The government of India does not seem to be interested in knowing how many poor there are in the country.

The missing poverty estimates

Until a few years ago, estimation of poverty in the country was always accorded a high priority, which has changed since 2014 when one official report on the measurement of poverty was ignored and then in 2019 when data collected by an autonomous government agency in 2017-18 was junked.

The absence of official data means that questions about who gains from economic growth and how the poor survive during times of slowdown and pandemic are now left to private researchers to answer. Two new sets of estimates on poverty and the trend since 2011-12 prepared by individual researchers have renewed focus on the absence of official data, as also on the quality of these two particular estimates.

The official estimates of poverty have hitherto been based on the consumption expenditure surveys (CES) of the National Statistical Office (NSO). The CES, part of the National Sample Surveys (NSS), are conducted by the National Sample Survey Organisation; the NSSO has now been made part of the NSO. The CES have been available since the 1950s and are very highly regarded for the quality of the survey as well as the independence of the survey agency.

The last such exercise was carried out in 2017-18 but the results of that CES survey were withheld by the government for reasons that are not entirely clear. While the official press release by the government mentioned problems with the quality of data collected, no details were provided about the nature of the problems. No study reports or any exercise that may have examined these problems have been made public.. This is against the standard practice of the NSO that used to release unit level data even for surveys that used experimental surveys and generated academic debate.

For example, the 1999-00 CES was mired in controversy due to its use of multiple reference periods for collecting consumption expenditure in the same survey schedule. Even though the survey results were finally not used by the government for official poverty estimation, the raw data as well as reports were made available to researchers. In fact, it was the availability of the unit level data which led to some of the finest research on poverty estimation and conceptual issues of measuring consumption expenditure using different reference period for eliciting information on consumption. A collection of papers on this issue and several issues related to poverty estimation was published as The Great Indian Poverty Debate by the World Bank, a volume edited by Angus Deaton and Valerie Kozel. The government of the day, led by Prime Minister Atal Bihari Vajpayee, accepted the wisdom of the research community and decided against using the survey for estimation of poverty given the controversial nature of the survey.

But at no point of time was there an attempt to suppress the data or to vilify the survey instruments or the agency. Contrary to the 1999-00 experiment, the 2017-18 survey was junked without any valid explanation. The 2017-18 survey was identical to the 2011-12 survey in methodology and concepts. One of the reasons for scrapping the survey results and the data could be that the preliminary results available from the report, which was subsequently leaked, suggested an increase in poverty in the country.

The absence of official data means that questions about who gains from economic growth and how the poor survive during times of slowdown and pandemic are now left to private researchers to answer.

Since the 1960s when the Planning Commission defined the official poverty lines, there have been numerous committees headed by eminent scholars who have periodically revised the poverty lines based on the CES data. The last such committee, under the chairpersonship of C Rangarajan, submitted its report in June 2014 to the present government. Unfortunately, that report is yet to be acted upon, with the current official poverty line continuing to be based on the 2009 Tendulkar committee. While NITI Aayog, the successor to the Planning Commission, has set up a committee for estimation of poverty, the idea of using consumption/income poverty has more or less been given up. The only available estimate of poverty is of multi-dimensional poverty using the National Family Health Survey (NFHS) data. However, this is based on a completely different methodology and is not comparable to the consumption-based poverty estimates. Unfortunately, even this poverty estimate remains unused with almost all government policies and programmes that use poverty estimates still relying on the 2011-12 estimates based on the 2009 Tendulkar committee.

Using the NSO consumption aggregates

Though the report of the 2017-18 CES was officially withheld by the government, it was leaked with data on the distribution of consumption expenditure for that year. Several estimates of poverty based on the leaked report have since been made.

One published in The India Forum is by Subramanian (2019) who used the data on the distribution of consumption expenditure across classes to estimate poverty. Using the 2014 Rangarajan committee’s poverty lines, Subramanian reports a rise in the poverty head count ratio to 35% from 31%. This is not unexpected given the fact that real monthly per-capita consumption expenditure (MPCE), according to the CES, declined to Rs 1,304 in 2017-18 from Rs 1,430 in 2011-12. The Subramanian poverty estimates remain the only ones consistent with the official methodology of the last four decades. In terms of population, the Subramanian estimates imply an increase in the number of poor by 52 million persons in the six years after 2011-12. India has not seen such a sharp rise in poverty and the number of poor in the last four decades even through a period of crisis.

[T]wo new studies have created a buzz as both have shown a decline in poverty since 2011-12 as against the estimates of Subramanian, and Mehrotra-Parida that showed an increase in poverty.

Other researchers have estimated poverty for more recent years by using different data from NSO surveys that have some measure of consumption. For example, Mehrotra and Parida (2021) have used the consumption estimate available in the Periodic Labour Force Surveys (PLFS) to estimate poverty for 2019-20, the last year for which PLFS estimates are available. Although the consumption estimates from the PLFS that they use are not comparable to those available from the CES, a single question on consumption has been included in various NSO surveys since 2015. Even though the instruction manual suggests a methodology to arrive at the estimate of consumption expenditure in the PLFS schedule, these are not comparable to the detailed consumption expenditure available in the CES surveys or even the abridged consumption schedule in employment-unemployement surveys (EUS) of 2009-10 and 2011-12 or earlier. Apart from the problem of bunching (responses are rounded and bunched around multiples of 100), these are also problems in their use for poverty estimation.

However, a strict comparison of similar aggregates over time is useful as a tool to understand evolution of consumption expenditure over time. Mehrotra and Parida use the Tendulkar committee poverty lines to arrive at an estimate of poverty of 25.9% for the country as a whole, with the number of poor increasing by 78 million between 2011-12 and 2019-20, broadly similar to what was reported by Subramanian (2019). Both these estimates are similar in terms of their use of consumption aggregates from the NSO surveys and use of poverty lines suggested by official committees. However, the Mehrotra-Parida approach is not entirely the right approach given the problem of non-comparability of the two consumption aggregates in the CES and PLFS. There has been a longstanding debate on comparability of consumption aggregates that have been collected using different level of aggregation and the choice of the recall period during the survey.

New hybrid estimates

In April 2022, two new estimates of poverty were published. The first was by Surjit Bhalla, Karan Bhasin and Arvind Virmani (hereafter BBV), of whom Surjit Bhalla happens to be affiliated with the International Monetary Fund (IMF) as the India representative on the board. The second also published in 2022 is by Sutirtha Sinha Roy and Roy van der Weide (hereafter RW), who are both affiliated to the World Bank (WB). Though both the papers appear as working papers of the IMF and World Bank, the poverty estimates they make are not the official estimates made by these institutions. For example, the official poverty estimate put out by the IMF and World Bank on their websites for India is different from those contained in the papers under discussion here.

The two new studies have created a buzz as both have shown a decline in poverty since 2011-12 as against the estimates of Subramanian, and Mehrotra-Parida that showed an increase in poverty. Both have used external benchmarks to adjust the NSO consumption expenditure aggregates to arrive at estimates of poverty after 2011-12. While the direction of poverty reduction in both is similar, they differ in terms of the magnitude of the poverty reduction. Further, unlike Subramanian (2019) and Mehrotra-Parida (2021) who used poverty lines based on the Tendulkar committee/Rangarajan committee, the most recent estimates have used the World Bank poverty lines of $1.90 and $ 3.2 converted into Indian rupees.

Bhalla, Bhasin and Virmani estimates of poverty

The 2022 paper by Bhalla with his co-authors is in many ways a continuation of Bhalla’s earlier work on poverty (See Bhalla 2002, Bhalla 2003 and Bhalla 2020) that used the methodology of adjusting the NSO consumption aggregates with the estimates of Private Final Consumption Expenditure (PFCE) of the National Accounts System (NAS). Bhalla has long argued that the consumption aggregates from the CES are gross underestimates of actual consumption, and are therefore not the right metrics to be used for poverty measurement. BBV use this argument to justify the non-release of the 2017-18 consumption survey of NSO. They yet use the distribution of consumption expenditure from the 2011-12 CES, which is methodologically and otherwise similar to the 2017-18 survey, to arrive at a poverty estimate for all the years since 2004-05.

BBV accept the role of the much-maligned PDS in reducing poverty.

The resultant methodology with the use of the $1.9 poverty line based on the Modified Mixed Recall Period (MMRP, explained below) distribution of consumption expenditure leads to their conclusion of a sharp decline in poverty in India from 12.2% in 2011-12 to 1.4% in 2019-20, before rising to 2.5% in 2020-21 during the pandemic. An alternative adjustment, using the consumption aggregates from the State Domestic Product (SDP) series, leads to a similar conclusion: poverty declined from 12.2% in 2011-12 to 2.2% in 2019-20, before rising to 4.1% in 2020-21. Note that both the methods yield the same estimate of poverty for 2011-12, irrespective of the methodology of updating the consumption expenditure, but diverge for estimates after 2011-12.

Using the $3.2 poverty line, BBV obtain estimates of poverty at 73.8% using PFCE adjustments and 76.8% using SDP adjustments for 2004-05. Incidentally, the estimates of poverty for 2004-05 are similar to the estimates of the size of the vulnerable population reported by the Arjun Sengupta-chaired National Commission for Enterprises in Unorganised Sector (NCEUS). For the same poverty lines, the estimates for 2011-12 are 53.6%, following both adjustments to the MMRP distribution of consumption expenditure. However, for the pre-pandemic year 2019-20, the PFCE adjusted estimates are 18.5%, compared to the SDP-adjusted estimate of 23.3%. Both report a rise in poverty in 2020-21 due to the pandemic.

BBV argue that the $3.2 poverty line may be better suited for poverty measurement given that the $1.9 poverty line leads to the absurd conclusion of almost no poverty in India. The second major finding of BBV is that in-kind transfers of foodgrains through the Public Distribution System (PDS) have a positive impact on poverty. While similar conclusions on the impact of the PDS have been reached in the past by several authors, BBV accept the role of the much-maligned PDS in reducing poverty. In the context of earlier writings of the authors criticising the PDS in favour of cash transfers (see Bhalla 2013a, Bhalla 2013b, Bhalla 2017), the 2022 study is a delayed but important recognition of the in-kind transfers made through the PDS.

NAS-NSO divergence in consumption

Ever since poverty began to be officially estimated in India, the CES of the NSO have been used for the purpose. This has been the practice, with no modification made to the expenditure reported by the CES. However, there have been experiments by the NSO to change the recall period of survey respondents for collecting this information. Until 2009, poverty was estimated in India primarily based on the Uniform Recall Period (URP) of expenditure. However, this practice was changed by the Tendulkar committee to make estimates from CES data that are based on the Mixed Recall Period (MRP). In the MRP consumption expenditure, the recall period of the survey respondents is 30 days for all items except for low-frequency items such as clothing, footwear, institutional health, education and durables on which information is collected on a 365-day recall period (yearly). The Rangarajan committee, however, recommended shifting to the use of the Modified Mixed Recall Period (MMRP), which uses 365-day recall for the low frequency items, a 7-day recall for some items of food expenditure and a 30-day recall period for the rest. Note, however, that all committees since Independence have recommended the use without any adjustment of consumption expenditure estimates reported by the NSO surveys.

While Bhalla has not been able to find any fault with the survey methodology or the survey instruments so far, he has been arguing for long that the consumption aggregates obtained through the CES are gross underestimates.

That there is a difference between NAS estimate of PFCE and survey-based estimate of consumption expenditure is not new in India or in any other country where such survey estimates exist. These differences have changed over the years depending on the methodology used to arrive at the PFCE estimate as well as the choice of recall period used in NSO surveys.

This controversy has been addressed numerous times by official committees on estimation of poverty as well as several other committees and independent researchers. Minhas (1988) studied these differences and concluded that one should not use the PFCE estimates to adjust NSS CES consumption estimates. A similar conclusion was reached by the Study Group on Non-sampling Errors, constituted by the Ministry of Statistics and comprising officers from the National Accounts Division (NAD), Central Statistical Office (CSO) and Survey Design and Research Division (SDRD), NSSO. The NSSO report Cross Validation Study of Estimates of Private Final Consumption Expenditure available from Household Survey and National Accounts arrived at the same conclusion (NSSO, 2005). This was followed by another committee by the National Statistical Commission (NSC) in 2008, which again included officials of NSSO and NAD under the chairpersonship of Savita Sharma. Most recently, another CSO committee under the chairpersonship of A K Adhikari examined this issue in the context of examining the methodology of GDP estimation and which in its 2015 report also reiterated the conclusions of the earlier committees.

That there is a difference between NAS estimate of PFCE and survey-based estimate of consumption expenditure is not new in India or in any other country where such survey estimates exist.

To a certain extent, the differences between the two aggregates of consumption are to be expected. They arise because of various factors including the differences in overage, concepts and the methodology used to arrive at these two estimates. The PFCE estimate of NAS are for households as well as private Non-Profit Institutes Serving Households (NPISHs) whereas the CES of the NSSO collects information from only households. In the NAS, the PFCE estimates are made using a commodity flow approach by subtracting intermediate consumption, government final consumption, exports, changes in stocks and investment from net availability, which is domestic production plus imports. To arrive at the final estimate of the PFCE, the NAS uses various rates and ratios, some of which are old and outdated. For example, the rates and ratios used for estimating wastage and marketable surplus for several commodities are based on the Directorate of Marketing and Inspection (DMI) reports of 1968-69. Further, given that the PFCE estimates rely on the production estimates used in estimation of GDP, all the problems related to GDP estimation are likely to be relevant in estimation of PFCE. This is more so with the current series of GDP estimates, which rely on data from the organised sector to estimate production and other aggregates for the unorganised sector. Several scholars including a former Chief Economic Advisor (CEA), Arvind Subramanian (2019), have raised doubts on the validity of the new GDP estimates. A part of the difference is also due to the inclusion of notional items in the NAS that are excluded in the NSO CES surveys. Two of these items are the imputed rentals of owner-occupied dwellings and Financial Intermediation Services Indirectly Measured (FISIM) that are not included in the NSO household surveys.

Despite overwhelming confirmation by all the expert committees that the two estimates of consumption are not comparable, Bhalla has continued to use the PFCE from the NAS to arrive at poverty estimates.

Most of these issues are well known and there have been efforts to correct some of the deficiencies in the PFCE estimates. One way PFCE estimates are adjusted to the extent possible is actually through the use of NSO consumption surveys whenever GDP estimates are revised with a new base year. This implies that even the PFCE estimates for several items in the NAS are derived from the consumption aggregates based on actual survey estimates from the CES and not the other way round. For example, in the new series of GDP with the 2011-12 base, expenditure on agricultural commodities, electricity, health and many others are revised based on the consumption expenditure estimates from the CES of 2011-12 (See CSO 2015b).

Despite overwhelming confirmation by all the expert committees that the two estimates of consumption are not comparable, Bhalla has continued to use the PFCE from the NAS to arrive at poverty estimates. Apart from PFCE being a derived estimate with its own set of issues, such a procedure is also problematic on several other counts. PFCE estimates by design are available only as an aggregate for the country as a whole and are not available either at the state level or even at the broad classification of rural and urban. Any adjustment, even item wise, therefore assumes the same divergence between the PFCE and CSE estimates across classes of households, across states and across rural/urban categories.

Clearly, PFCE estimates are not sufficient to generate any distributional outcome. The only alternative then is to use the CES of NSSO, which remains the only source of data on distribution of expenditure across households/persons. More importantly, it also assumes the distribution of households is constant over the years with the only change being the adjustment due to PFCE. Not only is this assumption unrealistic, it is also logically fallacious.

Sinha Roy and van der Weide estimates

RW use the consumption data of the Consumer Pyramid Household Survey (CPHS) of the Centre for Monitoring Indian Economy (CMIE) to analyse changes in poverty outcomes after 2011-12.

Since the CPHS data is only available after 2014, RW attempt to construct consumption aggregates from the CPHS data set that are comparable to the CES and estimate poverty for the years after 2011-12. They do this by using the non-consumption indicators available in both surveys to impute consumption expenditure in the CPHS survey that is consistent with the CES. They use the standard methodology applied in such cases where the common set of non-consumption variables are used to predict consumption estimates assuming a stable relation between the non-consumption variables and the consumption aggregates. They also adjust for the survey period since the CPHS surveys are collected on a quarterly basis and do not have the same reference period as the NSS surveys. However, given the issues with regard to the CPHS survey and its representativeness, they also re-adjust the sampling and population weights to arrive at consumption aggregates. As part of a robustness check they also use another approach by constructing a harmonized set of items of a consumption basket that is common in the CPHS and CES. They exclude items that are not there in either survey. Similar to BBV, they also use the World Bank poverty line of $1.9 and $3.2 to generate poverty headcount ratios.

However, unlike BBV who report negligible poverty in 2019-20, RW using the $1.9 poverty line report a poverty headcount ratio of 10.2% in 2019-20, down from 22.5% in 2011-12. Their analysis also confirms that demonetisation of 2016 did contribute to rising urban poverty and that the slowdown since 2016 has also contributed to a rise in rural poverty. Similar to BBV, they also report a sharp slowdown in the rate of poverty reduction, with poverty declining between 2011-18 at 1.3 percentage points per annum (ppa) compared to 2.5 ppa between 2004-05 and 2011-12. They also report a faster reduction in rural poverty compared to urban poverty after 2011-12.

CPHS survey

The CPHS survey of the CMIE has been carried out for quite some time now. However, very few studies have used it for measuring poverty or distributional outcomes. Part of the reason is the issue of reliability of the CPHS estimates of consumption/income as also of other indicators. Several authors have raised doubts about these estimates (Pais and Rawal, 2021; Somanchi 2021, and Sanyal 2021). Unlike the NSS surveys, CPHS surveys are conducted in waves of four months each with a total of three waves in a year. Apart from consumption, the CPHS also covers various other indicators on household demographics, income and assets. Being panel data, it also has issues of non-response and attrition of households, although these are not significant. RW undertake a comprehensive review of some of these objections and they provide different approaches to correcting the bias in the CPHS survey, including issues of sampling.

Unlike the NSS surveys that use a stratified sampling strategy based on house listing, CPHS surveys use a sampling strategy that selects villages at random with higher probabilities of households being selected if they are on the main street. The absence of a stratified sampling makes it doubtful that the CPHS survey can capture households at both ends of the distribution. For measurement of poverty and consumption expenditure, despite the use of a detailed consumption schedule, there are significant differences in coverage of items, with the CPHS using a far smaller number of items to arrive at consumption estimates. These also extend to the choice of the recall period, especially so when the surveys are not evenly staggered but can extend from one month to four months.

RW recognise the biases in the estimates reported by CPHS compared to other nationally representative surveys such as the NFHS and the PLFS. They use these differences to reweigh the CPHS sample observations through a two-stage procedure to eliminate the biases and make them comparable to the other surveys.

The absence of a stratified sampling makes it doubtful that the CPHS survey can capture households at both ends of the distribution.

While the method is technically sound, the adjustments are based on very strong assumptions. Inevitably, there are serious issues with the approach used to correct the survey responses in the CPHS to match with benchmark reference surveys such as the NFHS or the PLFS. Both the NFHS as well as PLFS surveys are designed to survey different sets of respondents and for specific purposes. Each of them comes with its own set of issues, starting from the sampling design to the data collection agency and the survey instruments chosen.

Even between the two external surveys used for validity tests, there are significant differences for most indicators or in the trend across time. The approach used by RW assumes the responses in the two external surveys to be correct with the differences primarily arising on account of the sampling weights and sampling strategy. Such an assumption is not just unrealistic but also ignores the critical issue of the difference in survey response by different socio-economic groups of households. More importantly, the RW exercise is unable to correct for the biases that appear because of a natural variation in survey response by different category of households.

It is these which matter for any identification strategy for imputing consumption aggregates comparable to the CES estimates of the NSS. The strong assumption implicit in the RW approach is the stability of the consumption characteristics of the household not only within individual years but also across time. This assumption is problematic in case the survey response itself is biased, which could well be the case. Reweighing only the sample weights is unlikely to take care of the biases in survey response. As RW themselves show, the reweighting exercise helps to reduce the difference in case of some variables but fails to do so in case of several others.

Choice of poverty line

The differences in the estimates of poverty are largely due to the income/consumption metrics chosen, which is different for each author. But some of the difference is also due to the choice of the poverty line. While Subramanian as well as Mehrotra and Parida use the official poverty line and update it with consumer price indices, there are differences among them with Subramanian using the Rangarajan poverty lines which are higher than the corresponding Tendulkar poverty lines. On the other hand, both BBV and RW use the World Bank poverty lines of $1.9 and $3.2. The World Bank poverty lines are available for the country as a whole but not separately for rural/urban or for states. While these can be arrived at by using the Tendulkar Committee rural/urban and state-wise price differentials, these have not been applied by either BBV or RV.

The use of the World Bank poverty line may be suitable for comparison across countries but using the same poverty line across households that face different prices across states and across rural/urban areas does pose a problem. Further, like all purchasing power parity (PPP)-based poverty lines, their ability to capture domestic inflation across geographical categories has always been an issue.

The choice of poverty lines is not a matter of convenience but is rooted in their representing some basic norms.

The choice of poverty lines also matters since the poverty lines are defined for a specific measure of consumption expenditure. The 1978 Lakdawala poverty line was meant to be used with with the URP-based distribution of consumption expenditure, which was the accepted recall period in consumption surveys of NSO until 2004-05. The Tendulkar poverty line was defined so as to be used on a distribution of consumption expenditure based on information collected by the MRP. This was changed by the Rangarajan committee that advocated the use of poverty lines based on the MMRP distribution since the NSO had decided to shift to collection of data on consumption expenditure by MMRP from 2011-12 onwards. As against these, the WB poverty lines of $1.9 and $3.2 do not specify any particular recall period to be used in the consumption distribution. This results in a peculiar situation where the same poverty line gives different poverty estimates depending on the recall period chosen. The difference in poverty estimates is a statistical artefact given that what changes is the estimate and ranking of households due to the choice of recall period even though the poverty line is the same. RW use the WB poverty line on the URP estimates while BBV use it on two alterative estimates of URP and MMRP. There are significant differences even using the same adjusted estimates of consumption expenditure of BBV depending on whether the adjustment has been made on the URP or MMRP distribution. This also changes the degree of adjustment required by BBV given that URP and MMRP distributions give different estimates with respect to PFCE estimates.

The choice of poverty lines is not a matter of convenience but is rooted in their representing some basic norms. These vary by recall period and the money-metric value of the same commodity bundle may be different given the choice of the recall period. Both the Tendulkar committee and the Rangarajan committee specify their poverty lines for a distribution of consumption expenditure that is based on a specific recall period.

Why did poverty decline according to BBV and RV?

Given the fact that the country saw back-to-back droughts in 2014 and 2015, followed by the policy-induced shocks of demonetisation and a hasty roll-out of Goods and Services Tax (GST) and a sharp slowdown in economic activity after 2016-17, the natural question to ask is how consistent are the BBV and RW estimates with the known facts on the Indian economy after 2011-12.

Both BBV and RW provide corroborative evidence from other data sources to substantiate their poverty estimates and the trend in poverty reduction. While RW have provided evidence on wages and other indicators to support their claim, they do recognise the negative impact of demonetisation as well as the GST roll-out. The impact was not uniform across rural and urban areas, and they point out that this did not continue in subsequent years. On the other hand, BBV argue that the real impact on poverty reduction has been through higher growth. This appears counterintuitive given the sharp slowdown in the GDP growth rate and a slowdown in private consumption after 2016-17, according to the NAS data.

Both BBV and RW provide corroborative evidence from other data sources to substantiate their poverty estimates and the trend in poverty reduction.

BBV do emphasise that in-kind transfer-based subsidies have played an important role in arresting the rise in poverty during the pandemic. They also come to the conclusion that a strengthening of these in-kind transfer-based subsidies is superior to the universal cash transfer programmes. The in-kind transfers had been analysed by many during the discussion preceding the enactment of the National Food Security Act (NFSA) (Himanshu and Sen 2013; Khera and Dreze 2013). However, most studies have found the impact to be much larger and also a greater impact after 2004-05 following expansion and reforms of the PDS in various states. On the other hand, the impact of the food subsidy in BBV’s exercise has been negligible with poverty rates after transfers only lower by 1-1.5% over the years. This is likely due to the flawed methodology used by BBV which, by using the PFCE of NAS, already takes into account the implicit subsidy in the PFCE. Nonetheless, irrespective of the quantum of the impact on poverty due to the in-kind food transfers, this is a delayed but important admission by BBV of the stellar role played by social security expenditure through in-kind transfers.

How robust are the adjusted poverty estimates?

There are now four different estimates of poverty for the years after 2011-12. Of these, only the estimates by Subramanian are consistent with the official methodology, since he uses the official poverty line and the unadjusted consumption expenditure aggregates from the CES. Both Subramanian, and Mehrotra and Parida use unadjusted consumption expenditure from NSO surveys and report an increase in poverty after 2011-12. On the other hand, the estimates by BBV and RW use adjusted estimates to arrive at an estimate of poverty that shows a reduction after 2011-12.

Despite the differences in estimates, which arise due to differences in the income/consumption estimate used for distribution and the choice of the poverty line, all four estimates agree that there has been a reversal in the trend of poverty reduction after 2011-12, compared to the period between 2004-05 and 2011-12. While Bhalla reports 30 million people moving out of poverty every year between 2004-5 and 2011-12, this number is only 13 million every year after 2011-12, without accounting for in-kind transfers. In terms of percentage point per annum (ppa) reduction in poverty, it is 2.9 ppa for 2004-11, and declines to almost one-third,  to 1.1 ppa after 2011-12. This is not very different from the RW estimate, which reports the rate of poverty reduction between 2004-05 and 2011-12 at 2.5 ppa, declining to almost half at 1.5 ppa for 2011-18. These trends are broadly similar even after accounting for in-kind transfers in case of BBV.

[A]ll four estimates agree that there has been a reversal in the trend of poverty reduction after 2011-12, compared to the period between 2004-05 and 2011-12.

Despite the confirmation of the fact that poverty reduction after 2011-12, in particular after 2014, has suffered a setback, there is no way to figure out the actual trend in poverty. The absence of any official data or a poverty line updated to reflect the contemporary reality is likely to lead to more efforts by private researchers to find alternative ways of estimating poverty. The use of CPHS data with all its flaws in terms of comparability and representativeness is only as good as the assumptions behind it. But the use of PFCE estimates to adjust CES estimates is not only flawed and outright wrong. It is mischievous to use this approach given the plethora of evidence against doing so. The attempt to revive the debate once again, however, fits in with the larger narrative of maligning and undermining official statistics, a trend that has gained prominence during the current government at the centre. This has been the case with the employment statistics for 2017-18, the official report of which was suppressed and released only after the general elections of 2019. The PLFS report for 2017-18 was leaked by a journalist before the general elections of 2019. In response to the leak, several government functionaries from the NITI Aayog and other ministers went on to vilify the survey. Ironically, the same government released the report after winning the 2019 elections,  and that report is now being used in all official documents of NITI Aayog and the Ministry of Finance. The political interference in the timing of the release of reports, and supressing official surveys such as the 2017-18 CES is part of a larger narrative of damning any statistics that are critical of the government.

The attempt by BBV is part of the same narrative that seeks to vilify the official statistical system to justify absurd claims of India having eradicated extreme poverty. While the issue of the differences between the CES and PFCE has been a long-standing one with efforts being made to reduce the gap between the two, such an exercise cannot be at the cost of the statistical integrity of the CES. In fact, there have been attempts by the NSO to try different recall periods as well as examine other issues related to household surveys. Similarly, the problems with PFCE estimates are well known and there are efforts to improve the data and ratios used in the NAS.

The political interference in the timing of the release of reports and the suppression of official surveys ... is part of a larger narrative of damning any statistics that are critical of the government.

Despite the availability of alternative data sources, estimates of poverty require a credible database and poverty lines that are defined taking into account the contemporary reality. Fortunately, the next CES survey is likely to start in July 2022. Though the availability of results is likely to be delayed given that general elections will be held in 2024, when it is made available it is likely to put to rest the debate on estimates of poverty after 2011-12. What is required is a decision on the poverty lines to be used to identify the extent of poverty.

Poverty estimates continue to remain relevant as inputs for decisions on distribution of resources across states and for various development programmes. Such an exercise begs an urgency on the part of the government. But poverty estimates are also important for a meaningful analysis of what happened to income distribution, poverty and vulnerability after 2011-12, a period of an economic slowdown and policy shocks in the Indian economy.

Himanshu teaches economics at Jawaharlal Nehru University, New Delhi.