Crying Wolf – Are We Over-counting the Number of Poor People in India?

April 15, 2011

By The Sanhati Collective

Abstract: The well-known and growing discrepancy between average consumption expenditure computed from the National Sample Survey Organization data and those reported by the National Account Statistics is probably reflecting the growing under-reporting by the rich and over-reporting by the poor. Once this is corrected in a meaningful manner, even the conservative estimates of the poverty ratios by the Planning Commission might increase, not decrease as supporters of neoliberalism claim; the reported measures of inequality will go up unambiguously.

1. Introduction

It is not uncommon for members of India’s elite, bouyed by two decades of neoliberal growth, to forget that there are still desperately poor people in the country, and extremely large numbers of them. Generalizing their own situation, marked by steadily rising incomes, wealth and consumption expenditures, the rising middle class forgets that the neoliberal growth process has structurally bypassed the vast majority. No wonder, economist Jean Dreze noted in an interview to Tehelka last year, that “the middle class has lost track of how poor this country is.” [1]

Of course, champions of neoliberalism in India go further than just losing track of “how poor this country is”, they make a positive claim. They claim that the country is actually much less poor that what most political activists, economists, and policymakers seem to accept. A case in point is the consulting editor of The Economic Times, Swaminathan S. Anklesaria Aiyar, the poster boy of neoliberalism in India.

In a recent column in the Times of India, he argues that India has been systematically over-counting its poor [2]. He thinks that the data on consumption expenditure that emerge from the periodic sample surveys conducted by the National Sample Survey Organization (NSSO) seriously underestimate “true” consumption expenditures of families, especially of the poor. If the consumption expenditure data were corrected for these underestimation errors, as in Mr. Aiyar’s opinion economist Surjit Bhalla has done, correct figures for poverty (for instance, the head count ratio) would emerge. These correct poverty ratios, according to Mr. Aiyar, would be about a third of even the conservative officially accepted figures. If Mr. Aiyar is right, then, we have been overestimating poverty in India by close to 200 percent!

Being a tall claim by all means, Mr. Aiyar’s assertions bear close scrutiny. Let us begin by noting that his argument rests on two separate claims: (a) there is serious underestimation of consumption expenditure figures, and (b) if we correct for this underestimation, measures of poverty like the head count ratio will go down substantially.

2. Underestimation of Consumption Expenditure?

Mr. Aiyar presents three arguments in support of his first claim: (a) the discrepancy between the average consumption expenditure as recorded by the National Account Statistics (NAS) and the National Sample Survey (NSS), (b) strong incentives for the poor to get them counted in the below poverty line (BPL) category, and thereby increase the poverty ratio, and (c) apparent growth of rural demand. Let us take up each.

2.1 NAS-NSS Discrepancy

Even though there is some discussion about the relative merits of the NAS versus the NSS, the first point is relatively uncontroversial: it is a well known fact that average consumption expenditure measured by the NAS is not only higher than that measured by the NSS but that the discrepancy has been increasing over time. This is a well-known fact and there is hardly any disagreement on this. But this fact of the discrepancy between the NAS and NSS by itself will not give Mr. Aiyar the conclusions he desires; the impact of the discrepancy depends on how under-reporting is distributed across expenditure classes. Mr. Aiyar realizes this and so proceeds to explain this relatively noncontroversial fact in the particular way – using the “incentive argument” – that will more or less deliver the conclusion that poverty ratios are overestimating true poverty. So let us take a closer look at this one.

2.2 The Incentive Argument

Mr. Aiyar’s second argument, the “incentive argument”, rests on the claim that “Government policies have made it profitable for rural folk to exaggerate their poverty. “ How does Mr. Aiyar reach this profound conclusion? By analogy. According to Mr. Aiyar, just like “the very announcement of a caste census could encourage people to claim, fraudulently, that they belong to a caste entitled to reservations”, the very act of a sample survey by the NSSO (to collect information on consumption expenditure, for instance) would strongly dispose people towards reporting themselves as poor once they knew that tons of “freebies” would be showered on the poor by the State. Mr. Aiyar’s explanation for the discrepancy between the NAS and NSS consumption figures, therefore, rests crucially on under-reporting by the poor. This explanation is flawed on several counts.

2.2.1 BPL Census-NSSO Confusion

The most important problem is the confusion, intended or otherwise, between the BPL category and poverty ratios. It is a well-known fact that government benefits provided to the poor are mostly dependent upon one’s inclusion in the BPL category. The term BPL, one must recall, is heard mainly in discussions of the public distribution system (PDS). After the government scrapped universal public distribution in favour of targetted PDS, the artefact of BPL was created. Not only food, but a number of other government handouts are channelled through the BPL list. Indira Awas Yojna, a housing scheme for the poor, for instance is a case in point. Within the BPL there is a category of super poor, who are provided food more regularly, at a lower price. This scheme, meant for a subsection of the BPL, is called Antodaya Anna Yojna. In a nutshell, many government benefits are routed through the BPL list. If villagers seek to fake their real economic status to corner handouts, they would do so by getting themselves into the BPL list.

And here is the catch. The BPL list is prepared on the basis of the BPL census. This is different from the survey which estimates the poverty ratio (head count ratio). The latter is conducted by the NSSO, the National Sample Survey Office. The calculation of the poor from the data falls under the aegis of the Planning Commission. The BPL census on the other hand is undertaken by the Union Ministry of Rural Development through the state Rural Development Department. Aiyar cleverly tries to invoke the incentive compatibility argument and show that people have an incentive to under-report their economic status, thus leading to overestimation of poverty. But the two ends of the logical chain do not join as the surveys are different. Indeed there are large differences between poverty estimates by these two methods (although there exists a curious government regulation that the BPL poverty figure of a state must be within 10% range of the Planning Commission figure!).

2.2.2 Exclusion and Inclusion Errors

What about the point that people may under-report in the BPL census to corner benefits? Although this may not affect poverty count, this would lead to what is known as inclusion error or type I error – the error which occurs when undeserved non-poor get the benefits. This itself is a serious concern even if we keep aside the poverty debate for the time being.

The first point to note in this regard is that BPL censuses – the last one was conducted in 2002 – are a score-based system. Households are assigned scores depending upon different criteria such as land ownership, indebtedness, income, consumer durables ownership, educational attainment, etc. Higher scores indicate lower poverty. Researchers have pointed out that the whole process is cumbersome and flawed. There exists type I error without any doubt. As the BPL census reeks of an arbitrary and convoluted bureaucratic process (apart from being imagined as a fit-for-all gateway through which all provisions must be distributed [3]) the non-poor households enjoy an upper hand in fudging the data in their favour. Caste, class, control of the village panchayat etc. play an important role here. It is however found that the error is far more grave on the other side. Exclusion of the poor – the type II error – is much higher than the inclusion of the non-poor. At the all-India level the numbers for type II and type I errors are 60.4% to 26.3%, although for two states the latter is higher [4]. If these errors in BPL census are corrected, rather than going down, the proportion of people in the BPL category would go up.

The second point to note about Mr. Aiyar’s argument about the incentive to under-report by poor people is the following: if families have an incentive to under-report their consumption expenditure to increase their chance of being counted as poor, then the quantum of under-reporting will be an increasing function of family income (or consumption expenditure). Families which are desperately poor will under-report by a smaller amount than a family which is closer to (or even above) the official poverty line. Why? Families which are extremely poor have monthly consumption expenditure so far below the poverty line that they will anyway be counted as poor and so their incentive to under-report is rather weak. On the other hand, families which are above the poverty line could be counted as poor only if they under-report by an amount which is the excess of their monthly expenditure over the poverty line. Since the poverty line is not known among the population at large, to increase the chances of being counted as poor, a richer family would need to under-report by a larger amount than a poorer family, other things being equal. But this is just another way of saying that the quantum of under-reporting increases with income (or expenditure).

2.2.3 Erosion of Non-market Entitlements

When we wish to study consumption expenditure distributions over time, another consideration will have to be brought into the picture: changing extent of non-market provisioning of subsistence needs of the rural poor. In a relatively less-developed capitalist context, a large portion of the subsistence needs of poor families are met outside the logic of the market (through their customary access to common property resources like forests, rivers, ponds, etc.). With the development of capitalism, as the rural economy becomes increasingly enmeshed in the logic of the market, the access of the poor to the common property resources is gradually curtailed; hence, the poor are forced to satisfy their subsistence needs increasingly through market transactions. Compared to non-market provisioning, market transactions are much more likely to be recorded in sample surveys; hence, increasing penetration of the market will, over time, inflate the recorded consumption expenditures of the rural poor (even though their real consumption might not increase). In any case, this will act as a counter-weight to any tendency to under-report consumption expenditures by the poor and might very well nullify the effect of under-reporting altogether. Over time, in our opinion, therefore, the poorer households will report their consumption expenditures with increasing accuracy or will, in fact, over-report changes in real consumption.

2.2.4 Under-reporting by the Rich

The upper ends of the expenditure class spectrum, i.e., the rich, have a different dynamic altogether. It is a well-known fact that families at the upper ends of the income/expenditure classes (a) are seriously under-sampled in sample surveys (by some accounts up to 70 percent of rich households don’t respond), and (b) systematically under-report their income or consumption expenditures (mainly to evade taxes). It does not help to point out, as Mr. Aiyar does, that tax rates have declined in India over time: as long as tax rates are positive, a rich person will always have the incentive to under-report, if she can, her income or consumption expenditure to save on tax payments. But more importantly, over the period when tax rates have declined, the ease with which money can be transferred abroad to tax heavens have also increased. According to a Washington-based think tank, Global Financial Integrity, India witnesses about Rs. 240 crore of illicit fund flow a day; that is more than the total food subsidy bill of the government [5]. Thus, the rich have enough incentives to under-report their true expenditures.

For both these reasons, i.e., under-sampling and under-reporting, when serious scholars wish to draw inferences about income (or consumption expenditure) distribution at the upper end of the income spectrum they use income tax return data and not sample surveys. Mr. Aiyar, we believe, will benefit from a study of a not-so recent paper on this issue by Abhijit Banerjee and Thomas Piketty [6].

2.3 Growth of Rural Demand?

Mr. Aiyar’s third argument related to the issue of underestimation of consumption expenditure is the apparent growth of aggregate demand arising from the rural sector of the economy. How can this square with growing poverty, questions Mr. Aiyar. Since aggregate rural demand is growing it must be the case that poverty is going down. This conclusion would probably be valid if rural society was homogeneous and was not divided along lines of class, caste and ethnicity. For, growing aggregate rural demand is compatible with rising inequality within rural society, so that there is not much positive impact on poverty. The point is simple: if growth is accompanied by rising inequality, the poverty reduction is severely muted. It is hardly surprising that Mr. Aiyar does not once refer to issues of inequality!

We can draw two conclusions from this discussion. First: there are no plausible grounds to suppose that poor people systematically under-report their consumption expenditures in the NSSO surveys; if anything, over time, the accuracy of their reported consumption expenditures have either become more accurate or have been over-reported. Second: serious scholars of development think that the problem of under-reporting occurs much more seriously on the other end of the income spectrum: rich people have incentives to under-report their income or consumption expenditure (mainly to evade taxes) [7].

3. How to deal with Underestimation?

This brings us to the second claim that underlies Mr. Aiyar’s argument: if the underestimation of consumption expenditure is corrected then the poverty ratios will inevitably go down. This is simply not true. Why? Because the effect of correcting for the underestimation of consumption expenditures on poverty ratios will depend on the method adopted for implementing the correction. For instance, if the entire distribution of consumption expenditure were shifted to the right by the amount of the discrepancy between average consumption expenditure figures from NAS and NSS, this would correct the problem of underestimation in a certain way; as a result, it would reduce the poverty ratio by a huge amount. This is precisely what Surjit Bhalla does and hence it comes as no surprise that he reports poverty ratios which are a third of the officially accepted figures.
But this method of “correcting” for the underestimation of consumption expenditure rests on a dubious assumption; it rests on the assumption that the under-reporting of consumption expenditure is uniformly distributed across expenditure classes. This is clearly not true; under-reporting is a function of family expenditure. In fact, by a plausible argument that we have sketched above, under-reporting is an increasing function of family expenditure. This means that increasingly larger shares of the discrepancy between the NSS and NAS figures for average consumption expenditure in India is accounted for by richer rather than poorer families. This means that one cannot simply shift the entire distribution of consumption expenditure (that we get from the NSS) to the right by an amount which makes the average of the NSS distribution equal to the average that we get from the NAS. Hence, the lower figures for poverty ratios that Mr. Aiyar refers to (borrowing from the work of economist Surjit Bhalla) are bogus.

It is true that Mr. Bhalla put forward a more “sophisticated” version of the same argument in a later paper [8]. The sophisticated argument ran as follows: in the first step, he computed the ratio of total expenditure on various broad categories of consumption items (like food, clothing, etc.) as reported in the NAS and the NSS; for all items of consumption, this ratio is larger than 1. In the second step, for each household in the NSS survey, he multiplied the expenditures on the broad categories of consumption by the ratio computed in step 1 and then added up the “corrected” expenditures to arrive at the total consumption expenditure of the household. He considers this sum to be the “true” consumption expenditure of the household. Since the ratio for each broad category of consumption was greater than 1, the “true” consumption expenditure was higher than the reported consumption expenditure for each and every household. Lo and behold, poverty goes down! By shifting the terrain of under-reporting from households to items of consumption, Mr. Bhalla has cleverly avoided the issue that under-reporting is a function of household expenditure. In sum, even the “sophisticated” version of Mr. Bhalla’s argument to deal with the problem of under-reporting seems totally unconvincing.

4. Conclusion: Counting the Poor

But even if we were to properly correct for under-reporting (by, for instance, accounting for the positive dependence of under-reporting on household expenditure), looking only at the proportion of the population below the official poverty line would still be seriously misleading. If one wishes to draw conclusions about the levels of deprivation faced by the Indian population we need to move beyond the official poverty line. This is because, by all reasonable accounts, the poverty line is much, much too low (which, by the way, even Mr. Aiyar accepts). A better way to measure the economically vulnerable population in India would be to use the much more intuitively appealing categorization presented by Sengupta, Kannan and Raveendran (2008): all households whose monthly per capita expenditure was below twice the official poverty line in 2004-05 could be considered the poor [9]. Given that in 2004-05, the official poverty line in rural India was Rs. 356.3 per capita per month, and the corresponding poverty line for urban India was Rs. 538.6 per capita per month, the cut-off proposed by Sengupta, Kannan and Raveendran (2008) seem eminently reasonable. Of course, we can further divide the population into finer groups as presented in Table 1.

Table 1: Categories of the Indian Population


Extremely PoorMPCE <= 0.75 OPL
Poor0.75 OPL < MPCE <= OPL
MarginalOPL < MPCE <= 1.25 OPL
Vulnerable1.25 OPL < MPCE <= 2 OPL
Middle Income2 OPL < MPCE <= 4 OPL
High Income> 4 OPL


MPCE = monthly per capita expenditure; OPL= official poverty line

Thus, what we wish to present as the broader category of poor (let us call it “broad poor”) is comprised of the following four categories in Table 1: (a) extremely poor, (b) poor, (c) marginal, and (d) vulnerable. How does India do in terms of these categories and the broad category of poor after the early 1990s?

Figure 1 and 2 shows the evolution of these categories graphically.

countingpoor1.gif

countingpoor2.gif

In terms of proportions, Figure 1 shows that though the share of the extremely poor and poor (which is what would be considered the BPL population) has declined, the proportion of the “broad poor” has remained essentially unchanged between 1993-94 and 200405 at close to 80%. Thus, by any reasonable measures, what could be considered the share of poor and economically vulnerable people has not changed over this period of blazing high economic growth. Figure 2 is more damning: the absolute number of the “broad poor” (poor and economically vulnerable) has increased between 1993-94 and 2004-05 from about 811 million to 836 million.

Why do we stop at 2004-05? Because right now we are awaiting the results of the next “thick” round of consumption expenditure survey by the NSS which was conducted in 2009-10 (the last one was conducted in 2004-05). If the past decade and a half is anything to go by, we will see a repeat of the pattern that is presented in Figure 1 and 2: the share of the “broad poor” will diminish only marginally if at all and their actual number will increase. It is only a supreme sleight of hand that can, in the face of this reality, claim that India has been overestimating its poor.

Notes

1. http://www.tehelka.com/story_main47.asp?filename=Ne061110The_middle_class.asp

2. http://blogs.timesofindia.indiatimes.com/Swaminomics/entry/making-profit-out-of-poverty

3. http://www.thehindu.com/opinion/lead/article406431.ece

4. Mahamallik, M. G. Sahu, 2011. “Identification of the Poor: Errors of Exclusion and Inclusion,” Economic and Political Weekly, February 26.

5. http://www.thehindu.com/opinion/columns/sainath/article1514987.ece?homepage=true

6. Banerjee, A. and T. Piketty. 2005. “Top Indian Incomes, 1922-2000,” The World Bank Economic Review, 19(1), pp. 1-20.

7. It might be noted in passing that the bad-mouthing of NSS data by Mr. Aiyar has a chequered recent history. Since the Arjun Sengupta Commission had used the NSS data (both consumption expenditure surveys and the employment-unemployment surveys) to arrive at the striking figure of 77% of the population spending less than Rs. 20 a day in 2004-05, neoliberals have scoffed at the “quality” of NSS data. No less than the home minister, Mr. P. Chidambaram, gave voice to the “argument” that NSS data gives an overestimate of the poverty ratio. Mr. Aiyar is, here, trying to substantiate the charge that Mr. P. Chidambaram had made casually.

8. Bhalla, S. 2003. “Recounting the Poor: Poverty in India, 1983-99,” Economic and Political Weekly, January 25-31.

9. Sengupta, A., Kannan, K. P. and G. Raveendran. 2008. “India’s Common People: Who Are They, How Many Are They and How Do They Live?” Economic and Political Weekly, March 15, pp. 49-63.

No Comments »

Leave a comment