poverty amongall Indian states, according to the adjusted1999-2000 estimates.17 Reassuringly, theanomalies noted earlier with respect torural-urban gaps in specific states tend todisappear as one moves from official toadjusted estimates.

Table 2b shows the corresponding pov-erty-gap indexes. The general patterns arevery much the same as with headcountratios; indeed the PGI series are highlycorrelated with the corresponding HCR

series, with correlation coefficients of 0.98for rural and 0.95 for urban. Even the HCRand PGI changes between the 50th and55th Rounds are highly correlated; thecorrelation coefficient between changes inHCR and changes in PGI is 0.95 for therural sector, and 0.96 for the urban sector.Thus, in spite of its theoretical superiorityover the headcount ratio, the poverty-gapindex gives us very little additional insightin this case.

In interpreting and comparing povertydeclines over time, it is useful to supple-

ment the poverty indexes with informationon the growth rate of average per capitaconsumption expenditure (hereafterAPCE). State-specific estimates of APCEgrowth between 1993-94 and 1999-2000are shown in Table 3, where states areranked in ascending order of APCE growthfor rural and urban areas combined.18

Here, a striking regional pattern emerges:except for Jammu and Kashmir, the low-growth states form one contiguous regionmade up of the eastern states (Assam,Orissa and West Bengal), the so-called

Figure 1: Official and Adjusted Headcounts Ratios

Source: Planning Commission, Press Releases ( March 11, 1997, and February 22, 2001), Deaton (2001a,b), and Table 2a of this paper.

1970 1980 1990 2000

Year

60

50

40

30

20

Headcountratio

(percent)

Official counts

Thin rounds

Adjusted counts


7/20


BIMARU states (Bihar, Madhya Pradesh,Rajasthan and Uttar Pradesh), and AndhraPradesh. The high-growth states, for their

part, consist of the southern states (exceptAndhra Pradesh), the western states(Gujarat and Maharashtra) and the north-western region (Punjab, Haryana andHimachal Pradesh). Further, it is interest-ing to note that this pattern is reasonablyconsistent with independent data on growthrates of per capita state domestic product(SDP); these are shown in the last columnof Table 3. With a couple of exceptionson each side, all the states in the lowAPCE growth set had comparatively lowrates of per capita SDP between 1993-94

and 1999-2000 (say below 4 per cent peryear), and conversely, all the states in thehigh APCE growth set had compara-tively high annual growth rates of percapita SDP (the correlation coefficientbetween the two series is 0.45).19

This broad regional pattern is a matterof concern, because the low-growth statesalso tend to be states that started off withcomparatively low levels of APCE or per-capita SDP. In other words, there has beena growing divergence of per capita ex-penditure (and also of per capita SDP)

across Indian states in the nineties.20 Thepoint is illustrated in Figure 2, which plotsthe average growth in APCE for each state

between 1993-94 and 1999-2000 againstthe geometric mean of APCE in 1993-94.

It is worth asking to what extent theseregional patterns, based on APCE data, arecorroborated by regional patterns of pov-erty decline. One difficulty here is thatthere is no obvious way of comparing theextent of poverty decline across states. Forinstance, looking at absolute changes in(say) HCRs would seem to give an unfairadvantage to states that start off withhigh levels of poverty, and where theretends be a large number of households

close to the poverty line. To illustrate, theabsolute decline of the rural HCR between1993-94 and 1999-2000 was about twiceas large in Bihar (7.4 percentage points)as in Punjab (3.8 points), yet over the sameperiod APCE grew by only 6.9 per centin Bihar compared with 20.2 per centin Punjab, with virtually no change indistribution in either case.21 The reasonfor this contrast is that Bihar starts off in1993-94 with a very high proportion ofhouseholds close to the poverty line, sothat small increases in APCE can produce

relatively large absolute declines in theheadcount ratio.

An alternative approach is to look atproportionate changes in HCRs or PGIs.These turn out to be highly correlated withthe corresponding growth rates of APCE.The point is illustrated in Figure 3, wherewe plot the proportionate decline in therural headcount ratio in each state against

the growth rate of APCE in rural areas. Thecorrelation coefficient between the twoseries is as high as 0.91. This reflects thefact that poverty reduction is overwhelm-ingly driven by the growth rate of APCE,rather than by changes in distribution weshall return to this point in Section III.From these observations, it follows that ifwe accept proportionate change in HCR(or PGI) as an index of poverty reduction,then the broad regional patterns identifiedearlier for the growth rate of APCE alsotend to apply to poverty reduction. In

particular: (1) most of the western andsouthern states (with the important excep-tion of Andhra Pradesh) have done com-paratively well; (2) the eastern region hasachieved very little poverty reductionbetween 1993-94 and 1999-2000; and(3) there is a strong overall pattern of di-vergence (states that were poorer to startwith had lower rates of poverty reduction).This reading of the evidence, however,remains somewhat tentative, since there isno compelling reason to accept the pro-portionate decline in HCR (or PGI) as a

definitive measure of poverty decline.We end this section with a caveat. From

Table 2 and Figure 1, it may appear thatthe pace of poverty decline in the ninetieshas been fairly rapid. It is important tonote, however, that the associated increasesin per capita expenditure have been rathermodest in most cases. For instance, thedecline of 6.6 percentage points in the all-India HCR (from 29.2 per cent to 22.7 percent) between 1993-94 and 1999-2000 isdriven by an increase of only 10.9 per centin average per capita expenditure not

exactly a spectacular improvement inliving standards. Similarly, Table 2a sug-gests that Bihar achieved a large step inpoverty reduction in the nineties, with therural HCR coming down from 49 per centto 41 per cent. Yet, as Table 3 indicates,average APCE in rural Bihar increasedby only 7 per cent between 1993-94 and1999-2000.

Why are small increases in APCE asso-ciated with substantial declines in povertyindexes? It is tempting to answer that thedistribution of consumer expenditure

Figure 2: Divergence of Per Capita Expenditure Across States, 1993-94 to 1999-00

Source: Authors calculations using unit record data from the 50th and 55th Rounds of the National Sample

Survey.

AI = All India; AP = Andhra Pradesh; AS = Assam; BI = Bihar; DE = Delhi; GU = Gujarat; HA = Haryana;HP = Himachal Pradesh; JK = Jammu and Kashmir; KA = Karnataka; KE = Kerala; MA = Maharashtra;MP = Madhya Pradesh; OR = Orissa; PU = Punjab; RA = Rajasthan; TN = Tamil Nadu; UP = Uttar Pradesh;

WB = West Bengal.

Geometric mean PCE in 1993-94, rupees per head

Averageannualgrowth

(percent)

200 300 400 500 600

6

4

2

0

BI

OR

MPUP

AS

WB

AP

AI

KA

TN

MA

GU

HP

KE PU

HA

DE

RA

JK


8/20


must have improved in the nineties. Asdiscussed in Section III, however, this isnot the case: indeed economic inequalityhas increased rather than decreased in thenineties. The correct answer relates to thedensity effect mentioned earlier (seeSection I.2): when many poor householdsare close to the poverty line, modest in-creases in APCE can produce substantial

declines in standard poverty indexes. Onereason for drawing attention to this is thatthe official poverty estimates have some-times been used to claim that the ninetieshave been a period of spectacular achieve-ments in poverty reduction. In fact, whenthe relevant adjustments are made, andthe poverty indexes are read together withthe information on APCE growth, povertyreduction in the nineties appears to bemore or less in line with previous rates ofprogress.

IIFurther Evidence

II.1 National Accounts Statistics

There has been much discussion of theconsistency between National SampleSurvey data and the national accountspublished by the Central StatisticalOrganisation (CSO).22 The latter includeestimates of private final consumer ex-penditure, which is frequently comparedwith NSS estimates of household

consumption expenditure. Over time, theCSO estimates have tended to grow fasterthan the NSS estimates, leading some com-mentators to question the reliability ofNational Sample Survey data.

It is important to note that these twonotions of consumer expenditure are not

exactly the same, and also that there aremajor methodological differences betweenthe two sources. The NSS figures are directestimates of household consumption ex-penditure. The CSO figures include sev-eral items of expenditure that are notcollected in the NSS surveys; examples are

expenditures by non-profit entreprises, aswell as imputed rent by owner occupiersand financial intermediation servicesindirectly measured (the last item is es-sentially the net interest earned by finan-cial intermediaries, which is counted asexpenditures on intermediation servicesby households). According to Sundaramand Tendulkar (2002), who quote a recentcross-validation study by the NationalAccounts Department, the last two itemsaccount for 22 per cent of the differencein levels between CSO and NSS estimates

of consumer expenditure. Further, the CSOestimates are residual figures, obtainedafter subtracting other items from thenational product. Leaving aside thesecomparability issues, there is indeed a gapbetween the CSO-based and NSS-basedgrowth rates of consumer expenditure.According to CSO data, per-capita con-sumer expenditure has grown at muchthe same rate as per capita GDP between1993-94 and 1999-2000 about 3.5 percent per year in real terms.23 The cor-responding NSS-based estimate associated

Table 3: Growth Rates of APCE and Per Capita SDP, 1993-94 to 1999-2000

Six-Year Growth of APCE (Adjusted), 1993-94 to 1999-2000 Annual Growth Rate ofRural Urban Combined Per Capita SDP,

1993-94 to 1999-2000

Assam 0.9 8.8 1.7 0.58

Orissa 1.4 0.0 3.3 2.34

West Bengal 2.1 11.5 3.3 5.48Jammu and Kashmir 5.4 8.0 5.3 2.49Bihar 6.9 4.8 7.1 2.10

Madhya Pradesh 6.6 14.1 7.8 2.78

Andhra Pradesh 2.8 18.5 8.3 3.57

Rajasthan 7.0 15.4 8.6 4.60Uttar Pradesh 8.3 10.1 9.0 2.99

India 8.7 16.6 10.9 4.36

Karnataka 9.5 26.5 14.0 5.82Maharashtra 14.1 16.7 15.9 3.53Gujarat 15.1 20.9 16.8 4.88

Himachal Pradesh 16.2 28.5 17.6 5.06

Tamil Nadu 15.7 25.1 18.9 5.39

Kerala 19.6 18.2 19.6 4.01Punjab 20.2 17.9 19.9 2.74Haryana 31.0 23.0 29.2 3.05

Delhi 30.7 30.7 5.69

Note: The states are arranged in ascending order of the growth rate of APCE for rural and urban areacombined.

Sources:For APCE: Authors calculations from unit record data for the 50th and 55th Rounds of the NationalSample Survey. For SDP: Authors calculations based on unpublished data kindly supplied by the

Planning Commission. The figures in the last column should be taken as indicative, given the

significant margin of error involved in SDP estimates.

Figure 3: HCR Declines and APCE Growth, 1993-94 to 1999-2000

(Rural)

Six-year growth of rural APCESource: Tables 2a and 3.

0 5 10 15 20 25 30

70

60

50

40

30

20

10

0

10

Proportionatedeclineinruralheadcou

ntratio

HA

PU

KEHP

TN

GU

MA

JK

KA

UPRA

INDIA

BI

MP

AP

WB

OR

AS


9/20


with our adjusted APCE figures is around2 per cent.24

This is quite different from the situationthat prevailed prior to the 55th Round,when consumer expenditure was hardlygrowing at all according to the NSS thinrounds but galloping forward accordingto the CSO data. Today, in the light of morerecent estimates, the discrepancy looks

much smaller. That discrepancy calls forfurther scrutiny and resolution, but mean-while, it can hardly be regarded as anindictment of National Sample Survey data.For one thing, the reference categories arenot the same. For another, there is noreason to believe that the CSO estimatesare more accurate than the NSS estimates;indeed the cross-validation exercise raisedserious questions about a number of theconsumption categories in the CSO data.25

II.2 Agricultural Wages

Agricultural wages provide an impor-tant source of further information onpoverty. There are, in fact, two ways ofthinking about the relevance of this infor-mation. First, real agricultural wages arehighly correlated with standard povertyindexes such as headcount ratios: wherepoverty is higher, wages tend to be lower,and vice versa. Based on this statisticalassociation, real wages can be used toprovide some information about otherpoverty indexes. Second, it is also possible

to think about the real wage as a roughpoverty indicator in its own right. The ideais that, if the labour market is competitive(at least on the supply side), then the realwage measures the reservation wage, i e,the lowest wage at which labourers areprepared to work. This has direct eviden-tial value as an indication of the deprivedcircumstances in which people live (themore desperate people are, the lower thereservation wage), independently of theindirect evidential value arising from thestatistical association between real wages

and standard poverty indexes such as theheadcount ratio.

Detailed information on agriculturalwages is available fromAgricultural Wagesin India (AWI), an annual publication ofthe Directorate of Economics and Statis-tics, Ministry of Agriculture. The datainitially come in the form of district-specific money wages.26 These are typi-cally aggregated using the numbers of agri-cultural labourers in different districts asweights, and deflated using the ConsumerPrice Index for Agricultural Labourers

(CPIAL). The quality of this informationis not entirely clear, but available evidencesuggests that it is adequate for the purposeof broad comparisons.27

As Figure 4 illustrates, real agriculturalwages in different states are highly cor-related with expenditure-based povertyindexes (here and elsewhere in this sec-tion, the focus is on rural poverty). The

main outlier is Kerala, where real wagesare far above the regression line; it seemsthat the power of labour unions in Keralahas raised agricultural wages well abovethe level found in any other Indian states,but that this does not translate into acorrespondingly low level of rural poverty,possibly because high wages are partlyoffset by high unemployment, or becauseother determinants of rural poverty are alsoat work. In 1999-2000, the correlationcoefficient between real wages and

headcount ratios in different states was0.79 in absolute value, rising to 0.91 ifKerala is excluded. In 1993-94, the cor-relation coefficient was 0.87 in absolutevalue, with or without Kerala. Interest-ingly, if official HCRs are used insteadof our adjusted HCRs, the correlationcoefficients come down quite sharply (e g,from 0.91 to 0.73 in 1999-2000 and from

0.87 to 0.54 in 1993-94, without Keralain both cases). This can be tentativelyregarded as a further indication of theplausibility of the proposed adjustments.

Given the close association between realwages and rural poverty, the growth ratesof real wages over time provide usefulsupplementary evidence on poverty trends.According to recent estimates based onAWI data, real agricultural wages weregrowing at about 5 per cent per year in theeighties and 2.5 per cent per year in the

Table 4: Growth and the Headcount Ratio, 1993-94 to 1999-2000HCR50 Derivative with Six Years Change in Change in

Respect to Growth Growth HCR55 HCR55,

Inequality Fixed Actual

RuralAndhra Pradesh

AssamBihar

GujaratHaryana

Himachal Pradesh

Jammu and KashmirKarnataka

KeralaMadhya Pradesh

Maharashtra

Orissa

PunjabRajasthan

Tamil Nadu

Uttar Pradesh

West Bengal 25.1 2.0 3.2

All-India 33.0 0.88 8.7 6.8 6.7UrbanAndhra Pradesh

Assam

BiharGujarat

Haryana

Himachal PradeshJammu and Kashmir

Karnataka

KeralaMadhya Pradesh

Maharashtra

OrissaPunjab

Rajasthan

Tamil Nadu

Uttar PradeshWest Bengal

DelhiAll-India 17.8 0.56 16.6 7.4 5.9

Source: Authors calculations from the unit record data of the 43rd , 50th , and 55th Rounds of the NSS.Note that the hypothetical all-India figures are calculated on the counterfactual asssumption thateach household received the state growth rate. They therefore do not show what would havehappened had growth been more equally distributed across the states: see the text for thisalternative calculation.

29.2

35.448.6

32.517.0

17.1

10.137.9

19.536.6

42.9

43.5

6.223.0

38.5

28.6

0.90

1.271.06

0.910.63

0.75

0.500.91

0.620.93

0.81

1.04

0.340.78

0.90

0.79

0.97

2.8

0.96.9

15.131.0

16.2

5.49.5

19.66.6

14.1

1.4

20.27.0

15.7

8.3

2.1

2.5

1.48.2

12.112.9

8.3

2.69.0

10.3

6.5

10.9

1.2

4.05.5

13.3

6.6

3.0

0.17.4

12.411.3

7.3

4.07.29.5

5.3

11.0

0.5

3.85.7

14.1

7.2

17.8

13.026.7

14.7

10.53.6

3.121.4

13.918.5

18.215.2

7.8

18.3

20.8

21.715.5

8.8

0.62

0.640.79

0.55

0.470.26

0.150.60

0.460.63

0.450.54

0.38

0.59

0.66

0.590.560.26

18.5

8.84.8

20.9

23.0

28.5

8.026.5

18.214.1

16.70.0

17.9

15.4

25.110.111.5

30.7

9.0

3.14.0

8.7

6.32.90.4

12.9

7.18.0

6.10.1

4.9

8.4

12.96.0

5.85.7

6.9

1.22.0

8.3

6.02.41.8

10.6

4.24.6

6.20.4

4.4

7.5

9.64.4

4.36.4


10/20


nineties.28 Thus, real agricultural wageswere growing considerably faster in theeighties than in the nineties. But even thereduced growth rate of agricultural wagesin the nineties, at 2.5 per cent per year,points to significant growth of per capitaexpenditure among the poorer sections ofthe population and reinforces our earlierfindings on poverty reduction. In fact, this

reduced growth rate is a little higherthanthe growth rate of average per capitaexpenditure (1.5 per cent per year) thatsustains our estimated declines of ruralheadcount ratios and headcount indexesbetween 1993-94 and 1999-2000.

The data on real wages also providesome independent corroboration of thestate-specific patterns of poverty decline.This is illustrated in Figure 5, where weplot state-specific estimates of the growthrate of real agricultural wages in the nine-ties against the estimated proportionate

decline in the headcount ratio (a very similarpattern applies to the poverty-gap index).Here the two main outliers are Punjab andHaryana, where the headcount ratio hasdeclined sharply without a correspond-ingly sharp increase in real wages (indeedwithout any such increase, in the case ofPunjab). Leaving out these two outliers,the association between the two series isremarkably close (with a correlation co-efficient of 0.88).

An interesting sidelight emerging fromFigure 5 is that a healthy growth of real

agricultural wages appear to be a suffi-cient condition for substantial povertydecline in rural areas: all the states wherereal wages have grown at more than, say,2.5 per cent per year in the nineties haveexperienced a comparatively sharp reduc-tion of the rural headcount ratio. Con-versely, in states with low rates of reduc-tion of the headcount ratio (say, 15 per centor less over six years), real wages haveinvariably grown at less than 2 per centper year. This applies in particular to theentire eastern region (Assam, Orissa, West

Bengal and Bihar) and also to AndhraPradesh and Madhya Pradesh.

Independent evidence on the growth ratesof real wages has recently been presentedby K Sundaram (2001a, 2001b), based onthe employment-unemployment surveys(EUS) of the National Sample Survey for1993-94 and 1999-2000. For the presentpurpose, these surveys are comparable.Sundaram estimates that the real earningsof agricultural labourers have grown atabout 2.5 per cent per year between 1993-94and 1999-2000. These are tentative

estimates, based as they are on data for twoyears only. Yet it is reassuring to find thatthey are consistent with the AWI-basedestimates.

II.3 The Employment-Unemployment Surveys

The National Sample Surveys 1993-94

and 1999-2000 employment-unemploy-ment surveys (EUS) also include consumerexpenditure data. These can be used forfurther scrutiny of poverty trends. Thistask has been undertaken in a recent paperby Sundaram and Tendulkar (2002). Theynote that the consumption survey in the1999-2000 EUS uses the traditional 30-day reporting period, but differs from thestandard questionnaire by only asking anabbreviated set of questions. However, theauthors find that, in those cases where thequestions have comparable coverage, the

means from the EUS, using the traditional30-day reporting period, are typically closeto those from the 30-day questionnaire inthe main consumption survey. Based onthis correspondence, they argue that the30-day questions in the main 1999-2000survey were not much distorted by theseven-day questions that were asked along-side them. In this version of events, themajor source of incomparability betweenthe 55th and 50th Rounds is not the con-tamination of the 30-day questions, butrather the revised treatment of the low

frequency items, for which the reportingperiod was 30 days in the 50th Round and365 days in the 55th Round. As we havealready noted, the 365-day reporting pe-riod for these items pulls up the lower tailof the consumption distribution, and thusbiases down the headcount ratio comparedwith earlier methods. However, Sundaramand Tendulkar note that the 50th Round

contained both 30-day and 365-day report-ing periods for the low frequency items.Hence, by recalculating the 50th Roundheadcounts using the 365-day responses,they can put the 50th and 55th Rounds ona roughly comparable basis. When they dothis, they find that, in both rural and urbansectors, they can confirm a little more thanthree-quarters of the official decline in theheadcount ratios between the two rounds[Sundaram and Tendulkar 2002:Table III.8]. These calculations are notidentical to our first-step adjustments (see

Table 1a), but they are close enough toinspire some confidence that both sets ofresults are in the right range.

To sum up, the all-India poverty indexespresented earlier in this paper are broadlyconsistent with independent evidence fromthe national accounts statistics and theemployment-unemployment surveys, aswell as with related information on agri-cultural wages. There is also some congru-ence between the inter-state contrastsemerging from NSS data and independentinformation on state-specific growth rates

Figure 4: Agricultural Wages and Rural Poverty, 1999-2000

Source: Drze and Sen (2002), Statistical Appendix, Table A 3, and Table 2a of this paper. The realagricultural wage is a three-year average ending in 1999-2000.

0 2 3 4 5 6 7 8

Real agricultural wage

50

40

30

20

10

0

Ruralheadcountratio(adjusted)

OR

BI

AS

KA

MP

MA

APINDIA

UP

TN

WBGU

RA

PUHA

KE


11/20


of state domestic product and real agri-cultural wages. The combined evidencefrom these different sources is fairly strong,even though each individual source hassignificant limitations.

IIIEconomic Inequality in the

Nineties

III.1 Growth, Povertyand Inequality

It is possible to think about povertydecline, as captured by standard povertyindexes, in terms of two distinct compo-nents: a growth component and a distri-bution component. The growth compo-nent reflects the increase of average percapita expenditure. The distribution com-ponent captures any change that may takeplace in the distribution of per capita ex-penditure over households.

This decomposition exercise is pursuedin Table 4, with reference to the headcountratio (very similar results apply to thepoverty-gap index). The first column re-peats the headcount ratio for 1993-94from Table 2. The second column (labelledderivative with respect to growth) showsour estimate of the percentage-point re-duction in HCR associated with a distri-bution-neutral, 1 per cent increase in APCEin the relevant state. To illustrate, in ruralAndhra Pradesh a 1 per cent increase inAPCE in 1993-94, with no change in

distribution, would have led to a declineof 0.9 percentage points in the ruralheadcount ratio.29This derivative dependspositively on the fraction of people whoare at or near the poverty line, which istypically larger in the poorer states. Thefigures in column 2 vary from 1.27 inrural Assam to 0.15 in urban Jammu andKashmir. Column 3 reproduces the totalpercentage growth between 1993-94 and

1999-2000 from Table 3.If we multiply the second column (the

derivative with respect to growth) by thethird column (the amount of growth), weget an estimate of the amount of povertyreduction that we would expect from growthalone, in the absence of any change in theshape of the distribution. This is an ap-proximation, because the derivative is likelyto change as the headcount ratio falls. Incolumn 4, we report a more precise cal-culation: an estimate of what the headcountratio would have been in 1999-2000 if the

distributions of consumption in each statewere identical to those in 1993-94, but hadbeen shifted upwards by the amount ofgrowth in real per capita expenditure thatactually took place. This can be readilycalculated by reducing the 1993-94 pov-erty lines by the amount of growth, andre-estimating the headcount ratios fromthese adjusted lines and the 1993-94 ex-penditure data. These hypothetical changescan then be compared with the actualreductions in the headcount ratios, shownin the final column. The difference

between these last two columns is thechange in the headcount ratio that is at-tributable to changes in the shape of theconsumption distribution.

It is important to note that the last twocolumns are highly correlated. The corre-lation coefficients across the states are0.97 (rural) and 0.93 (urban), so that growthalone can predict much of the cross-state

pattern of reduction in HCRs. Nevertheless,the estimates are far from identical. Inparticular, the all-India calculations showthat growth alone would have reduced thepoverty rate by more than actually hap-pened, implying that there was an increasein inequality that offset some of the effectsof growth, or put differently, that APCEgrowth among the poor was less than theaverage. These inequality effects varysomewhat from state to state and are muchweaker in rural than in urban areas. Inurban India, increasing inequality moder-

ated the decline in the headcount ratio inall states except Delhi, Maharashtra, andJammu and Kashmir. In some cases, suchas urban Kerala and Madhya Pradesh, themoderating effect is pronounced, withactual rates of reduction only a little over halfthose predicted by the growth in the mean.

For the urban sector as a whole (the lastrow of the table), the actual decline in theHCR is one and a half points lower (5.9versus 7.4 per cent) than would have beenthe case had growth been equally distrib-uted within each state. This estimate, which

is the population-weighted average of thecorresponding numbers for each state,calculates what would have happened ifeach household in each state had experi-enced the average growth for that state. Analternative, and equally interesting, counter-factual is what would have happened if,between 1993-94 and 1999-2000, eachhousehold in the country had experiencedthe countrywide growth rate of 10.9 percent. Such a calculation yields an all-IndiaHCR of 21.4 per cent (for rural and urbanareas combined), compared with an actual

all-India HCR of 22.7 per cent based onthe 55th Round. In other words, the all-India HCR in 1999-2000 was 1.3 percent-age points higher than it would have been(with the same growth rate of APCE) inthe absence of any increase in inequality.

III.2 Aspects of Rising Inequality

Three aspects of rising economic in-equality in the nineties have come up sofar in our story. First, we found strongevidence of divergence in per capita

Figure 5: Wage Growth and Poverty Decline, 1993-94 to 1999-2000

Source:See Figure 4.

4 2 0 2 4 6 8 10

Growth rate of real agricultural wage

80

60

40

20

0

Proportionatedeclineinrura

lHCR PU

HA

KE

GU

TN

KA

RA

UP

INDIA

MA

WB

AP

MPBI

AS

OR


12/20


consumption across states. Second, ourestimates of the growth rates of per capitaexpenditure between 1993-94 and 1999-2000 (Table 3) point to a significant in-crease in rural-urban inequalities at the all-India level, and also in most individualstates. Third, the decomposition exercisein the preceding section shows that risinginequality within states, particularly in the

urban sector, has moderated the effects ofgrowth on poverty reduction.

Table 5 provides more systematic evi-dence on recent changes in consumptioninequality within each sector of each stateusing two different measures of inequality.We show the logarithm of the differenceof the arithmetic and geometric means(approximately the fraction by which thearithmetic mean exceeds the geometricmean), as well as the variance of thelogarithm of per capita expenditure.

The table shows that the correction for

questionnaire design is critical for under-standing what has been happening. (Notethat the correction for prices has no effectwithin sectors and states.) The direct useof the unit record data in the 55th Round,with no adjustment, shows a substantialreduction in inequality within the ruralsectors of most states, with little or noincrease in the urban sectors. With thecorrection, we see that within-state ruralinequality has not fallen, and that therehave been marked increases in within-stateurban inequality. We suspect that the main

reason why the unadjusted data are somisleading in this context is the changefrom 30 to 365 days in the reporting periodfor the low frequency items (durable goods,clothing and footwear, and institutionalmedical and educational expenditures). Thelonger reporting period actually reducesthe mean expenditures on those items, butbecause a much larger fraction of peoplereport something over the longer reportingperiod, the bottom tail of the consumptiondistribution is pulled up, and both inequal-ity and poverty are reduced. Whether 365-

days are a better or worse reporting periodthan 30-days could be argued either way,but the main point here is that the 55th and50th Rounds are not comparable, and thatthe former artificially shows too littleinequality compared with the latter. Oncethe corrections are made, we see that, inaddition to increasing inequality betweenstates, there has been a marked increasein consumption inequality within theurban sector of nearly all states.

Two further pieces of evidence areworth mentioning in this context. First,

our findings on rising economic inequalitywithin the urban sector are consistent withrecent work by Banerjee and Piketty (2001),who use income tax records to documentvery large increases in income among thevery highest income earners. They showthat, in the 1990s, real incomes among thetop one per cent of income earners in-creased by a half in real terms, while those

of the top 1 per cent of 1 per cent increasedby a factor of three in real terms.

Second, it is interesting to compare thegrowth rate of real wages for agriculturallabourers with that of public sector sala-ries. As we saw earlier, real agriculturalwages have grown at 2.5 per cent or soin the nineties. Public sector salaries, fortheir part, have grown at almost 5 per centper year during the same period.30 Giventhat public-sector employees tend to bemuch better off than agricultural labourers,this can be taken as an instance of rising

economic disparities between different

occupation groups. Since agriculturallabourers and public sector employeestypically reside in rural and urban areas,respectively, this finding may just beanother side of the coin of rising rural-urban disparities. Even then, it streng-thens the evidence presented earlier onaspects of rising economic inequality inthe nineties.

To sum up, except for the absence ofclear evidence of rising intra-rural in-equality within states, we find strong indi-cations of a pervasive increase in eco-nomic inequality in the nineties. This isa new development in the Indian economy:until 1993-94, the all-India Gini coeffi-cients of per capita consumer expenditurein rural and urban areas were fairly stable.31

Further, it is worth noting that the rate ofincrease of economic inequality in thenineties is far from negligible. For in-stance, the compounding of inter-state

divergence and rising rural-urban

Table 5: Inequality Measures

logAMlogGMa Variance of Logs

50th Round 55th Round 55th Round 50th Round 55th Round 55th RoundAdjusted Adjusted

Andhra Pradesh

Assam

Bihar

GujaratHaryana

Himachal Pradesh

Jammu and Kashmir

Karnataka

KeralaMadhya PradeshMaharashtra

Orissa

PunjabRajasthan

Tamil Nadu

Uttar PradeshWest BengalAll-India Rural 0.14 0.11 0.14 0.23 0.21 0.24

Andhra Pradesh

Assam

Bihar

GujaratHaryana

Himachal Pradesh

Jammu and Kashmir

KarnatakaKerala

Madhya PradeshMaharashtra

Orissa

PunjabRajasthan

Tamil Nadu

Uttar PradeshWest BengalDelhi

All-India Urban 0.19 0.20 0.21 0.34 0.34 0.37

All-India 0.17 0.18 0.19 0.29 0.29 0.32

Note:a AM is the arithmetic mean and GM is the geometric mean: the difference in their logarithms is themean relative deviation, a measure of inequality.

0.14

0.050.08

0.100.16

0.13

0.100.12

0.150.13

0.16

0.10

0.130.12

0.16

0.13

0.11

0.09

0.07

0.07

0.090.10

0.10

0.06

0.10

0.140.100.11

0.10

0.100.07

0.14

0.100.09

0.13

0.060.08

0.110.23

0.14

0.070.12

0.160.12

0.16

0.12

0.140.10

0.15

0.12

0.08

0.24

0.10

0.160.170.28

0.22

0.16

0.21

0.260.220.27

0.180.22

0.20

0.27

0.230.17

0.17

0.13

0.130.180.19

0.17

0.12

0.18

0.240.180.20

0.180.19

0.14

0.23

0.180.15

0.22

0.11

0.160.180.31

0.24

0.14

0.22

0.270.220.28

0.210.24

0.18

0.24

0.210.15

0.170.13

0.150.14

0.13

0.38

0.130.16

0.20

0.18

0.21

0.150.130.14

0.21

0.170.19

0.29

0.16

0.16

0.170.14

0.14

0.160.09

0.180.17

0.17

0.21

0.140.14

0.13

0.27

0.18

0.200.21

0.170.14

0.170.14

0.15

0.42

0.120.170.22

0.18

0.21

0.160.140.14

0.20

0.19

0.190.30

0.300.25

0.270.25

0.24

0.37

0.240.310.31

0.29

0.40

0.290.230.25

0.39

0.310.34

0.43

0.290.30

0.300.25

0.27

0.29

0.160.320.32

0.29

0.36

0.260.250.23

0.34

0.310.31

0.39

0.330.27

0.300.26

0.28

0.40

0.210.34

0.37

0.33

0.40

0.290.250.26

0.35

0.340.35

0.46


13/20


disparities produces very sharp contrastsin APCE growth between the rural sectorsof the slow-growing states and the urbansectors of the fast-growing states (Table 3).This is further compounded by the accen-tuation of intra-urban inequality, which isitself quite substantial, bearing in mindthat the change is measured over a shortperiod of six years (Table 5).

It might be argued that a temporaryincrease in economic inequality is to beexpected in a liberalising economy, andthat this trend is likely to be short-lived.

Proponents of the Kuznets curve mayeven expect it to be reversed in due course.However, Chinas experience of sharp andsustained increase in economic inequalityover a period of more than 20 years, aftermarket-oriented economic reforms wereinitiated in the late 1970s, does not inspiremuch confidence in this prognosis.32 It is,in fact, an important pointer to the pos-sibility of further accentuation of economicdisparities in India in the near future.

IV

Qualifications and Concerns

IV.1 Food Consumption

There have been major changes in Indiasfood economy in the nineties. The eightieswere a period of healthy growth in agri-cultural output, food production, and realagricultural wages. During the nineties,however, productivity increases sloweddown in many states. The quantity indexof agricultural production grew at a lame2 per cent per year or so. The growth of

real agricultural wages slowed down con-siderably. And cereal production barelykept pace with population growth.33

The virtual stagnation of per capita cerealproduction in the nineties has been accom-panied by a gradual switch from net importsto net exports, and also by a massiveaccumulation of public stocks. Correspond-ingly, there has been no increase in esti-mated per capita net availability of cereals(Table 6). If anything, net availabilitydeclined a little, from a peak of about 450grams per person per day in 1990 to 420

grams or so at the end of the nineties. Thisis consistent with independent evidence,from National Sample Survey data, of adecline in per capita cereal consumptionin the nineties. Between 1993-94 and 1999-2000, for instance, average cereal con-sumption per capita declined from 13.5 kgper month to 12.7 kg per month in ruralareas, and from 10.6 to 10.4 kg per monthin urban areas.34 This comparison isbased on the uncorrected 55th Rounddata, and the true decline may be largerstill, given the changes in questionnaire

design (Section I.1).The reduction of cereal consumption in

the nineties may seem inconsistent withthe notion that poverty has declined duringthe same period. Indeed, this pattern hasbeen widely invoked as evidence of im-poverishment in the nineties. If cereal con-sumption is declining, how can povertybe declining?

It is worth noting, however, that thedecline of cereal consumption is not new.A similar decline took place (according toNational Sample Survey data) during the

seventies and eighties, when poverty wascertainly declining. Hanchate and Dysons(2000) recent comparison of rural foodconsumption patterns in 1973-74 and1993-94 sheds some useful light on thismatter. As the authors show, during thisperiod per capita cereal consumption inrural areas declined quite sharply on av-erage (from 15.8 to 13.6 kgs per person

per month), but rose among the pooresthouseholds. The decline in the average isdriven by reduced consumption among thehigher expenditure groups.35

The average decline is unlikely to bedriven by changes in relative prices;indeed, there has been little change in foodprices, relative to other prices, in the inter-vening period. Instead, this pattern ap-pears to reflect a substitution away fromcereals to other food items as incomes rise(at least beyond a certain threshold). Theconsumption of superior food items such

as vegetables, milk, fruit, fish and meat didrise quite sharply over the same period,across all expenditure groups. Seen in thislight, the decline of average cereal con-sumption may not be a matter of concernper se. Indeed, average cereal consump-tion is inversely related to per capita in-come across countries (e g, it is lower inChina than in India, and even lower in theUnited States), and the same applies acrossstates within India (e g, cereal consump-tion is higher in Bihar or Orissa than inPunjab or Haryana).

Food intake data collected by the Na-tional Nutrition Monitoring Bureau(NNMB) shed further light on this issue.Aside from detailed information on foodintake, the NNMB surveys include roughestimates of household incomes. These areused in Figure 6 to display the relationbetween per-capita income and food in-take, for different types of food. The sub-stitution from cereals towards other fooditems with rising per-capita income emergesquite clearly.36 This pattern, if confirmed,would fit quite well with the data on change

over time.37

It also implies that the declineof average cereal consumption in the nine-ties is not inconsistent with our earlierfindings on poverty decline.38

IV.2 Localised Impoverishmentand Hidden Costs

The overall decline of poverty in thenineties does not rule out the possibilityof impoverishment among specific regionsor social groups. That possibility, of course,is not new, but it is worth asking whether

Source: Calculated from National Nutrition Monitoring Bureau (1999), Table 6.9. The data relate to rural

areas of eight sample states.

Figure 6: Food Intake for Different Per Capita Income Groups, as a Proportion(Per Cent) of Average Intake (1996-97)

Cereals and millets Fats and oils Sugar and Jaggery Milk and milk products

Per capita income groups (Rs/cap/month) less than 90 90-150

150-300 300-600 more than 600

250.0

200.0

150.0

100.0

50.0

0.0


14/20


its scope has expanded during the lastdecade. As the economy gives greater roomto market forces, uncertainty and inequal-ity often increase, possibly leading toenhanced economic insecurity among thosewho are not in a position to benefit fromthe new opportunities, or whose liveli-hoods are threatened by the changes in theeconomy. The increase of economic in-equality in the nineties, noted earlier,suggests that tendencies of this kind maywell be at work in India today. Adversetrends in living standards could take several

distinct forms, including: (1) impoverish-ment among specific regions or socialgroups, (2) heightened uncertainty ingeneral, and (3) growing hidden costs ofeconomic development.

In connection with the first point, wehave already noted that some of the poorerstates, notably Orissa and Assam, have notfared well at all in the nineties. It is quitepossible that the poorer regions withinthese states have done even worse, to thepoint of absolute impoverishment forsubstantial sections of the population. In

the case of Orissa, there is some indepen-dent evidence of localised impoverishmentin the poorer districts, due inter alia to thedestruction of the local environmental baseand to the dismal failure of state-spon-sored development programmes [Drze2001].39

Similarly, the overall improvement ofliving standards may hide instances ofimpoverishment among specific occupa-tion groups. The nineties have been a periodof rapid structural change in the Indianeconomy, leading in some cases to

considerable disruption of earlier liveli-hood patterns. Examples include a deeprecession in the powerloom sector, a se-rious crisis in the edible oil industry afterimport tariffs were slashed, periodic wavesof bankruptcy among cotton growers, thedisplacement of traditional fishing bycommercial shrimp farms, and a numberof sectoral crises associated with the abruptlifting of quantitative restrictions on im-ports in mid-2001.40 The destruction oflocal environmental resources is anothercommon cause of disrupted livelihoods in

many areas.A related issue is the possibility of hidden

hardships associated with recent patternsof economic development. To illustrate,there is much evidence that, in many ofthe poorer regions of India, further impov-erishment has been avoided mainly throughseasonal labour migration.41 The latteroften entails significant social costs that

are poorly captured, if at all, in standardpoverty indexes or for that matter in theother social indicators examined in thispaper. Examples of such costs includeirregular school attendance, the spread ofHIV/AIDS, the disruption of family life,and rising urban congestion.42 Similarly,involuntary displacement of persons af-fected by large development projects such

as dams and mines tends to have enormoushuman costs. These, again, are largelyhidden from view in income-based analy-ses of poverty. In fact, the incomes ofdisplaced persons often rise (with cashcompensation) even as their lives are beingshattered.43The informalisation of labourmarkets is another example of economicchange with substantial hidden costs (e g,longer working hours, higher insecurity,lower status, and deteriorating work con-ditions).44 These issues are not new, butit is important to acknowledge the possi-

bility that the hidden costs of economicgrowth have intensified in the nineties.

This acknowledgement helps to recon-cile the survey-based evidence reviewedearlier with widespread media reports, inrecent years, of sectoral economic crisesand localised impoverishment.45 Thisissue calls for further scrutiny, based onmore focused analysis of survey data aswell as on micro-studies.

IV.3 The Thin Rounds: AnUnresolved Puzzle?

We have so far said very little about thethin rounds, and the poverty estimatesthat can be calculated from them. YetFigure 1 shows that the recent thin rounds,from the 51st through the 54th Round,generate poverty estimates that are hard toreconcile with the quinquennial thickrounds. If we were to connect up these

Source: Drze and Sen (2002), chapter 9.

Table 6: Cereal Availability in the Nineties(Grams per person per day)

Net Production Net Imports Net Change in Net Availability

Public Stocks (1+2-3)1985-89 422.7 2.0 -5.3 430.11990 456.9 0.3 5.0 452.11991 447.9 -1.4 0.4 446.1

1992 446.8 1.3 4.2 443.8

1993 446.4 2.5 16.6 432.3

1994 456.3 0.2 16.6 439.9

1995 448.6 -5.9 -2.3 445.11996 451.7 -6.9 -11.7 456.4

1997 445.5 -6.7 -4.2 443.0

1998 455.3 -4.7 11.0 439.6

1999 456.3 -5.4 25.3 425.62000 452.7 -5.2 30.7 416.8

Note: All figures (except first row) are three-year averages centred at the year specified in the first column.Source: Calculated from Government of India (2002), p S-21.

Figure 7:Progress of Selected Social Indicators in the 1980s and 1990s(Per Cent Per Year)

Increase of real Decline of infant Decline of total Decline of

agricultural wages mortality rate fertility rate illiteracy rate

6.0

5.0

4.0

3.0

2.0

1.0

0.0

5.5

2.5

2.9

1.51.7 1.8

1.5

2.8

1980s

1990s


15/20


points with the official HCR estimates, wewould get a series in which poverty rosebetween 1993-94 and 1994-95, fell from1994-95 to the end of 1997, rose verysharply in the first half of 1998, and thenfell with extraordinary rapidity in 1999-2000. As we have seen, the official esti-mate for 1999-2000 is too low, and the lastthin round, the 54th Round, ran for only

the first six months of 1998, and maytherefore not be fully comparable withother rounds. Even so, and with due al-lowance for corrections, it is very hard tointegrate the poverty estimates based onthe thin rounds with the picture that emergesfrom the thick rounds as well as from othersources surveyed in this paper.

The story is further complicated by thefact that these thin rounds were run in twoversions, one of which resembled thestandard questionnaire up to and includingthe 50th Round, and one of which the

experimental questionnaire had differentreporting periods for different goods.Headcount ratios based on the experimen-tal questionnaire (not shown in Figure 1)are lower than those from the standardquestionnaire, because the experimentalquestionnaire generated higher reports ofper capita expenditure. However, they alsoshow rising HCRs from the 52nd throughthe 54th Rounds, and the increase contin-ues into the 55th Round if we use com-parable reporting periods from that round.Based on the experimental questionnaire,

a case could be made that the all-IndiaHCR has been rising since 1995-96 [Sen2000]. As we have seen, there are goodgrounds for distrusting the experimentalquestionnaire in the 55th Round, becauseof the juxtaposition of the seven-day recalland 30-day recall data for food-pan andtobacco. Quite likely, the reconciliationeffect (see Section I.1) pulled down theestimates of per capita expenditure fromthe experimental questionnaire, thus exag-gerating poverty by this count. Even so,if poverty were genuinely falling, there is

no obvious explanation why the experi-mental questionnaire should show a risein poverty from 1995 through 1998.

The Planning Commission has neverendorsed poverty counts from the thinrounds. In part, this has been because ofthe smaller sample sizes. The PlanningCommission needs estimates of HCRs, notjust for all-India, but for individual states,and the thin rounds are not large enough tosupport accurate estimates for some of thesmaller (of the major) states. But in-adequate sample size generates variance,

not bias, and in any case, the thin roundsample sizes are perfectly adequate togenerate accurate estimates for the all-India HCRs. The discrepancies in Figure 1cannot be explained by inadequatesample sizes.

There are other differences between thickand thin rounds. For example, the sam-pling frame for the 51st, 53rd, and 54th

Rounds was not the census of population,but the economic census. In the popu-lation census, each household is asked ifit has a family business or enterprise, andonly such households are included in thefirst-stage sampling from the economiccensus when first-stage units are drawnwith probability proportional to size. Thismeans that a village with few or no suchhouseholds has only a small or no chanceof being selected as a first-stage unit. Evenso, when the team reaches the village, allhouseholds are listed and have a chance

of being in the sample, so it is unclear thatthis choice of frame makes much differ-ence. Indeed, comparison of various socio-economic indicators (e g, literacy rates,years of schooling, landholding, or familysize) from the surveys suggests no obviousbreaks between the 51st and 53rd Roundson the one hand, and the 52nd Round(which used the population census) on theother. Conversations with NSS and Plan-ning Commission staff sometimes suggestthat there may be other (non-documented)differences in the sampling structure of the

thin rounds. Certai

deaton dreze poverty india

Documents