Poverty & Inequality

Statistical priorities for the ‘Great Indian Poverty Debate 2.0’

  • Blog Post Date 15 October, 2022
  • Perspectives
  • Print Page
Author Image

Himanshu .

Jawaharlal Nehru University

himansu2@gmail.com

In the final post of a six-part series on the estimation of poverty in India, Himanshu summarises attempts by researchers to estimate poverty using three varied approaches, given the lack of official consumption expenditure data. He considers the validity of recent estimates against the reality of declining wages. His view remains that if the outstanding statistical issues outlined here aren’t resolved, the upcoming NSS estimates on consumption expenditure will not end, but likely spark, a third round of the poverty debate.

Controversies surrounding poverty estimates in India are not new. These have concerned the poverty line as much as the measure of distribution used for estimating poverty, which remains consumption expenditure from the National Statistical Organisation (NSO). The long history of the debate by academics and policy makers in India and abroad has certainly helped to enrich the methodology of collecting consumption expenditure data, but has also contributed to repeated revisions in the poverty line, both of which remain crucial for any estimate of poverty. The last such controversy was regarding the 1999-2000 consumption survey which deviated from the existing practice of collecting consumption expenditure data by using a survey schedule which had questions from two different recall periods, leading to contamination of the consumption estimates and its distribution.

More recently, a number of estimates have been presented on poverty in India. Unlike the first ‘Great Indian Poverty Debate’ which arose due to the completely new approach to estimating consumption expenditure by the NSO, the current ‘Great Indian Poverty Debate 2.0’ is an attempt by scholars to fill the gap arising from the absence of any consumption survey since 2011-12. Part of the reason for the gap is the unexplained and abrupt decision by the Indian government to withhold the 2017-18 consumption survey, which was almost similar to the previous one in 2011-12. The claims of doubts regarding ‘data quality’ were neither explained, nor was the data made available to the research community. Such a long gap is unusual in the history of the statistical system of the country. The next survey of 2022-23 is now in the field and so any concrete assessment of what happened to poverty is only possible after the 2022-23 data is made available.

Poverty estimates using NSO data

However, absence of data has not deterred attempts by individual researchers to estimate poverty after 2011-12. Three kinds of attempts have been made: the most straight forward one is the estimation of poverty using the leaked data of 2017-18 Consumer Expenditure Survey (CES). Although not accepted officially, this remains the only valid comparison of any estimate of poverty after 2011-12. Based on leaked data on grouped distribution of consumption expenditure from 2017-18, Subramanian (2019) reported a rise in poverty by 4 percentage points, with poverty at 35% for 2017-18.

The second type of exercise is again using NSO consumption data – not from the usual consumption expenditure rounds, but from consumption estimates from abridged consumption questions as part of other socio-economic surveys of the NSO. Mehrotra and Parida (2021), and Debroy, Barnwal and Sinha (2022) use consumption data from the Periodic Labour Force Survey (PLFS) of 2019-20 and 2020-21 respectively, to arrive at opposite conclusions. Mehrotra and Parida report a rise in poverty, with 2019-20 with poverty headcount at 25.9%. On the other hand, Debroy et al.  report a decline in poverty in 2020-21 compared to 2011-12, with the estimate at 17.9%. While there are certainly issues of comparability with regard to the approaches adopted by the latter two estimates, these are not very different from comparable estimates of consumption expenditure from 2014.

Poverty estimates using a hybrid approach

It is, however, the third approach of poverty estimation which has recently attracted attention. These are two separate estimates which use a hybrid approach using external data sources to construct estimates of consumption expenditure comparable to the 2011-12 estimates, to arrive at strikingly different results. The first paper in this set is by Bhalla, Bhasin and Virmani (hereafter BBV). While the results of the paper are striking and provocative, to say the least, the methodology adopted by BBV is an old one. This method, which uses consumption aggregates from the Private Final Consumption Expenditure (PFCE) aggregates from the National Accounts Series (NAS) to adjust the consumption estimates obtained from the CES, leads to the conclusion that by the commonly used $1.90 poverty line, India has eliminated extreme poverty (with the poverty headcount in 2019-20 at 1.4% only).

This exercise of adjusting NSO consumption aggregates by using PFCE from the National Accounts (NA) as an external benchmark has been examined multiple times, not only by individual researchers but also by several expert government committees. Most of these committees included experts from the NSO and the NA, and have reiterated the conclusion that the two estimates are conceptually different and hence not comparable. The last such committee which gave its report was the Adhikari committee, which examined this question and submitted its report in 2015 (see Central Statistics Office, 2015). The previous such exercise was by the National Statistical Commission, incidentally during the Bhalla’s tenure as a member of the commission. All these reports, starting from the seminal paper by Minhas (1988), are in the public domain.1

The BBV approach of using this method is not only incorrect in the light of the humongous evidence against it, but is also disingenuous in the absence of any reference to any of these studies. The selective use of evidence by BBV is also evident in the case of their second claim regarding the imputation of in-kind transfers, primarily through the Public Distribution System (PDS). A large literature exists which quantifies the impact of in-kind transfers on poverty estimates (see Drèze and Khera 2013, Himanshu and Sen 2013). While BBV use the imputation of in-kind transfers to justify their estimates, this again is based on assumptions on leakages and access to the PDS that are difficult to verify using secondary sources. 

The second study in this series is the approach by Roy and van der Weide (hereafter RW), who are both affiliated with the World Bank (WB). The RW study also uses a hybrid approach, but unlike BBV they used the consumption data of the Consumer Pyramid Household Survey (CPHS) of the Centre for Monitoring Indian Economy. RW recognise the issue of bias as far as the CPHS data is concerned, an issue which has been brought out by several authors (see Pais and Rawal 2021, Sanyal 2021, Somanchi 2021). However, their approach to correct the biases is again based on strong assumptions, the validity of which has been questioned. Their method assumes the external survey benchmarks to be correct and uses a reweighting mechanism to force the CPHS to arrive at estimates similar to the benchmark estimates. While the approach may be technically valid, the underlying assumptions of a stable relationship between non-monetary and monetary aggregates2 across surveys and across time is not entirely justified. This was clear even during the controversy surrounding the 1999-00 consumption survey estimates. Their method with all the caveats results in a poverty headcount ratio of 10.2% in 2019-20, compared to 22.5% in 2011-12.

Unlike the poverty estimates, which use only the NSO estimates, the hybrid methods have tried to create a synthetic distribution of consumption expenditure that is closest to the actual consumption expenditure surveys of the NSO. The poverty estimates using only NSO estimates have their own problems – most importantly the comparability of consumption aggregates using different survey instruments and recall period. It is well known that use of an abridged consumption schedule not only changes the aggregate estimate of consumption expenditure, but the distribution of consumption expenditure is also very different. This poses problems of comparing consumption expenditure and poverty estimates over time.

On the other hand, the hybrid estimates pose a different set of problems. The issue of comparability across NSO and PFCE estimates is well known and does not need repetition here, but even the RW approach suffers from serious issues of comparability. The underlying assumption in the case of RW is that there exists a true estimate of consumption expenditure and a distribution across households that can be obtained by reweighing any distribution of consumption expenditure, as long as there is a stable relationship between non-monetary variables. The success of the method depends not only on the validity of the correctness of the estimates from the benchmark surveys – in this case, the CPHS – but also on the stability of the relationship between non-monetary variables and monetary variables. Both the assumptions have been found to be doubtful using past exercises with limitations imposed by the choice of the non-monetary variables.

What happened to poverty after 2011-12?

Despite multiple estimates of poverty after 2011-12, there is no consensus on the level of poverty in recent years. However, almost all estimates suggest a reversal of the trend of a fast decline in poverty between 2004 and 2012. While some estimates suggest an increase in recent years, even those which suggest a poverty decline have confirmed the slowdown in that decline. While any conclusion on official poverty estimates may have to wait for the new estimates of consumption expenditure currently underway in the field, there are certainly other indicators which confirm the reversal of the trend of poverty reduction.

In fact, the strongest indicator of poverty, at least in rural areas, remains casual daily wages. The trend suggests a sharp deceleration in growth rate of wages, with a decline in real wages of non-agricultural labourers since 2013. The latest data from the Labour Bureau, which regularly publishes data on casual wages, is from June 2022. Based on that data, real wages in non-agricultural occupations have declined at 4.5% per annum in the last two years, while agricultural wages have declined at 2.7% per annum. On a long-term basis, the decline in real wages since 2013 for non-agricultural occupations has been 0.5% per year, whereas it has increased only marginally at 0.4% per year for agricultural occupations. Given the trend in real wages, it is unlikely to be the case that poverty decline would have accelerated.

Outstanding statistical issues for robust poverty estimates

In a country like India, where a significant population is still poor, poverty estimates matter. The debate on poverty is as much an intellectual exercise on statistical properties or the aggregates and distribution across regions and households, as it is the question of what the appropriate yardstick to measure poverty should be– which is the poverty line. On both counts, there is historical antecedent with engagement from the official government machinery and researchers. The evaluation of any such exercise is not independent of the reality emerging from various other indicators. Even otherwise, the debate on poverty estimates has been a lively and rich debate on measurement of key economic indicators, sampling and estimation strategy. This debate has also thrown up some of these issues which need to be dealt with, irrespective of whether official consumption surveys are available or not.

The first is obviously the privileging of some estimates and indicators over others. The implicit assumption in BBV’s estimation is that the NA estimates of PFCE are superior to the NSO consumption surveys. This issue has certainly attracted attention, not just for the purpose of comparison of the two estimates. but the validity of the NA estimates. Questions have been raised on how robust the NA estimates are, including by former Chief Economic Advisor Arvind Subramanian, and several others (see Subramanian 2019). Changes in the methodology of estimation of NA is not rare, and the new series of national accounts, which is based on new methodology, has attracted its fair share of criticism. However, its use in poverty estimation is problematic irrespective of the quality of NA data, given that it is a national aggregate and PFCE is a derived estimate. Poverty estimates are after all a distributional measure, something which is entirely absent in the National Accounts. Even with the same growth of consumption expenditure aggregates and PFCE aggregates, changes in distribution across households and across regions can produce completely different outcomes.

RW’s estimates also raise statistical questions which emerge from their approach of correcting the bias in the CPHS estimates. Consumption expenditure estimates are available from different nationally representative surveys. Other than the full consumption expenditure survey rounds, NSO also collects consumption aggregates with varying degrees of detail in other surveys, including the PLFS. These are available for a long period and were a standard schedule in the Employment-Unemployment Surveys (EUS) of the NSO. But private surveys also collected consumption data. The India Human Development Survey (IHDS) has consumption expenditure estimates for both the surveys in 2005 and 2012; so is the case with the CPHS. Even though the IHDS surveys were coterminous with the NSO consumption surveys, the aggregates as well as the distribution were widely different. The differences arise not only because of the sampling strategy but also the choice of recall period, level of detail in the questions, and so on. While these aggregates are different, even though they correspond to same year, it is unfair to privilege one over the other. Moreover, any method of adjusting the estimates based on reweighing or other methods essentially involves the use of strong assumptions and an inherent preference of individual researchers on what constitutes the true estimate. Nonetheless, RW’s method does raise important questions on why estimates differ across nationally representative surveys for the same period. This will require more research into the statistical properties of the distributions, but also reasons for differences arising from non-sampling errors and biases.

Conclusion

While the ‘Great Indian Poverty Debate 2.0’ may not answer the vexed question of what happened to poverty after 2011-12, it has certainly raised several issues which should catch the attention of the academic community engaged in research on consumption, income and its distributional impact. Such research is a necessity given that the new consumption survey by the NSO that is currently underway is using a new framework of collecting information on consumption expenditure – using multiple rounds, as against the past practice of a single visit and single questionnaire. With no such previous pilots or studies available in the public domain, the availability of the NSO consumption expenditure data of 2022-23 may not provide the much needed resolution of the poverty debate. Most likely, it will be the beginning of the Great Indian Poverty Debate 3.0.

Notes: 

  1. A selective reference to the various studies and the issues in comparing NSO estimates of consumption expenditure and national accounts is available in Himanshu (2022).
  2. The method used relies on a stable relationship between monetary aggregates such as consumption, income and wages, which are collected as part of various surveys, and non-monetary aggregates such as demographics, education, employment, dwelling characteristics, and asset ownership.  

Further Reading

Tags:
No comments yet
Join the conversation
Captcha Captcha Reload

Comments will be held for moderation. Your contact information will not be made public.

Related content

Sign up to our newsletter