GROWTH IN A POOR WORLD) - Princeton Universitydeaton/downloads/measuring poverty... · 2008-12-17 · The Review of Economics and Statistics V OL.LXXXVII NF EBRUARY 2005 UMBER 1 MEASURING

The Review of Economics and StatisticsVOL LXXXVII NUMBER 1FEBRUARY 2005

MEASURING POVERTY IN A GROWING WORLD (OR MEASURINGGROWTH IN A POOR WORLD)

Angus Deaton

AbstractmdashThe extent to which growth reduces global poverty has beendisputed for 30 years Although there are better data than ever beforecontroversies are not resolved A major problem is that consumptionmeasured from household surveys which is used to measure povertygrows less rapidly than consumption measured in national accounts in theworld as a whole and in large countries particularly India China and theUnited States In consequence measured poverty has fallen less rapidlythan appears warranted by measured growth in poor countries Oneplausible cause is that richer households are less likely to participate insurveys But growth in the national accounts is also upward biased andconsumption in the national accounts contains large and rapidly growingitems that are not consumed by the poor and not included in surveys Soit is possible for consumption of the poor to grow less rapidly thannational consumption without any increase in measured inequality Cur-rent statistical procedures in poor countries understate the rate of globalpoverty reduction and overstate growth in the world

I Introduction

A central issue in the debate about globalization is theextent to which economic growth reduces poverty

When economic growth benefits everyone in equal propor-tion the incomes of the poor grow at the same rate as doesmean income The fraction of the population whose incomesare below a fixed poverty line must then decline withgrowth although the rate at which it does so depends on theposition of the poverty line in the income distribution withgrowth in the mean generating more rapid poverty reductionthe greater the fraction of the population who are near thepoverty line If economic growth is unequally distributedthe effects of growth on poverty reduction will be less (ormore) depending on whether the incomes of the poor growby less (more) than average So much but perhaps not muchmore is common ground

Early debates on growth and poverty much influenced bySimon Kuznetsrsquos (1955) dictum that inequality would in-crease in the early stages of development tended to arguethat growth did little to reduce poverty Writing in the

1970s Chenery et al (1974) Adelman and Morris (1973)Fishlow (1972) and Bardhan (1973) all argued that eco-nomic development either left the poor behind or actuallymade them worse off see Cline (1975) for a contemporarysurvey Taylor and Bacha (1976) constructed a growthmodel of ldquoBelindiardquo a tiny rich Belgium in a huge poorIndia as an example of ldquothe unequalizing spiralrdquo that theysaw as fitting the stylized facts of development AhluwaliaCarter and Chenery (1979) who were among the first tomeasure global poverty using now standard methods ar-gued that the effect of growth was limited both by therelatively low growth of the poorest countries and byexpanding inequality within them When Fields (1977)argued that in the Brazilian economic miracle of the 1960sthe poor had actually done better than average he wasrobustly challenged by Ahluwalia et al (1980) who showedthat Fieldsrsquos conclusions were not warranted by his datawhich were consistent with an uninformatively wide rangeof differential growth rates of incomes of the poor andnonpoor This was surely the truth of the matter in 1980 thedata were not available to provide a clear answer to thequestion whether the poor did better the same or worsethan average during the unprecedentedly high rates ofgrowth in many poor countries in the immediate postwarperiod Researchers were forced to rely on a scattering ofpublished distributional measures whose provenance andreliability were often unclear and indeed Kuznetsrsquos famousarticle used distributional data for only three rich countrieswith a smaller amount of information for three poor ones

The paper by Ahluwalia et al (1980) was an importantimpetus to the establishment of the Living Standards Mea-surement Study (LSMS) at the World Bank The originalpurpose of the LSMS was to measure the living standards ofthe poor in a standardized way to remedy the paucity ofdistributional data in the third world and to set up a systemof household surveys that would both support and cross-check the national accounts as well as replicating for livingstandards measurement what the UNrsquos System of NationalAccounts (SNA) had done for national income accountsaround the world see for example Pyatt (2003)

Thirty years later the data situation has been transformedThere are two key innovations First internationally com-parable national accounts based on purchasing power parity

Received for publication June 9 2003 Revision accepted for publicationApril 7 2004

Woodrow Wilson School Princeton UniversityThe Review of Economics and Statistics Lecture presented at Harvard

University April 15 2003 I am grateful to Daron Acemoglu BettinaAten Barry Bosworth Francois Bourguignon Shaohua Chen RusselFreeman Paul Glewwe Carol Graham Tom Griffin Ivo Havinga AlanHeston Michael Kremer Martin Ravallion Dani Rodrik T N SrinivasanNick Stern John Williamson and Jeronimo Zettelmeyer for help in thepreparation of this paper as well as comments on an earlier draft AnneCase made many invaluable suggestions and corrected several errors Theviews expressed here are those of the author alone

The Review of Economics and Statistics February 2005 87(1) 1ndash19copy 2005 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology

(PPP) exchange rates allow comparisons of average livingstandards across countries in a way that is not vitiated by thegross inadequacies of conversions at market exchange ratesMaking comparisons in PPP units corrects or at leastdiminishes the gross understatement of living standards inpoor countries relative to rich and removes the spuriouscomponent of growth among poor countries that comesfrom the elimination of those differences with economicdevelopment PPP exchange rates were first used for globalpoverty estimates by Ahluwalia et al (1979) and their useis by now almost universal Second there has been anextraordinary growth in the number of household surveysavailable to the research community including severaldozen LSMS surveys For example the World Bankrsquos mostrecent set of poverty calculations use data from 297 surveysfrom 88 developing countries (Chen amp Ravallion 2001)Deininger and Squire (DS) (1996) have collected and tab-ulated data on more than 2600 Gini coefficients as well asmany measures of quintile shares the WIDER extensionincludes more than 5000 Gini coefficients The unit recorddata from many household surveys are now routinely avail-able to researchers including such previously inaccessibletroves as nearly 20 years of data from the Indian NationalSample Surveys back to the early 1980s Notable by itsexclusion is any similar access to Chinese official surveys

Yet the controversies are no more settled than they were30 years ago although there is certainly more commonground among economists than there is in the world at largeThe professional consensus based on the DS data and onwork by them and many others is that contrary toKuznetsrsquos hypothesis and contrary to beliefs in the 1970sthere is no general relationship between inequality andgrowth and certainly not one in which growth systemati-cally widens inequality as would be the case of growth leftthe poor behind From this two important propositionsfollow First at least on average (and much depends onwhether we are averaging over countries or people) growthis good for the poor (Dollar amp Kraay 2002 Ravallion2001) as is the growth that is arguably generated by greateropenness (Berg amp Krueger 2003) Second and again onaverage the fraction of people in poverty should decline asif growth were neutrally distributed In particular the rela-tively rapid growth in the developing world from 1980 to2000 must have brought about a rapid reduction in thefraction of the worldrsquos population that is poor And indeedcalculations using the Penn World Tables combined withinequality measuresmdashthe technique first used by Ahluwaliaet al (1979)mdashshow rapid poverty reduction in the 1980sand 1990s see Bhalla (2002) Sala-i-Martin (2002) andBourguignon and Morrisson (2002) According to thesecalculations not only has the proportion of poor in theworld declined but the decline has been rapid enough tooffset population growth so that the actual number of poorpeople in the world has fallen According to Bhalla the firstof the United Nations Millennium Development Goals

halving the number of people living on less than $1 a daybetween 1990 and 2015 had already been met when thegoal was announced

These optimistic calculations are starkly at odds with theWorld Bankrsquos numbers on global poverty The World Bankwhich is endorsed by the UN as official scorekeeper for thepoverty Millennium Development Goal uses householdsurvey data to measure the living standards of the poorignoring national accounts estimates and its calculationsshow relatively little poverty reduction in the 1990s Chenand Ravallion (2001) which provides the details of theBankrsquos calculations shows a reduction in the proportion ofthe poor living on less than $1 a day from 1987 to 1998from 283 to 235 they argue that this modest reductioncomes not from any expansion in inequality within coun-tries but from relatively slow growth in mean consumptionAcross their 88 countries the population-weighted rate ofgrowth in mean consumption was only 090 from 1987 to1998 compared with 33 growth in real per capita con-sumption in the Penn World Tables over the same periodThese estimates exclude the latest (1999ndash2000) Indian datawhose inclusion will increase the growth of the surveymeans over the 1990s There remains a large gap betweenon the one hand the direct assessment of the growth ofconsumption of the poor through surveys and on the otherhand the growth that is implied by the growth in averageaccompanied by no general increase in inequality

The plethora of new data has not resolved the contro-versy because the new sources are mutually contradictoryAccording to direct measurement in household surveysgrowth among the poor of the world has been sluggishcompared with the average growth rates of the countries inwhich they live Yet there is no documented increase ininequality that would resolve the discrepancy If we are toaccept the surveys growth in the world is a good dealslower than we are used to thinking from the nationalaccounts data and what growth there has been in the latesttwo decades has made only a modest dent in the level ofworld poverty If we accept the national accounts and donot challenge the conclusion that there is no general in-crease in inequality nor any correlation between growth andchanges in inequality then official poverty numbers areoverstated and we have already made rapid progress towardreducing poverty in the world This paper explores thesecontradictions empirically with an aim to providing asharper characterization and to advancing some first hypoth-eses about causes and possible remedies

A note of caution at the outset Because countries havevastly different populations statements about averages of-ten depend sharply on whether or not they are population-weighted A third of the worldrsquos poor live in two countriesIndia and China and the global poverty counts are muchaffected by what happens there When we are interested inthe well-being of the people of the world and in the effectsof statistical practice and statistical discrepancies on global

THE REVIEW OF ECONOMICS AND STATISTICS2

poverty measurements we must weight by populationThere is no reason to downweight the well-being of aChinese peasant relative to a Ghanaian cocoa farmer nor tobelieve that the world is a better place when an Africanmoves out of poverty and an Indian moves in Howevermany of my concerns are about the relationship betweenmeasurement and the level of development in which casethe appropriate procedure is to take each statistical systemas the unit and to ignore population sizes Beyond thatmany of the political negotiations about poverty and aboutmeasurementmdashfor example those in the councils of theUnited Nations and the World Bankmdashare carried on nationby nation In consequence I shall typically present bothweighted and unweighted results

II Surveys versus National Accounts All Countries

In this section I consider the cross-country and intertem-poral relationships between survey and national accountsestimates of consumption expenditure per capita Manycommentators have noted the (sometimes substantial) dis-crepancies between survey estimates and their nationalaccounts counterparts As we shall see below there are alsolong-standing literatures in India and the United States notonly on level differences but also on the fact that surveymeans grow less rapidly than means in the national ac-counts My analysis and data overlap with those of Raval-lion (2003) whose main concern is with regional and globalanalyses of the statistical significance of discrepancies in thelevels and growth rates of the ratios of survey to nationalaccounts consumption For consumption surveys Ravallioncomes to the optimistic conclusion that the significant dis-crepancies can be traced back to the disarray in the statis-tical systems of the transition economies The lack ofsignificant differences elsewhere reflects the large cross-country variation in the ratios as well as the fact that whensurveys are not weighted by population the low and fallingratio in India where approximately a third of the worldrsquospoor live is lost in the variation of the ratios elsewhere Inconsequence it is possible for the survey-to-national-accounts ratios to be insignificantly different from one eventhough the surveys and national accounts data have radi-cally different implications for trends in global poverty

National accounts estimates of consumption are availablefor most countries in most recent years so the countries anddates of the comparison are set by the availability of thesurveys The surveys used here come from a conveniencesample assembled from various sources In most cases Ihave survey estimates of mean income or mean consump-tion from the estimates assembled by Chen and Ravallion(2001) and which appear on the World Bankrsquos povertymonitoring Web site To these I have added my own esti-mates for India most of which appear in Deaton and Dreze(2002) a number of OECD surveys particularly from theConsumers Expenditure Survey and Current Population

Surveys in the United States and the Family ExpenditureSurvey (now the Expenditure and Food Survey) in theUnited Kingdom as well as a number of additional surveyestimates supplied by the Bank but not used in their povertycounts for example estimates of mean consumption perhead from the official Chinese surveys In all I have 557survey-based estimates of mean consumption per head ormean income per head (occasionally both) Table 1 showsthat these come from 127 countries that the earliest year is1979 and the latest 2000 The number of surveys in the dataset grows steadily larger over time I have only 3 in 1979and 7 in 1980 but 57 in 1998 (the peak year) There are 22surveys for 1999 and 26 in 2000 but this diminution innumbers after 1998 reflects merely the delay in processingand obtaining survey data rather than any slackening in thegrowth of usable surveys around the world For a singlecountry consumption and income estimates may comesfrom the same survey (for example in China) or fromdifferent surveys (for example in the United States) Thefraction of the worldrsquos population covered by the surveysshows a strong upward trend more than doubling from 1980to 1998 but there is fluctuation in the fraction from year toyear as individual countries move in and out of the countsMuch depends on whether or not there is an Indian surveyin a specific year China is included in 1980 and from 1985onward

Table 2 shows information on the ratios of survey esti-mates of consumption or income per head to consumptionor income per head from the national accounts The ratios

TABLE 1mdashDESCRIPTION OF SURVEYS USED IN THE ANALYSIS

YearNo of

SurveysNo of

CountriesPopulation

Covered (bn)Fraction of World

Population ()

1979 3 3 035 901980 7 6 133 3411981 5 5 027 671982 3 3 033 811983 2 2 097 2361984 7 5 048 1141985 13 11 159 3741986 21 18 175 4031987 23 21 182 4121988 26 24 292 6491989 31 28 214 4571990 23 20 169 3511991 29 26 195 3861992 37 34 216 4211993 45 41 261 4991994 35 30 322 6021995 51 45 367 6771996 48 44 382 6831997 43 38 338 6111998 57 53 386 7031999 22 19 216 3912000 26 23 354 636

All 557 127 mdash mdash

Notes Surveys are a convenience sample where survey means were readily available When thenumber of surveys exceeds the number of countries some countries have estimates of both mean incomeand mean consumption per capita China is included in 1980 1985 and every year thereafter India in1983 1988 1994 1995 1996 1997 1998 and 2000 but not in 1999 (Indian surveys that run frommidyear to midyear have been arbitrarily allocated to the second year) There are 278 estimates of meanconsumption and 281 estimates of mean income

MEASURING POVERTY IN A GROWING WORLD (OR MEASURING GROWTH IN A POOR WORLD) 3

are calculated using nominal values in local currency units(lcu) for both the numerator and denominator Nationalaccounts estimates of household final consumption are theobvious counterparts to survey consumption For incomemost countries do not publish data on disposable householdincome so that possible counterparts are GDP or onceagain household consumption The argument for the latteris that much of saving may not be done by households butby corporations government or foreigners so that house-hold income may be closer to household consumption thanto national income The top panel shows summary statisticsfor ratios of survey to national accounts consumption perhead the second panel is for the ratios of survey income tonational accounts consumption and the third panel is forratios of survey income to GDP

Consumption estimated from the surveys is typicallylower than estimated from the national accounts the aver-age ratio is 0860 (with a standard error of 0029) or 0779(0072) when weighted by population (India has particu-larly low ratios) The exception is sub-Saharan Africawhere the average ratio of survey to national accountsconsumption is 1 in the unweighted and greater than 1 in the

weighted calculations For the OECD where survey andnational accounts quality are presumably the highest thesurveys pick up only a little more than three-quarters ofconsumption in the national accounts These differencescome in part from differences in definitionmdashfor examplenational accounts consumption includes such items as theimputed value of owner-occupied housing which is nearlyalways excluded from the surveysmdashbut they also reflecterrors and omissions in both surveys and national accountsIn consequence that the ratios for the Middle East andNorth Africa (MENA) and sub-Saharan Africa (SSA) areclose to 1 says nothing about the quality of the surveys inthose two regions Indeed it is possible that the perfectlymeasured ratio is less than 1 but is actually measured asgreater than 1 because there is understatement in the na-tional accounts And it is entirely possible that the highratios for SSA come from large-scale underestimation in thenational accounts

Income measured in the surveys is on average larger thanconsumption measured in the surveys but is in most cases lessthan national accounts consumption and much less than GDPSurvey income is less than 60 of GDP on average

TABLE 2mdashRATIOS OF SURVEY MEANS TO MEANS FROM NATIONAL INCOME ACCOUNTS

No ofSurveys

Unweighted Population-Weighted

MeanRatio

StandardError

StandardDeviation

MeanRatio

StandardError

StandardDeviation

Consumption to Consumption

All 277 0860 (0029) 0306 0779 (0072) 0191

EAP 42 0819 (0069) 0224 0863 (0031) 0110EECA 59 0847 (0038) 0230 0796 (0040) 0184LAC 26 0767 (0094) 0329 0585 (0078) 0193MENA 20 0955 (0104) 0300 0867 (0111) 0270OECD 33 0781 (0052) 0097 0726 (0032) 0076SA 23 0649 (0063) 0122 0569 (0036) 0103SSA 74 1000 (0061) 0415 1089 (0089) 0459

Income to Consumption

All 266 0904 (0034) 0290 1008 (0044) 0174

EAP 32 1036 (0065) 0244 1057 (0019) 0105EECA 47 0852 (0038) 0231 0811 (0030) 0196LAC 100 0893 (0084) 0392 1004 (0143) 0416OECD 75 0891 (0020) 0137 0910 (0011) 0084SA 8 0892 (0028) 0118 0874 (0009) 0101SSA 4 1000 (0136) 0420 1023 (0204) 0359

Income to GDP

All 272 0569 (0023) 0203 0542 (0023) 0113


Notes EAP is East Asia and Pacific EECA is eastern Europe and central Asia LAC is Latin America and the Caribbean OECD comprises the countries of the OECD SA is south Asia MENA is Middle Eastand north Africa and SSA is sub-Saharan Africa There are no income surveys for MENA in the sample Numbers differ slightly from table 1 because the relevant national income magnitudes are not always availablePanel 1 shows the ratio of consumption from the survey to consumption from the national accounts panel 2 the ratio of income from the surveys to consumption from the national accounts and panel 3 the ratioof income from the surveys to GDP from the national accounts Standard errors are calculated so as to allow for correlations within countries


The standard deviations of the ratios provide one crudeindicator of combined survey and national accounts accu-racy including both sampling and nonsampling errorsWithout prejudging the relative accuracy of national ac-counts and the surveys the latter are more likely to varyfrom year to year for example because of sampling andchanges in survey design and from country to countrybecause survey protocols are less standardized internation-ally than are national accounts By this measure the surveysin sub-Saharan Africa are the most problematic though thesurveys in Latin America and the Caribbean also show greatvariance particularly the income surveys OECD surveyshave the lowest variance followed by south Asia wherehigh-quality household surveys have been in existence formany years In spite of the difficulties of collecting data intransition economies the eastern Europe and central Asiaregion does not show particularly high variance In severalcountries in both EECA and LAC high inflation poses greatproblems for both survey and national accounts data

Figures 1 (weighted by population) and 2 (unweighted)show how the same three ratios are related to the level ofGDP here GDP per head in 1995 PPP dollars (This is the

World Bankrsquos current PPP series divided by the implicitprice deflator of GDP in the United States) Cross-countryand time series data are pooled in these graphs

There are two points to take away from these figuresFirst the top left panels in both figures show a negativerelationship between the ratio of survey to national accountsconsumption on the one hand and the GDP per capita on theother This relationship is steepest among the poorest coun-tries is flatter in middle income countries but resumes itsdownward slope among the rich countries The continuouslines in the two top left graphs are locally weighted non-parametric regressions of the relationship using a bandwidthof 15 (units of real log GDP in PPP) Second there is nosimilar relationship among the income surveys either forthe ratio of survey income to national accounts consumptionor for the ratio of survey income to GDP At least some of thepattern in figure 1 must come from the fact that consumptionis typically much easier to measure in surveys than is incomein poor countries where many people are self-employed inagriculture whereas the opposite is true in rich countrieswhere most people are wage earners and are more reluctant tocooperate with time-consuming consumption surveys

FIGURE 1mdashRATIO OF SURVEY ESTIMATES OF MEAN INCOME OR CONSUMPTION PER CAPITA TO COMPARABLE NATIONAL ACCOUNTS ESTIMATES

498 Surveys 124 Countries Years from 1979 to 2000 The diameter of the circles is proportional to national population in the year of the survey


For assessing trends in global poverty and growth themost important feature of these data is the behavior of theratios over time This issue is explored in figure 3 and table3 Because the subset of countries for which we have surveymeans differs from year to year it is not useful to calculaterates of growth of the survey means country by country andthen weight by population to obtain estimates of globalgrowthfromthesurveysInsteadIhavecomputedpopulation-weighted averages for each year over whatever subsets ofcountries have survey data First the local-currency con-sumption and income means are converted to PPP dollars bydeflation by the consumption PPP exchange rate from thePenn World Tables version 61 (PWT61) and then into realterms by deflation by the US CPI They are then weightedby population and averaged excluding the wealthy coun-tries of the OECD The resulting series are plotted as thebottom two lines in figure 3 They differ in their treatmentof country-years where there are both a consumption and anincome mean For the broken lower line I have chosen theconsumption survey whenever both are available and forthe solid upper line I have chosen the income survey (The

results of choosing income means are almost identical if wetake income means for China and consumption meanselsewhere)

For comparison with these survey-based estimates I haveused real consumption from PWT61 calculated by applyingthe consumption share to the chain-weighted GDP seriesThe top solid line in the figure shows the populationweighted average of PWT61 consumption for all of thecountries that ever appear in the survey data set excludingonly the OECD The broken line is also a population-weighted average of PWT61 consumption but for eachyear is averaged only over the countries for which there aresurvey data This calculation allows a comparison with thesurvey calculations in which the two series are affectedsimilarly by the variation that comes from the fact thatsurvey countries (and thus the composition of the sumacross the world income distribution) changes from year toyear And indeed the year-to-year variation in the broken-line version of the PWT61 consumption series is highlycorrelated with both survey measures Of course the year-to-year (or cyclical) fluctuations in all the series in figure 3


498 surveys 124 countries years from 1979 to 2000 Unweighted


(except for the top line) comes as much from the changingselection of countries with different living standards as itdoes from any genuine fluctuations in the unobservablesurvey mean over all countries so that we can use theseseries only to examine long-run growth not differences ingrowth rates over subperiods

Figure 3 shows that national accounts consumption innon-OECD countries here taken from the PWT61 andshown in the top two lines grew more rapidly over the1990s than did consumption from poor countries measuredfrom the surveys shown in the bottom line Table 3 showsthat growth of survey consumption is 23 a year if we

simply take average growth over the decade or 19 a yearif we regress its logarithm on a time trend the difference inthe two estimates coming from the variability in the seriesThis difference is induced by countries with different in-come levels particularly India moving in and out of thesurvey averages and is also seen in the comparison growthrates from national accounts consumption which are 38and 45 a year Whether we take the two low or two highestimates the growth rate of survey consumption is approx-imately half of the growth rate of national accounts con-sumption If instead of using consumption estimates fromthe surveys we take income estimates when they are avail-able the situation is reversed and we get a rate of growthfrom the surveys that is larger that the correspondinggrowth rates in national accounts consumption The highergrowth rate when we give preference to income surveyscomes almost entirely from the Chinese data The WorldBankrsquos global poverty estimates use income surveys forChina because there are no distributional data for theChinese consumption figures However in the Bankrsquos cal-culation the Chinese income distribution is scaled down bythe ratio of consumption to income in the Chinese nationalaccounts a ratio that has been rising over time so that thefirst column in table 3 and the bottom graph in figure 3 arethe relevant ones for thinking about trends in global povertyas measured by the dollar-a-day counts

III Surveys versus National Accounts India and China

Figure 4 shows the ratios of survey to national accountsestimates for China and for India The Chinese data whichhave a discontinuity in 1990 for which there are twoestimates are from the same survey data base discussedabove the national accounts data are taken from the 2002

TABLE 3mdashPOPULATION-WEIGHTED GROWTH RATES 1990ndash2000 REAL

CONSUMPTION OR REAL INCOME VARIOUS MEASURES NON-OECD COUNTRIES

Growth Rate (yr)

Surveys withConsumption

Preference

Surveys withIncome

Preference

PWT61Matching

Surveys byYear andCountry

PWT61all SurveyCountries

Regression oflog on time 19 40 38 28

Average rateof growth 23 50 45 28

Notes Columns 1 and 2 show the growth rates of population-weighted survey means In column 1whenever there is both an income and a consumption mean for a country year pair consumption is usedIn column 2 whenever there are two surveys preference is given to income In both cases survey meansare converted to a constant-price PPP basis by dividing by the product of the US CPI and theconsumption PPP exchange rate from the Penn World Tables Version 61 (PWT61) For each year from1990 to 2000 a population-weighted average of the survey means is calculated note that these averagesinvolve different countries in different years (see table 1) The growth rates are then calculated in twoways by regression of the logarithm on a time trend (first row) and by calculating the average changein the logarithm over the period These can be quite different when the series is noisy as is the case herebecause countries come in and out of the average Columns 3 and 4 show comparable population-weighted growth rates for real PPP (chain-weighted) consumption from PWT61 In column 3 con-sumption from PWT61 is used only for country year pairs for which a survey mean exists this columntherefore shares the variability in columns 2 and 3 that comes from the varying selection of countriesColumn 4 shows the population-weighted growth rates for consumption from PWT61 using all countriesfor which there ever exists a survey

FIGURE 3mdashLOGARITHMS OF POPULATION-WEIGHTED AVERAGES OF CONSUMPTION OR INCOME

Household surveys and Penn World Tables v 61


Edition of the World Development Indicators In China theratio of survey to national accounts consumption has beendeclining since around 1990 from a peak of 95 in 1990 to80 in 2000 the growth rates of the two series thus differby approximately 17 a year in the 1990s The ratio ofsurvey income (from the same surveys as consumption) tonational consumption did not decline over the same periodHowever there is a great deal of household saving in China(which shows up in the surveys in that the top line is muchhigher than the bottom line) so national consumption is notthe relevant comparison Ideally income should be com-pared with GDP or better still some national accountsestimate of household income Although I do not have thedata to calculate that ratio there is little doubt that it wouldalso be declining over time Given the population of Chinaits increasing discrepancies between survey and nationalaccounts is a major contributor to the global differencesHowever it should be noted that many commentators haveargued that the growth rates in the Chinese national ac-counts are too high The discussions by Maddison (1998)Wu (2000) and Keidel (2001) are all consistent with anoverestimation in the rate of GDP growth by between 2 and4 percentage points a year and Rawski (2001) argues formuch larger overestimation in the last few years Removing2 percentage points a year from NAS consumption growthwould eliminate the difference in the growth rates betweenthe NAS and the surveys

Figure 4 also shows the data from India in this casetaken for national consumption directly from the latestavailable edition of the national accounts (Government ofIndia 2003) and for the survey estimates from my owncalculations from the unit record data The Indian NationalSample Survey (NSS) conducted its latest full-scale house-hold expenditure survey in 1999ndash2000 but because thequestionnaire design was changed from earlier similar sur-veys there has been controversy about the interpretation ofthe results The estimate of average consumption used here

was calculated according to the methods laid out in Deaton(2003) but differs relatively little from the official calcula-tions much less than is the case for the poverty estimates InIndia survey consumption is much lower relative to na-tional accounts consumption than it was in China Howeveras in China the ratio of the two estimates of consumptionhas been declining over time In 1983 the ratio was 068which declined in 1999ndash2000 to 056 so that nationalaccounts consumption has been growing at 11yr morerapidly than survey consumption India like China ac-counts for a large share of the worldrsquos population and aneven larger share of those who live on less than $1 a day

The Indian consumption ratio in figure 4 calls for someadditional comment particularly the erratic behavior from1995 through 1998 The Indian NSS carries out largehousehold expenditure surveys only once every 6 years orso the two most recent being in 1993ndash1994 and 1999ndash2000The estimates between those dates come from four smallerNSS surveys that also collect expenditure data Although thesample sizes of those surveys are sufficient to obtain reliableestimates of the national headcount ratio there have beenquestions about their design The 1998 survey in particularlasted only for half a year and it is arguable that thepenultimate observation in the graph should be ignoredUnfortunately circumstances have conspired to give thisand the immediately preceding observation a great deal ofweight Because the 1999ndash2000 survey was arguably con-taminated by changes in the questionnaire the 1997 and1998 surveys did not fade into history as quickly as theyotherwise would have done In addition these were thelatest observations for India available to the World Bank forthe most recent set of global poverty counts constructed forthe 2000ndash2001 World Development Report on poverty Theuse of the new data in the next round of global povertycounts will give a more optimistic picture of the rate ofglobal poverty decline though not as optimistic as would be

FIGURE 4mdashRATIOS OF SURVEY MEANS TO NATIONAL ACCOUNTS MEANS OF CONSUMPTION ANDOR INCOME PER HEAD INDIA AND CHINA


the case if survey growth had been as rapid as growth innational accounts

The internal Indian debate on discrepancies betweensurveys and national accounts has flared up sporadically forat least thirty years see in particular the papers in Sriniva-san and Bardhan (1974) as well as those of Minhas (1988)and Minhas and Kansal (1989) The recent spate of interesthas generated a great deal of important detailed workincluding collaborative efforts between the NSS and theNational Accounts Division of the Central Statistical OfficeMuch can be learned from that work not only for India butalso for other countries and I postpone discussion untilsection V

IV Surveys versus National Accounts the UnitedKingdom and the United States

Although my primary concern is with the measurement ofglobal poverty and thus with measurement in poor coun-tries the issue of statistical discrepancies between surveysand national accounts is a general one and there is a greatdeal to be learned by looking at the issue at the other end ofthe global income distribution Rich countries tend to havefuller data so that it is sometimes possible to test generalhypotheses about surveys that cannot readily be tested infor example India or China

Figure 5 presents the results of survey and nationalaccounts comparisons for the United States and the UnitedKingdom The right-hand panel shows results for the FamilyExpenditure Survey (FES) since 1995 subsumed into theExpenditure and Food Survey (EFS) Data on real consump-tion per head were taken from the EFS reports and werescaled up using population and retail price data from theAnnual Abstract of Statistics (Office for National Statistics2003) The figure shows the ratios of these numbers to finalconsumersrsquo expenditure from the national accounts Theredesign of the survey and the switch from the FES to theEFS results in a discontinuity at 1995 for which year thereare two estimates The left-hand panel shows correspondingdata from the United States using two different surveys the

Current Population Survey (CPS) and the Consumer Expen-diture Survey (CEX) The CEX is the main consumptionsurvey in the United States although it also collects incomedata and the two solid lines in the left-hand panel show (a)the ratio of consumption from the CEX to consumption inthe national accounts (the upper solid line) and (b) the ratioof pretax income from the CEX to personal income from thenational accounts (the lower solid line) The CEX incomeand consumption estimates are calculated by the Bureau ofLabor Statistics from the CEX and include estimates fromboth the diary and interview components of the survey aswell as an estimate of the rental equivalents of owner-occupied homes The CPS which is the main income surveyin the United States and which is used by the Bureau of theCensus to calculate the official estimates of poverty doesnot collect data on consumption The broken line in thefigure is the ratio of income from the CPS to personalincome in the national accounts

Both sets of consumption figures show the now familiarpattern of declining ratios of survey to national accountsconsumption In the United Kingdom the decline is farfrom uniform over time and if the break in the survey in1995 had been in 1994 it could perhaps have been attrib-uted to the change in design As it is the ratio declines byapproximately 10 percentage points over the 25 years from1976 to 2001 so that survey consumption is growingapproximately half a percent a year less rapidly than con-sumption in the national accounts The decline in the cor-responding ratio in the United States is a good deal moredramatic from 080 to 064 from 1984 to 2001 so that thedifference in the two growth rates is 13yr a little morethan in India and a little less than in China Income from theCEX is also a declining fraction of personal income in thenational accounts although the rate of decline is muchslower less than 5 percentage points over 17 years Andincome in the CPS shows no trend relative to personalincome in the national accounts

More careful comparisons between the CEX and nationalaccounts consumption data have been made by Triplett

FIGURE 5mdashRATIOS OF SURVEY MEANS TO NATIONAL ACCOUNTS MEANS OF CONSUMPTION ANDOR INCOME PER HEADUNITED STATES AND UNITED KINGDOM


(1997) and by Garner et al (2003) After making a numberof corrections to try to put the two series on a comparablebasis Triplett estimates that from 1984 to 1994 personalconsumption expenditures grew at 10yr more rapidlythan consumers expenditure from the CEX Garner et alcomparing only comparable items calculate that the ratio ofCEX to national accounts values was 89 in 1992 In 1997and 2000 the comparable ratio was only 80 so that thedifferential growth rate was 24yr until 1997 and 15yrto 2000

The differential behavior of income and consumptionratios may have something to do with the fact that in theUnited States consumption is much harder to collect than isincome The CEX costs a great deal more per interview thandoes the CPS and whereas for most people (those who arenot self-employed) income can be collected with only a fewquestions consumption requires a long interview or exten-sive recordkeeping in diaries The nonresponse rate in theCEX has been rising over time (Groves amp Couper 1998)while that for the CPS has been constant It is possible thatpeople are less and less willing to cooperate with the CEXover time but those who do so are prepared to answer theincome questions more fully and more accurately than theconsumption questions The CPS which does not ask con-sumption questions may suffer from fewer problems Itssample size is also much larger 60000 households versusonly 5000 up to 1999 and 7500 thereafter

V Why Do Surveys and National Accounts Diverge

The previous sections have documented the fact thatconsumption measured from surveys frequently grows lessrapidly than consumption measured from the national ac-counts Consistent with this general relationship the ratio ofthe two magnitudes is highest in the poorest countries andlowest in the richest Within countries as diverse as ChinaIndia the United Kingdom and the United States the ratiofalls over time as real income increases Taking non-OECDcountries as a whole population-weighted survey consump-tion in PPP constant dollars grew at only half the rate ofpopulation-weighted consumption in the Penn World Ta-bles There are conceptual differences between the twoconcepts of consumption but these do not account for thedifferences in growth rates so that one or both of the growthrates are incorrect If the surveys are wrong and the nationalaccounts right either inequality has been widening in waysthat our data do not appear to show or poverty has beenfalling more rapidly than shown by the $1-a-day counts Ifthe surveys are right there was less growth in the world inthe 1990s than usually supposed Extreme positions apartwe have some combination of underestimation of povertydecline underestimation of a widening in the distribution ofconsumption and overestimation of growth Quantifyingthe contribution of each is an urgent task for anyone inter-

ested in growth poverty and inequality In this section I layout some of the possible explanations

It is important to note that there can be no generalpresumption in favor of one or other of the surveys and thenational accounts In particular that national accounts arefamiliar widely used and in principle comparable (theytypically conform to the UNrsquos System of National Ac-counts) does not imply that the divergences between themand the surveys must be attributed to the latter Although itis certainly the case that there exist ldquofailedrdquo surveys whoseexecution is known to have been faulty where fieldworkwas disrupted or inadequately supervised where samplingprocedures were flawed or where changes in survey designmade it impossible to compare the results with earliersurveys national accounts estimates are also subject tomany errors some of which will be discussed below

A Unit Nonresponse

Not everyone who is asked to participate in a surveyagrees to do so and failure to respond (unit nonresponse) isknown to be different for households with different house-hold characteristics (Groves amp Couper 1998) Of particularinterest is the case where better-off households are lesslikely to respond Groves and Couper report that in richcountries the probability of response is negatively related toalmost all measures of socioeconomic status and thoughsurvey organizations in poor countries can usually collectdata in very poor areas (albeit under difficult conditions) itis often impossible to penetrate the gated communities inwhich many rich people live Suppose then that the proba-bility that consumption y is recorded in the survey is (y)and that (y) is monotone declining in y This situation hasbeen discussed in a recent paper by Mistiaen and Ravallion(2003) who also show how to use aggregate measures ofnonresponse (for example by region) to correct estimates ofpoverty and inequality

If the true (untruncated) density of consumption (orincome) is f(y) the density for observed (truncated) con-sumption is

f y y f y

y0

y1

f y y dy

y fy

(1)

where y0 and y1 are the bottom and top levels of consump-tion and is the mean response in the population Fromequation (1) the difference in the true and actual densities is

f y f y y

f y (2)

so that the observed density is higher or lower according towhether the householdrsquos response rate is below or above themean Because (y) is monotone decreasing the truncated


density is higher at low levels of y and lower at high valuesso that the distribution function is shifted to the left that is

F y F y (3)

This inequality says that the truncated distribution is first-order stochastic dominated by the untruncated distributionwhich implies that the estimated poverty rate from theactual data will be no less than the estimated poverty rate inthe population no matter what the poverty line and that theestimated mean consumption will be no larger than thepopulation mean

Mistaen and Ravallion also consider the effects of thetruncation on the Lorenz curve L(p) In general the deriv-ative of the Lorenz curve satisfies (see for example Kak-wani 1987)

L p F1 p

y

(4)

where y is the pth quantile of consumption and is itsmean Provided that y0 0 and y1 and provided(y) 0 for all y in the support so that the support of thetruncated distribution is identical to that of the original thereduction in the mean by the nonresponse implies that thetruncated Lorenz curve is at least as steep as the true Lorenzcurve both at the origin and at (11) so that either theLorenz curves are identical or they must cross at least onceThis result although obtained under special assumptions(for example if y0 0 it is possible to construct caseswhere the curves need not cross) tells us that with greaternonresponse by the rich there can be no general suppositionthat estimated inequality will be biased either up or down bythe selective undersampling of richer households (Theintuition that selective removal of the rich should reducemeasured inequality which is sometimes stated as obviousin the literature is false perhaps because it takes no accountof reduction in the mean from the selection)

If we are prepared to place restrictions on the compliancefunction (y) we can analyze the effect of inequality oncompliance In particular suppose (a) that (y) in additionto being monotone decreasing is convex and (b) that y(y)is monotone increasing and concave Then if F1 and F2 aretwo distributions of income with the same mean such thatF1 second-order stochastically dominates F2 we have

y dF1 y y dF2 y (5)

so that the average compliance is lower for the more equaldistribution In addition

y y dF1 y y y dF2 y (6)

so that dividing equation (5) by (6) we have

1 2 (7)

Provided the two monotonicity assumptions and convexityand concavity assumptions are satisfied a mean-preservingincrease in spread in the true distribution will decrease thetruncated mean The monotonicity assumption in (b) guar-antees that in spite of the noncompliance reported incomeincreases with actual income The concavity and convexityassumptions guarantee the result but do not appear to berequired by the logic of the problem

To sharpen intuition further consider the following illus-trative but not unrealistic case in which a log normaldistribution of income is combined with a probability ofcompliance that is nonincreasing in income Suppose that xis the logarithm of income or consumption and that thedistribution prior to truncation is log normal with mean (oflogs) v and variance (of logs) 2 Suppose too that theprobability of responding to the survey is unity up to someincome level exp(v 13) for some number 13 but thatabove exp(v 13) the logarithm of the compliance prob-ability declines linearly in the logarithm of income the kinkin the response function is needed to prevent the probabilitybeing greater than unity Hence if (x) is the probability thata household with (log) income x agrees to cooperate wehave

x1 x v 13exp[x v 13)] x v 13

(8)

so that the probability of response is unity at the bottom ofthe distribution The parameter is nonnegative and is(minus) the elasticity of compliance with respect to income

In the Appendix I show that provided 13 is large enoughso that noncompliance begins far enough below the meanthe observed (truncated) distribution of incomes is approx-imately log normal and that the variance of log income isunchanged but the mean of logs is shifted downward fromv to v 2 Although this result is entirely driven byassumption it illustrates a number of important pointsFirst we have a case where nonresponse drives the differ-ence between the national accounts and the surveys andwhere the mean is biased down but the Lorenz curve iscorrect Second the ratio of survey consumption to trueconsumption depends on the variance of the true (andtruncated) distribution In particular If and are thetruncated and true means of income the ratio satisfies

ln

2 (9)

so that the understatement of income will be greater inplaces and at times where inequality is higher In particularincreasing inequality of incomes will drive down the surveyestimates in relation to the truth even though the ratio ofsurvey to the true mean is independent of the level of meanincome Third in this case the ratio of the truncated to thetrue mean is independent of mean income so that althoughcompliance is declining in income and although average


compliance is declining as the economy expands [at least ifthe compliance probability in equation (8) is scaled torespond to actual income rather than the deviation ofincome from the mean] the fraction of total income cap-tured by the survey does not decrease with growth

The compliance probability in equation (8) can be gen-eralized for example by introducing a quadratic term in thesecond branch of (8) so that

x 1 x v 13

expx v 13

2x v 132 x v 13

(10)

The parameter can be positive or negative in the lattercase equation (10) needs to be modified at high levels of xto stop the probability exceeding 1 Although I do not dealwith the complication here high values of x can be handledin the same way as low values of x in equation (8) Underthe same condition as before that 13 is large enough equa-tion (8) also implies that the truncated distribution will belog normal but now both mean and variance of logs arechanged Similar algebra to the linear case gives

2 2

1 2 (11)

for the variance of logs in the observed distribution whichcan be greater than or less than 2 depending on the sign of For the mean of logs we have

v v 2 13

1 2 (12)

Once again the inequality of income affects the ratio of theobserved to the true mean However it is no longer appro-priate to replace the mean by its true value leaving thevariance unchanged because if is nonzero the variance isnow also affected by the noncompliance something that wewould generally expect to be the case Note that as in theoriginal case the ratio of true to measured income does notvary with the true mean so that noncompliance can increasewith income without the ratio of measured to true incomefalling with increases in mean income

There are no ideal aggregate data for testing the extent towhich mean income and income inequality affect surveymeans through noncompliance Although there is a greatdeal of distributional information in the DS data set theinformation for developing countries is neither reliable initself nor well matched to the surveys in the sample dis-cussed above For the smaller subset of 111 consumptionand 77 income surveys for which Gini coefficients areprovided on the World Bankrsquos poverty-monitoring web sitethere is no significant (unweighted as is appropriate here)relationship between the log of the ratio of survey to NAS

mean and the Gini coefficient whether or not real GDP percapita is controlled for (Region by region there is a mar-ginally significant effect in South Asia where the dataquality is probably highest note that the OECD countriesare not represented in the poverty monitor countries) This isalso true when the Gini is replaced by the log variancecalculated from the formula for the log standard deviation 21g1)2] which holds when the distribution islog normal (Aitchison amp Brown 1969)

Another place to look is across the states of India wherethere exist state net domestic product data which can becompared with the state means from the household surveysAgain this comparison is far from ideal the state domesticproduct accounts are widely believed to be measured withconsiderable error and even without error the ideal com-parison would not be with net domestic product but withconsumption An offsetting advantage compared with theinternational data is that the state survey means and in-equality measures are derived from the same surveys usingidentical questionnaires and procedures in each state Itshould also be noted that the Indian National Sample SurveyOrganization consistently maintains that noncompliance israre and that numerators make repeated visits until peopleare available or it is convenient for them The data from thesurveys also carry a notation for whether the householdactually surveyed was the one originally intended orwhether it is a substitute for the household targeted forsampling In the 1999ndash2000 survey only 1200 out of morethan 70000 rural households are listed as substitutes with1900 out of 48900 urban households Approximately two-thirds of the substitutions are attributed to the informantbeing away and less than a quarter to informants being busyor uncooperative

The state survey means are well correlated with the stateestimates across the 43rd (1987ndash1988) 50th (1993ndash1994)and 55th (1999ndash2000) rounds of the NSS and using meansfor only the 17 largest states the correlations are 088 orhigher if Delhi is included and 070 or higher if is excludedIf we use the log of the ratio of survey consumption perhead to state net domestic product per head as the left-handside of equation (6) and the variance of logs from thesurveys as the right-hand side the regression coefficient onthe variance of logs is 139 (t 33) in a pooledregression of the three rounds (54 observations from 18states in each of the rounds) including round dummiesTaking each round separately the corresponding coeffi-cients (and t-values) are 069 (08) 078 (26) and144 (22) for the 43rd 50th and 55th rounds respec-tively Figure 6 shows the corresponding plots with eachstate identified Taken literally these estimates suggest thatthe elasticity of noncompliance has almost doubled in the 12years between 1987ndash1988 and 1999ndash2000 which is cer-tainly consistent with a fall in the fraction of aggregateconsumption captured by the surveys Inequality withinurban areas and between urban and rural areas has also


been rising in India (Deaton amp Dreze 2002) which wouldagain depress the ratio of survey to NAS means Of coursethese results are consistent with a wide range of otherpossibilities for example as suggested by a referee stateswith more inequality could have higher savings rates and inthe absence of good capital markets a higher share ofinvestment and a lower share of consumption in state GDPor higher inequality could generate more government ex-penditure with similar consequences So the empirical ev-idence is weak at best

These data are not suitable for investigating the importantquestion of whether the ratios are lower when mean con-sumption is higher This is because the state net domesticproduct is used in the calculation of the log ratio so that toinclude it in the regression is to guarantee a negativecorrelation whether or not one actually exists Anothervariable that is plausibly important is the degree of urban-ization if enumerators have greater difficulty contacting orobtaining compliance from urban households In fact withthe Indian state data it is difficult to tell the urbanizationand inequality explanations apart Urbanization (the fractionof population in the urban sector) can be used to replace thevariance of logs in the regression with similar t-values andwhen both urbanization and the variance are entered to-

gether neither is significantly different from 0 Urbanizationand inequality are highly correlated in these data and wecannot tell whether it is high income that poses the problemfor the surveys With only 18 states I am almost certainlypushing these data too far Nevertheless the question ofcompliance is central to the analysis of survey versusnational accounts and the Indian experience provides somesupport for the idea that income-related noncomplianceexplains some part of the shortfall between the surveys andthe national accounts and perhaps even a part of why theshortfall is increasing

B Issues Involving National Accounts

Although noncompliance almost certainly explains atleast some of the discrepancies between surveys and na-tional accounts and although there are other problems withthe surveys beyond noncompliance there are serious qualityissues with the national accounts estimates of consumptionand GDP I discuss some of the most important in thissubsection

Discrepancies between survey and national accounts es-timates of consumption can arise both through differencesin definition and through differences in the relative success

FIGURE 6mdashRATIO OF SURVEY MEAN OF CONSUMPTION TO STATE NET DOMESTIC PRODUCT AND VARIANCE OF LOG PER CAPITA TOTAL HOUSEHOLD

EXPENDITURE (PCE) INDIA THREE NSS SURVEY ROUNDS


of meeting those definitions National accounts statistics arecompiled according to protocols laid down in the 1993version of the Systems of National Accounts SNA93 TheSNA93 establishes what is known as the production bound-ary which defines what is and is not part of consumptionand GDP The production boundary includes all goods andservices that are exchanged as well as goods that are notexchanged such as food produced for home consumptionbut excludes services that are not exchanged such as foodpreparation home education of children or minor homerepairs with the important exception of housing servicesconsumed by owner-occupiers Because the excluded ser-vices tend to be replaced by exchanged services as peoplebecome better off and substitute market for home produc-tion the measured growth rate of consumption and GDPwill be too high at least when the true rate of growth ispositive Yet this bias cannot explain any of the growingdiscrepancy between surveys and national accounts be-cause the nonexchanged services are not included in thesurveys either

Yet the degree of effective coverage of the nonexchangeditems will almost always differ between the surveys andnational accounts Surveys almost never directly collectdata on implicit rents for owner-occupiers (other than themaintenance component) though it is sometimes possible touse data from the surveys on housing characteristics toestimate hedonic rental regressions provided there is a localrental market But few of the surveys used for povertyanalysis contain such estimates which undoubtedly contrib-utes to the finding that survey-to-NAS consumption ratiosare typically less than 1 and if the share of the NASconsumption attributable to rents to owner-occupiers isincreasing over time it will also help explain the increasingdivergence I have not made any general study of the latterbut in India the share appears to be more or less constantover time In the United States the comparison of the CEXand the NAS in figure 5 includes imputed rents in bothnumbers so that this cannot be the source of the increasingdiscrepancy between them There are also likely to bedifferences in coverage of nonexchanged goods Consump-tion of own production gifts and wages in kind are animportant part of the total in poor countries and many goodsurvey offices devote a great deal of attention to collectingsuch information For example the Indian NSS distin-guishes purchases own production and gifts for severalhundred items The coverage of nonexchanged goods in theNAS will depend on the methodology employed Somecountries use the survey estimates but in many and perhapsmost cases consumption is calculated as a residual in aprocess that begins from production In principle this is nota problem but in many countries it would be extremelyoptimistic to suppose that the measurement of productionaccurately captures home production

Consumption surveys as opposed to income surveys arelikely to capture a good deal of illegal or legal but con-

cealed (for example to avoid taxes or regulation) activitiesPurchasers of such goods and services unlike their produc-ers often have no incentives to conceal their transactionsand individuals who have substantial income from sourcesthat they are unlikely to report may nevertheless report theconsumption that is financed by that income see Blades andRoberts (2002) OECD (2002) Because of this and becausemany surveys collect comprehensive data on nonexchangedproduction it is not surprising that in some of the poorestcountries consumption measured in the surveys is some-times much larger than consumption estimated in the na-tional accounts As nonexchanged production becomes lessimportant with economic development the effect will wearoff and could thus account for at least some of the declinein the ratio of the two measures However as Blades andRoberts emphasize claims that the existence of variousnonobserved activities means that a large share of GDP asmuch as a quarter or a third is missed in the nationalaccounts are almost certainly exaggerated National incomeaccountants understand the nature of the problems andalthough their estimates for the nonobserved economy maynot be very accurate they do not omit it

In addition to imputed rents of homeowners there are twoother important items of consumption that are included inthe NAS but not in the surveys One is financial servicesindirectly imputed or FISIM which is an estimate of theconsumption value of financial intermediation FISIM ismeasured as the interest paid to banks and other intermedi-aries less interest paid by them The idea is that interestcharged to borrowers contains in addition to the market rateof interest a charge for intermediation services to lendersin addition to the market rate of interest whereas interestpaid to lenders is lower than market with the differenceattributed to financial intermediation services to depositorsThe difference between interest paid and interest received istherefore a measure of the value of financial intermediationand since the 1993 revision of the SNA has been added tonational accounts estimates of household consumption Asimilar item is included for risk-bearing services measuredfrom the profits of insurance companies In India the valueof FISIM increased from close to 0 in 198384 to 25 ofconsumption in 1993ndash1994 (Kulshreshtha amp Kar 2002) sothat this item alone accounts for a quarter of a percentagepoint per year of the difference in annual growth ratesbetween NAS and survey consumption in India Note alsothat to the extent we are interested in measuring the livingstandards of the poor it can reasonably be doubted whetherthe value of such financial intermediation is relevant Inconsequence even if we accept the argument for the inclu-sion of FISIM in NAS consumption neither it nor its rate ofgrowth contributes to the living standards of the poor

The second potentially important item of consumptionincluded in NAS but not in the surveys is consumption bynonprofit institutions serving households (NPISH) whichin most countries cannot be separated from household


consumption It is unclear how large these items are orwhether such expenditures are growing relative to totalconsumption In the United Kingdom NPISH in 2001 was39 of total consumption almost double the 1970 share of21 It is possible that NPISH are much more widespreadin poorer countries (in India it is sometimes claimed thatthere is an NGO in every village) but I know of no data onthe subject

National accounts consumption is typically estimated as aresidual using the commodity flow method Starting from anestimate of domestic production of each commodity netexports and government consumption are deducted as arethe amounts used in investment and intermediate consump-tion with the residual attributed to household (and NPISH)consumption Many of these calculations are done in phys-ical volumes so that estimation of consumption in currencyunits which is what can be compared with the surveysrequires the use of prices and price indices There are manyopportunities for error along this chain of calculation andin general there is no means (other than surveys) of cross-checking the final answer The measurement of prices is asurvey-based activity with its own sampling and nonsam-pling errors and it is sometimes difficult to be sure thatprices are those actually paid by consumers Not surpris-ingly the monetary value of NAS estimates of consumptionare subject to errors and to occasional large revisionSundaram and Tendulkar (2003) report that the Indian NASestimate of consumption of fruits and vegetables in 1993ndash1994 in nominal rupees more than doubled between the1998 and 1999 versions of the national accounts Theestimate for clothing fell by approximately a half and thatfor rent fuel and power rose by more than 40 Even withsome canceling out of pluses and minuses total consump-tion was revised upward by 14 an amount which if usedto calculate poverty rates would cut the Indian poverty rateby a little less than half

For food which is a large share of consumption in poorcountries domestic production is typically estimated bymultiplying the acreage of land under cultivation by anestimate of yields per acre The former comes from a landcensus or survey which in many countries is done quiteinfrequently whereas the latter comes from crop-cuttingsurveys themselves of mixed quality Data on governmentconsumption are usually relatively accurate as are importsand exports which typically are subject to direct monitoringby the government The same cannot be said for interme-diate (business) consumption which is often assessed byapplying various ratios to measured production These ratioscome from enterprise surveys or from input-output tablesOnce again these measures are often outdated For IndiaKulshreshtha and Kar (2002) write that their NAS consump-tion estimated ldquodepend on an assortment of direct andindirect estimates along with various rates and ratios someof which are based on the results of studies carried out in thedistant pastrdquo

The use of outdated ratios and correction factors isparticularly problematic when the economy is growing andits structure changing Kulsheshtra and Kar in their detailedcommodity-by-commodity comparison of food consump-tion in the NAS and the NSS in India find that one of thelargest discrepancies is for vanaspati a vegetable cookingoil that is widely used in restaurants This intermediate useof cooking oil should be deducted in the commodity flowcalculations but in fact there is no such correction in theIndian national accounts In consequence and because con-sumersrsquo expenditure in restaurants is already included inNAS consumption restaurant use of vanaspati is double-counted in the national accounts And because consumersswitch from domestic cooking to purchased meals as theyget better off the more rapid is the growth of the economythe larger will the overstatement of consumption becomeThe vanaspati example is an extreme case in that there is nocorrection for intermediate business consumption but thesame exaggeration of growth will be generated by the use ofoutdated rates and ratios to assess intermediate consumptionin an economy where growth is reallocating economicactivity from own production to the market

Overstatement of consumption and consumption growththrough a failure to capture intermediate consumption willalso lead to an overstatement of the level and growth rate ofexpenditure-based measures of GDP This exaggeration is inaddition to the exaggeration associated with the generalmovement of activity such as services from a nonex-changed to an exchanged basismdashfor example as a greatershare of food preparation is done by food vendors which iscounted in GDP rather than by family members which isnot Both come from the same fundamental trend which isthe increasing marketization complexity and roundabout-ness of production with economic development Note thatnot all of these errors in constructing consumption neces-sarily find their way into GDP For example how a com-modity flow is allocated between consumption and capitalformation will affect the estimation of both but not of theirsum

C Other Survey Issues

The two previous sections have documented what areperhaps the most likely candidates for explaining the diver-gence between national accounts and survey-based esti-mates of consumption However it should also be empha-sized that there are many other problems some of which areon the survey side It is clear that details of survey designmatter for the results and that protocols are not the sameacross countries or sometimes within countries over timeMany of these are discussed in more detail in Deaton andGrosh (2000)

Surveys often have less than complete coverage exclud-ing for example students the military and institutionalizedpersons expenditures by whom are included in NAS esti-mates of consumption In some cases survey coverage


excluded rural households or parts of the country that areexpensive or dangerous to visit

Survey questionnaires differ in the length of the recallperiod over which respondents are asked to report theirconsumption The choice of recall period is often thought toinvolve a tradeoff between accuracy of memory which callsfor a short period and the match between consumption andpurchases which is more accurate when averaged over along period But there is little understanding of the effects ofdifferent recall periods particularly in poor agriculturalsocieties In India between 1989 and 1998 the NSS exper-imented with different recall periods replacing the tradi-tional 30 day recall period for all goods with a 7 day recallperiod for food and tobacco and with a 365 day period fordurable goods and some other infrequently purchased itemsThe sample was randomly divided and half were given theold questionnaire and half the new so that it is possible tomake a clean evaluation of the effects of the change Theshorter reporting period increased reported expenditures onfood by around 30 and total consumption by about 17very much in the right direction to help resolve the discrep-ancy with the NAS Because there are many Indians close tothe poverty line the 17 increase was enough to reduce themeasured headcount ratio by a half removing almost 200million people from poverty What might seem to be anobscure technical issue of survey design can have a majoreffect on the measurement of poverty not only in India butin the world It should be noted however that the higherconsumption totals associated with the shorter recall periodalthough closer to the NAS estimates are not necessarilymore accurate Indeed the NSS has carried out a series ofcontrolled experiments in which for many foods the 30 dayreference period appears to be more accurate then the 7 dayperiod see NSSO Expert Group on Sampling Errors (2003)

Survey questionnaires also vary in the number of itemsthat are separately distinguished and there is some evidencethat the greater the degree of disaggregation the greater isthe measured consumption in total There is also no consis-tency in the treatment of seasonality some surveys visiteach household on several occasions throughout the yearbut most simply rely on spreading data collection through-out a calendar year a procedure that should not bias themean though there will be biases in higher-order statisticsIn some surveys respondents keep diaries of their purchasesover a period in others they make oral responses to inter-viewers based on recall Surveys vary on who is chosen asrespondent and whether one or more household membersare interviewed However well informed is the householdmember who reports purchases proxy reports on the pur-chases of other family members are likely to be less accu-rate than reports about the respondentrsquos own behaviorIndeed proxy reporting can plausibly contribute to a pro-gressively large share of consumption being missed overtime In a poor rural community where everyone eats fromthe same pot and food is nearly all of the budget the

housewifersquos report will be quite accurate This is much lessso in more diverse and better-off households with somefamily members working outside of the home and maintain-ing partial budgetary independence

In addition to the unit nonresponse discussed in section VA there is item nonresponse where household members failto report at least some expenditures or provide deliberatelymisleading reports for example on alcohol consumption oron various illegal items Finally and in parallel with thenational accounts there are difficulties in finding adequateprices for consumption items that are not purchased in themarket some surveys use market prices to impute homeproduction some use farm-gate prices and some use valu-ation techniques that are not clearly documented

It would be desirable if the international statistical com-munity could agree on a common set of best-practiceprotocols for household income expenditure surveys as aparallel to the SNA for the national accounts Unfortunatelymost of the problems discussed here are not sufficiently wellresearched or understood to admit of uncontroversial solu-tions and many statistical offices are stout defenders of theirown particular practices Yet as I shall argue in the nextsection only household surveys allow us to measure pov-erty so that the task of harmonization must be undertaken ifwe are to put global poverty measurement on a sound basis

VI Conclusions and Implications for theMeasurement of Poverty

The standard measures of poverty are based on countingthe number of people who live in households whose mea-sured per capita consumption is less than a poverty lineWhen rich households are less likely to cooperate with thesurvey than poor people survey-based estimates of con-sumption will understate mean consumption and overstatethe fraction of people in poverty Under some conditionsthe amount by which average consumption is understatedwill be larger the greater is the inequality of the truedistribution of consumption Unless consumption inequalityis increasing over time or the fraction of noncooperatinghouseholds is increasing income-based noncooperationdoes not in and of itself imply that ratio of measured to trueconsumption is increasing over time

National accounts estimates of consumption are typicallyalthough not always larger than survey-based estimatesand there is a tendency both across countries and over timewithin important countries for the NAS estimate of con-sumption to grow more rapidly than does the survey-basedestimate Survey-based estimates are subject to numerouserrors and inaccuracies but there are also problems withnational accounts estimates These are likely to understateconsumption in the poorest countries and to overstate therate of growth of average consumption both over time inpoor countries and in comparisons between poor and richcountries at a moment in time In part these systematicproblems in measuring the rate of growth of consumption


carry through to GDP whose growth rate is also systemat-ically biased upward I know of no plausible estimates of thesize of the bias

Given the conflict between survey and NAS estimates ofconsumption it is tempting to allow the NAS estimates toplay at least some role in poverty measurement instead ofusing only the survey data Indeed the combination ofmeans from the national accounts and Lorenz curves fromthe surveys has a long history including Ahluwalia et al(1979) the Indian government prior to 1993 and most ofLatin America until today In some cases this procedurewas adopted because the survey means were unavailableand in others such as the Indian case the practice wasabandoned after searching criticism of the quality of thenational accounts see in particular Minhas (1988) whomemorably describes the earlier practice as ldquomindless tink-eringrdquo In general there is argument on averaging of mul-tiple estimates (although only estimates of the same thing)so that extensive prior adjustment of any NAS mean wouldbe required before using it to scale up survey estimatesHowever there is need for a good deal of caution andmechanical use of unadjusted NAS means combined withsurvey-based estimates of the distribution around the meanwill certainly give the poor measures of poverty There areat least three reasons why

First and most generally the national accounts are de-signed to generate estimates of macroeconomic aggregatesnot estimates of poverty and the SNA rules are designedwith that in mind National accounts track money notpeople To take an example the SNA recognizes that pro-duction for own consumption is difficult to measure andrecommends that the effort be made only ldquowhen the amountproduced is likely to be quantitatively important in relationto the total supply of the good in the countryrdquo (OECD 2002p 179) Such a rule makes little sense when our primeobjective is to measure poverty At the other end of thespectrum items like FISIM and the rental value of owner-occupier homes are (properly) included although in mostcases they are either not consumed by the poor or make upless of their budgets In general the NAS is more likely tocapture larger transactions than smaller ones which is closeto the opposite of what happens in the surveys where largetransactors are the least likely to be included

Second the differences in coverage and definition be-tween NAS and surveys mean that even if everything wereperfectly measured it would be incorrect to apply inequalityor distributional measures which are derived from surveysthat measure one thing to means that are derived from thenational accounts which measure another When nationalaccounts and surveys are measuring different things it isperfectly possible for the poor to do less well than theaverage without any increase in measured inequality

To illustrate suppose that we are interested in measuringconsumption growth among the bottom 100p of the pop-ulation We have data on mean consumption from the

national accounts and data on the share sp of the bottom100p of households from surveys Mean consumption ofthe bottom 100p is then estimated to be

p spp (13)

so that the growth rate of consumption for the bottom groupis

p

p

sp

sp

(14)

In equation (14) the first term on the right-hand side comesentirely from the surveys and the second entirely from thenational accounts The survey mean and the direct surveymeasure of p sp the average consumption of thebottom 100p are discarded even though the poor rarelyrefuse to respond and provide accurate estimates of theirconsumption Moreover the validity of equations (13) and(14) depends on being able to apply the survey shares to theNAS means which assumes not only that the NAS meansare perfectly measured but that both are measuring thesame thing So even if we were to accept that NAS con-sumption is the concept that we want and even if we wereto believe that it is accurately measured the shares from thesurvey would be shares of consumption excluding con-sumption on rents of owner-occupiers excluding FISIM andthe profits of insurance companies and excluding the ex-penditures of NPISH Using the survey shares to allocateNAS consumption to the poor and nonpoor assumes thatthese items are distributed between poor and nonpoor in thesame way as are the goods measured in the survey anassumption that is not true

Third we must recognize that neither the mean consump-tion nor its distribution are accurately measured either inthe surveys or in the NAS A particular difficulty comesfrom the mechanical use of the distributional shares andGini coefficients that come from the DS and WIDER com-pilations (Shares can be calculated from Gini coefficients ifa particular distribution is assumed for example the lognormal) For most poor countries these measures are ofdubious quality as indeed is recognized by DS And neitherDS nor the WIDER compilation provides the informationthat would be required to make an informed judgment onthe way their numbers were calculated So if equation (14)is used to construct pp and the measures of sp are noisya regression of pp on will have a coefficient that isclose to 1 essentially by construction and the worse is themeasurement error the closer the estimate will be to 1 Sothere is no credibility to the claim that globalization hasbeen good for the poor based on a calculation that appliesbadly measured distributional shares to (upward-biased)measures of growth from the national accounts The glob-alization debate is serious enough that we must genuinelymeasure the living standards of the poor not simply assumethem We cannot prove that growth trickles down by


assuming that growth trickles down nor argue that global-ization has reduced poverty without measuring the livingstandards of the poor

If the task were the purely statistical one of estimatingmean consumption there would be much to be said forusing the average of the mean consumption from the sur-veys and that from the adjusted national accounts (Deaton2001) But if we need to measure poverty in a way that willconvince those who are skeptical of the idea that averagegrowth reaches the poor there is little choice but to use thesurveys This argument is reinforced by the fact that formany countries it is impossible to adjust NAS consumptionestimates to make them comparable with the survey totalsNone of this says that the surveys are correct nor thatcurrent measures of global poverty are doing a good job ofmeasuring the trends And because not every country has asurvey in every year they are clearly unsuitable for mea-suring year-to-year variations (see figure 3) There is toomuch incompatibility in survey design across countries Thedownward bias in survey measures of consumption almostcertainly biases upward the World Bankrsquos global povertyestimates and in as much as it is unlikely that all of thegrowth discrepancy between the surveys and the NAS is dueto faults in the latter the rate of poverty decline is likelydownward biased We need an international initiative toprovide a set of consistent international protocols for surveydesign as well as deeper study of the effects of nonsamplingerrors particularly noncompliance

REFERENCES

Adelman Irma and Cynthia Taft Morris Economic Growth and SocialEquity in Developing Countries (Stanford CA Stanford Univer-sity Press 1973)

Ahluwalia Montek S Nicholas G Carter and Hollis Chenery ldquoGrowthand Poverty in Developing Countriesrdquo Journal of DevelopmentEconomics 6 (1979) 299ndash341

Ahluwalia Montek S John H Duloy Graham Pyatt and T N Sriniva-san ldquoWho Benefits from Economic Development CommentrdquoAmerican Economic Review 70 1 (1980) 242ndash245

Aitchison J and Alan Brown The Lognormal Distribution (CambridgeUK Cambridge University Press 1969

Bardhan Pranab K ldquoOn the Incidence of Poverty in Rural India in theSixtiesrdquo Economic and Political Weekly 8 February special issue(1973) 245ndash254

Berg Andrew and Anne Krueger ldquoTrade Growth and Poverty ASelective Surveyrdquo IMF working paper no 0330 (2003)

Bhalla Surjit S Imagine There Is No Country Poverty Inequality andGrowth in the Era of Globalization (Washington DC Institute forInternational Economics 2002)

Blades Derek and David Roberts ldquoMeasuring the Non-observed Econ-omyrdquo OECD Statistics Brief no 5 (November 2002) 1ndash8

Bourguignon Francois and Christian Morrisson ldquoInequality amongWorld Citizens 1820ndash1992rdquo American Economic Review 92 4(2002) 727ndash744

Chen Shaohua and Martin Ravallion ldquoHow Well Did the Worldrsquos PoorestFare in the 1990srdquo Review of Income and Wealth 47 3 (2001)283ndash300

Chenery Hollis Montek S Ahluwalia Clive L G Bell John H Duloyand Richard Jolly Redistribution with Growth (Oxford OxfordUniversity Press for the World Bank 1974)

Cline William R ldquoDistribution and Development A Survey of theLiteraturerdquo Journal of Development Economics 1 4 (1975) 359ndash400

Deaton Angus 2001 ldquoCounting the Worldrsquos Poor Problems and PossibleSolutionsrdquo World Bank Research Observer 16 2 (Fall 2001)125ndash147ldquoAdjusted Indian Poverty Estimates for 1999ndash2000rdquo Economicand Political Weekly (25 January 2003) 322ndash326

Deaton Angus and Margaret Grosh ldquoConsumptionrdquo (vol 1 pp 91ndash133)in M Grosh and P Glewwe (Eds) Designing Household Ques-tionnaires for Developing Countries Lessons from Fifteen Years ofthe Living Standard Measurement Study (Washington DC WorldBank 2000)

Deaton Angus and Jean Dreze ldquoPoverty and Inequality in India aReexaminationrdquo Economic and Political Weekly (September 72002) 3729ndash3748

Deininger Klaus and Lyn Squire ldquoA New Data Set Measuring IncomeInequalityrdquo World Bank Economic Review 10 (1996) 565ndash591

Dollar David and Aart Kraay ldquoGrowth is Good for the Poorrdquo Journal ofEconomic Growth 7 (2002) 195ndash225

Fields Gary S ldquoWho Benefits from Economic Development A Reex-amination of Brazilian Growth in the 1960srdquo American EconomicReview 67 4 (1977) 570ndash582

Fishlow Albert ldquoBrazilian Size Distribution of Incomerdquo American Eco-nomic Review 62 12 (1972) 391ndash402

Garner Thesia I George Janini William Passero Laura Paszkiewicz andMork Vandemia ldquoThe Consumer Expenditure Survey in Compar-ison Focus on Personal Consumption Expendituresrdquo Paper pre-sented at the Federal Economic Statistics Advisory CommitteeMeeting Bureau of Labor Statistics Washington DC March 212003

Government of India National Accounts Statistics Ministry of Statisticsand Programme Implementation httpmospinicinnational_account_mainhtm (accessed March 1 2003)

Groves Robert M and Mick P Couper Nonresponse in HouseholdInterview Surveys (New York Wiley 1998)

Kakwani Nanak ldquoLorenz Curverdquo (vol 3 pp 242ndash244) in J Eatwell MMilgate and P Newman (Eds) The New Palgrave A Dictionaryof Economics (London and Basingstoke Macmillan 1987)

Keidel Albert ldquoChinarsquos GDP Expenditure Accountsrdquo China EconomicReview 12 4 (2001) 355ndash367

Kulshreshtha A C and A Kar ldquoEstimates of Food Consumption Ex-penditure from Household Surveys and National Accountsrdquo WorldBank Working paper httpwwwworldbankorgindiapoverty-workshop (2002)

Kuznets Simon ldquoEconomic Growth and Income Inequalityrdquo AmericanEconomic Review 45 1 (1955) 1ndash28

Maddison Angus Chinese Economic Performance in the Long Run(Paris OECD 1998)

Minhas B S ldquoValidation of Large-Scale Sample Survey Data Case ofNSS Household Consumption Expenditurerdquo Sankhya Series B50 Supplement (1988) 1ndash63

Minhas B S and S M Kansal ldquoComparison of NSS and CSO Estimatesof Private Consumption Some Observations Based on 1983 DatardquoThe Journal of Income and Wealth 11 (1989) 7ndash24

Mistiaen Johan A and Martin Ravallion ldquoSurvey Compliance and theDistribution of Incomerdquo Policy Research Working Paper 2956World Bank Development Research Group Washington DC2003

NSSO Expert Group on Sampling Errors ldquoSuitability of Different Refer-ence Periods for Measuring Household Consumption Results of aPilot Studyrdquo Economic and Political Weekly (January 25 2003)307ndash321

OECD Measuring the Non-observed Economy A Handbook (ParisOECD 2002)

Office for National Statistics Annual Abstract of Statistics (LondonHMSO 2003)

Pyatt Graham ldquoDevelopment and the Distribution of Living Standards ACritique of the Evolving Data Baserdquo Review of Income and Wealth49 3 (2003) 333ndash358

Ravallion Martin ldquoGrowth Inequality and Poverty Looking beyond theAveragesrdquo World Development 2911 (2001) 1803ndash1815ldquoMeasuring Aggregate Welfare in Developing Countries HowWell Do National Accounts and Surveys Agreerdquo this REVIEW 85(3) 645ndash652

Rawski Thomas G ldquoWhat is Happening to Chinarsquos GDP StatisticsrdquoChina Economic Review 12 4 (2001) 347ndash354


Sala-i-Martin Xavier ldquoThe disturbing lsquorisersquo of global income inequalityrdquoNBER working paper no 8904 (2002)

Srinivasan T N and Pranab Bardhan Poverty and Income Distributionin India (Calcutta Statistical Publishing Society 1974)

Sundaram K and Suresh Tendulkar ldquoNAS-NSS Estimates of PrivateConsumption for Poverty Estimation A Further Comparative Es-timationrdquo Economic and Political Weekly (January 25 2003)376ndash384

Taylor Lance and Edmar L Bacha ldquoThe Unequalizing Spiral A FirstGrowth Model for Belindiardquo Quarterly Journal of Economics 902 (1976) 197ndash218

Triplett Jack E ldquoMeasuring Consumption The Post-1973 Slowdown andthe Research Issuesrdquo Federal Reserve Bank of St Louis ReviewMayJune 1997) 9ndash42

Wu Harry X ldquoChinarsquos GDP Level and Growth Performance AlternativeEstimates and the Implicationsrdquo Review of Income and Wealth 464 (2000) 475ndash499

APPENDIX

Log-normally Distributed Income with Selective Compliance

Suppose that x is the logarithm of income and that x is normallydistributed in the population with mean and variance 2 The compli-ance probability as a function of income is given by equation (1) in themain text In general if the true density if f(x) the density function of thetruncated distribution is given by

f x p x f x

0

ps dFs

x f x (A-1)

In this normal case with the response function given by equation (1) in themain text the truncated density is

f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)

where p is the population-average compliance probability The second partof equation (A-2) can be rewritten

f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)

If we integrate f(x) over the full range of x we can derive an expressionfor the mean compliance probability

p (13) 13 exp 13 1

222 (A-4)

where the first term comes from integrating the first part of equation (A-2)and the second from integrating equation (A-3) These three equationscompletely characterize the truncated density f(x)

If we substitute equation (A-4) into (A-3) we get the density of anormal distribution with mean 2 and variance 2 scaled by thefactor

exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)

As 13 becomes large the second term in the denominator on the right goesto 1 while the first term goes to 0 Hence for large 13 with little of thedensity to the left of 13 the truncated density is approximatelyequation (A-3) which is approximately

f x 1

22exp 1

2 x 2

2 x 13 (A-6)

so that the truncated distribution of log income is also normal with thesame variance 2 as the true distribution but with mean 2 insteadof


(PPP) exchange rates allow comparisons of average livingstandards across countries in a way that is not vitiated by thegross inadequacies of conversions at market exchange ratesMaking comparisons in PPP units corrects or at leastdiminishes the gross understatement of living standards inpoor countries relative to rich and removes the spuriouscomponent of growth among poor countries that comesfrom the elimination of those differences with economicdevelopment PPP exchange rates were first used for globalpoverty estimates by Ahluwalia et al (1979) and their useis by now almost universal Second there has been anextraordinary growth in the number of household surveysavailable to the research community including severaldozen LSMS surveys For example the World Bankrsquos mostrecent set of poverty calculations use data from 297 surveysfrom 88 developing countries (Chen amp Ravallion 2001)Deininger and Squire (DS) (1996) have collected and tab-ulated data on more than 2600 Gini coefficients as well asmany measures of quintile shares the WIDER extensionincludes more than 5000 Gini coefficients The unit recorddata from many household surveys are now routinely avail-able to researchers including such previously inaccessibletroves as nearly 20 years of data from the Indian NationalSample Surveys back to the early 1980s Notable by itsexclusion is any similar access to Chinese official surveys

Yet the controversies are no more settled than they were30 years ago although there is certainly more commonground among economists than there is in the world at largeThe professional consensus based on the DS data and onwork by them and many others is that contrary toKuznetsrsquos hypothesis and contrary to beliefs in the 1970sthere is no general relationship between inequality andgrowth and certainly not one in which growth systemati-cally widens inequality as would be the case of growth leftthe poor behind From this two important propositionsfollow First at least on average (and much depends onwhether we are averaging over countries or people) growthis good for the poor (Dollar amp Kraay 2002 Ravallion2001) as is the growth that is arguably generated by greateropenness (Berg amp Krueger 2003) Second and again onaverage the fraction of people in poverty should decline asif growth were neutrally distributed In particular the rela-tively rapid growth in the developing world from 1980 to2000 must have brought about a rapid reduction in thefraction of the worldrsquos population that is poor And indeedcalculations using the Penn World Tables combined withinequality measuresmdashthe technique first used by Ahluwaliaet al (1979)mdashshow rapid poverty reduction in the 1980sand 1990s see Bhalla (2002) Sala-i-Martin (2002) andBourguignon and Morrisson (2002) According to thesecalculations not only has the proportion of poor in theworld declined but the decline has been rapid enough tooffset population growth so that the actual number of poorpeople in the world has fallen According to Bhalla the firstof the United Nations Millennium Development Goals

halving the number of people living on less than $1 a daybetween 1990 and 2015 had already been met when thegoal was announced

These optimistic calculations are starkly at odds with theWorld Bankrsquos numbers on global poverty The World Bankwhich is endorsed by the UN as official scorekeeper for thepoverty Millennium Development Goal uses householdsurvey data to measure the living standards of the poorignoring national accounts estimates and its calculationsshow relatively little poverty reduction in the 1990s Chenand Ravallion (2001) which provides the details of theBankrsquos calculations shows a reduction in the proportion ofthe poor living on less than $1 a day from 1987 to 1998from 283 to 235 they argue that this modest reductioncomes not from any expansion in inequality within coun-tries but from relatively slow growth in mean consumptionAcross their 88 countries the population-weighted rate ofgrowth in mean consumption was only 090 from 1987 to1998 compared with 33 growth in real per capita con-sumption in the Penn World Tables over the same periodThese estimates exclude the latest (1999ndash2000) Indian datawhose inclusion will increase the growth of the surveymeans over the 1990s There remains a large gap betweenon the one hand the direct assessment of the growth ofconsumption of the poor through surveys and on the otherhand the growth that is implied by the growth in averageaccompanied by no general increase in inequality

The plethora of new data has not resolved the contro-versy because the new sources are mutually contradictoryAccording to direct measurement in household surveysgrowth among the poor of the world has been sluggishcompared with the average growth rates of the countries inwhich they live Yet there is no documented increase ininequality that would resolve the discrepancy If we are toaccept the surveys growth in the world is a good dealslower than we are used to thinking from the nationalaccounts data and what growth there has been in the latesttwo decades has made only a modest dent in the level ofworld poverty If we accept the national accounts and donot challenge the conclusion that there is no general in-crease in inequality nor any correlation between growth andchanges in inequality then official poverty numbers areoverstated and we have already made rapid progress towardreducing poverty in the world This paper explores thesecontradictions empirically with an aim to providing asharper characterization and to advancing some first hypoth-eses about causes and possible remedies

A note of caution at the outset Because countries havevastly different populations statements about averages of-ten depend sharply on whether or not they are population-weighted A third of the worldrsquos poor live in two countriesIndia and China and the global poverty counts are muchaffected by what happens there When we are interested inthe well-being of the people of the world and in the effectsof statistical practice and statistical discrepancies on global









YearNo of

SurveysNo of

CountriesPopulation


Population ()

1979 3 3 035 901980 7 6 133 3411981 5 5 027 671982 3 3 033 811983 2 2 097 2361984 7 5 048 1141985 13 11 159 3741986 21 18 175 4031987 23 21 182 4121988 26 24 292 6491989 31 28 214 4571990 23 20 169 3511991 29 26 195 3861992 37 34 216 4211993 45 41 261 4991994 35 30 322 6021995 51 45 367 6771996 48 44 382 6831997 43 38 338 6111998 57 53 386 7031999 22 19 216 3912000 26 23 354 636









No ofSurveys


MeanRatio

StandardError

StandardDeviation

MeanRatio

StandardError

StandardDeviation


All 277 0860 (0029) 0306 0779 (0072) 0191



All 266 0904 (0034) 0290 1008 (0044) 0174


Income to GDP

All 272 0569 (0023) 0203 0542 (0023) 0113
























Growth Rate (yr)


Preference

Surveys withIncome

Preference

PWT61Matching































A Unit Nonresponse



f y y f y

y0

y1

f y y dy

y fy

(1)


f y f y y

f y (2)




F y F y (3)



L p F1 p

y

(4)



y dF1 y y dF2 y (5)




1 2 (7)



x1 x v 13exp[x v 13)] x v 13

(8)



ln

2 (9)





x 1 x v 13

expx v 13

2x v 132 x v 13

(10)


2 2

1 2 (11)


v v 2 13

1 2 (12)
















































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)










YearNo of

SurveysNo of

CountriesPopulation


Population ()

1979 3 3 035 901980 7 6 133 3411981 5 5 027 671982 3 3 033 811983 2 2 097 2361984 7 5 048 1141985 13 11 159 3741986 21 18 175 4031987 23 21 182 4121988 26 24 292 6491989 31 28 214 4571990 23 20 169 3511991 29 26 195 3861992 37 34 216 4211993 45 41 261 4991994 35 30 322 6021995 51 45 367 6771996 48 44 382 6831997 43 38 338 6111998 57 53 386 7031999 22 19 216 3912000 26 23 354 636









No ofSurveys


MeanRatio

StandardError

StandardDeviation

MeanRatio

StandardError

StandardDeviation


All 277 0860 (0029) 0306 0779 (0072) 0191



All 266 0904 (0034) 0290 1008 (0044) 0174


Income to GDP

All 272 0569 (0023) 0203 0542 (0023) 0113
























Growth Rate (yr)


Preference

Surveys withIncome

Preference

PWT61Matching































A Unit Nonresponse



f y y f y

y0

y1

f y y dy

y fy

(1)


f y f y y

f y (2)




F y F y (3)



L p F1 p

y

(4)



y dF1 y y dF2 y (5)




1 2 (7)



x1 x v 13exp[x v 13)] x v 13

(8)



ln

2 (9)





x 1 x v 13

expx v 13

2x v 132 x v 13

(10)


2 2

1 2 (11)


v v 2 13

1 2 (12)
















































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)








No ofSurveys


MeanRatio

StandardError

StandardDeviation

MeanRatio

StandardError

StandardDeviation


All 277 0860 (0029) 0306 0779 (0072) 0191



All 266 0904 (0034) 0290 1008 (0044) 0174


Income to GDP

All 272 0569 (0023) 0203 0542 (0023) 0113
























Growth Rate (yr)


Preference

Surveys withIncome

Preference

PWT61Matching































A Unit Nonresponse



f y y f y

y0

y1

f y y dy

y fy

(1)


f y f y y

f y (2)




F y F y (3)



L p F1 p

y

(4)



y dF1 y y dF2 y (5)




1 2 (7)



x1 x v 13exp[x v 13)] x v 13

(8)



ln

2 (9)





x 1 x v 13

expx v 13

2x v 132 x v 13

(10)


2 2

1 2 (11)


v v 2 13

1 2 (12)
















































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)























Growth Rate (yr)


Preference

Surveys withIncome

Preference

PWT61Matching































A Unit Nonresponse



f y y f y

y0

y1

f y y dy

y fy

(1)


f y f y y

f y (2)




F y F y (3)



L p F1 p

y

(4)



y dF1 y y dF2 y (5)




1 2 (7)



x1 x v 13exp[x v 13)] x v 13

(8)



ln

2 (9)





x 1 x v 13

expx v 13

2x v 132 x v 13

(10)


2 2

1 2 (11)


v v 2 13

1 2 (12)
















































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)
















Growth Rate (yr)


Preference

Surveys withIncome

Preference

PWT61Matching































A Unit Nonresponse



f y y f y

y0

y1

f y y dy

y fy

(1)


f y f y y

f y (2)




F y F y (3)



L p F1 p

y

(4)



y dF1 y y dF2 y (5)




1 2 (7)



x1 x v 13exp[x v 13)] x v 13

(8)



ln

2 (9)





x 1 x v 13

expx v 13

2x v 132 x v 13

(10)


2 2

1 2 (11)


v v 2 13

1 2 (12)
















































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)










Growth Rate (yr)


Preference

Surveys withIncome

Preference

PWT61Matching































A Unit Nonresponse



f y y f y

y0

y1

f y y dy

y fy

(1)


f y f y y

f y (2)




F y F y (3)



L p F1 p

y

(4)



y dF1 y y dF2 y (5)




1 2 (7)



x1 x v 13exp[x v 13)] x v 13

(8)



ln

2 (9)





x 1 x v 13

expx v 13

2x v 132 x v 13

(10)


2 2

1 2 (11)


v v 2 13

1 2 (12)
















































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)

























A Unit Nonresponse



f y y f y

y0

y1

f y y dy

y fy

(1)


f y f y y

f y (2)




F y F y (3)



L p F1 p

y

(4)



y dF1 y y dF2 y (5)




1 2 (7)



x1 x v 13exp[x v 13)] x v 13

(8)



ln

2 (9)





x 1 x v 13

expx v 13

2x v 132 x v 13

(10)


2 2

1 2 (11)


v v 2 13

1 2 (12)
















































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)



















A Unit Nonresponse



f y y f y

y0

y1

f y y dy

y fy

(1)


f y f y y

f y (2)




F y F y (3)



L p F1 p

y

(4)



y dF1 y y dF2 y (5)




1 2 (7)



x1 x v 13exp[x v 13)] x v 13

(8)



ln

2 (9)





x 1 x v 13

expx v 13

2x v 132 x v 13

(10)


2 2

1 2 (11)


v v 2 13

1 2 (12)
















































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)









A Unit Nonresponse



f y y f y

y0

y1

f y y dy

y fy

(1)


f y f y y

f y (2)




F y F y (3)



L p F1 p

y

(4)



y dF1 y y dF2 y (5)




1 2 (7)



x1 x v 13exp[x v 13)] x v 13

(8)



ln

2 (9)





x 1 x v 13

expx v 13

2x v 132 x v 13

(10)


2 2

1 2 (11)


v v 2 13

1 2 (12)
















































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)




F y F y (3)



L p F1 p

y

(4)



y dF1 y y dF2 y (5)




1 2 (7)



x1 x v 13exp[x v 13)] x v 13

(8)



ln

2 (9)





x 1 x v 13

expx v 13

2x v 132 x v 13

(10)


2 2

1 2 (11)


v v 2 13

1 2 (12)
















































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)





x 1 x v 13

expx v 13

2x v 132 x v 13

(10)


2 2

1 2 (11)


v v 2 13

1 2 (12)
















































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)












































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)



































p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)




























p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)



















p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)









p spp (13)


p

p

sp

sp

(14)






REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)





REFERENCES












































APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)









APPENDIX



f x p x f x

0

ps dFs

x f x (A-1)


f x 1

p 22exp 1

2 x

2 x 13

1

p 22exp 1

2 x

2 exp[x 13 x 13

(A-2)


f x 1

p 22exp 1

2 x 2

2 exp 13

1

222 x 13

(A-3)


p (13) 13 exp 13 1

222 (A-4)



exp13 1

222

(13) 13 exp 13 1

222

1

(13) exp13 1

222 13

(A-5)


f x 1

22exp 1

2 x 2

2 x 13 (A-6)



Documents

GROWTH IN A POOR WORLD) - Princeton Universitydeaton/downloads/measuring poverty... · 2008-12-17 · The Review of Economics and Statistics V OL.LXXXVII NF EBRUARY 2005 UMBER 1 MEASURING