The Great Indian Poverty Debate, 2.0

TL;DR: Has the Modi government accelerated or decelerated poverty reduction? It’s hard to know, as India has effectively stopped measuring poverty. A new World Bank paper using private-sector survey data finds the share of people living below $1.90 per day has been falling, but is higher than we thought, at about 10 percent. A rival paper from India's representative at the IMF says India has joined China in (almost) fully eradicating extreme poverty. Both estimates seem optimistic given other economic indicators.

For the past decade, India hasn’t released any official poverty numbers

In the early 2000s, economists engaged in a heated debate about how much India's 1990 liberalization had reduced poverty. Christening it The Great Indian Poverty Debate, Nobel Laureate Angus Deaton and Valerie Kozel noted "the various claims have often been frankly political, but there are also many important statistical issues." Twenty years later, history repeats.

The underlying question this time is whether Narendra Modi, who assumed office in 2014, has been good for poverty reduction. From a technical perspective, this debate is interesting because the data are so bad. If the data were good, there'd be no need for fancy imputations. But because the integrity of India’s statistical system has come into question over recent years, facts are up for grabs.

Official poverty measurement essentially stopped in 2011/12. There are hints of bad news behind that pause. Leaked data from the 2017/18 National Sample Survey showed a startling 3.7 percent decline in real consumption over six years, and the survey was never released. Though in fairness, as discussed on this blog in the past, those preliminary numbers seemed implausibly dire.

In the absence of official survey data, two new papers released this month offer creative solutions to fill in the gaps, and reach starkly different conclusions about what might be happening to poverty.

A new paper by World Bank researchers estimates extreme poverty in India is higher than previously thought at 10.2 percent in 2019

Sutirtha Roy and Roy van der Weide use data from the Consumer Pyramid Household Survey (CPHS) run by the Centre for Monitoring the Indian Economy to come up with a new estimate of Indian poverty. That's no small feat. The official data stop in 2011 and the CPHS only started in 2014, its sample appears to undercount poor households, and even small differences in how consumption questions are asked can make a big difference. So Roy and van der Weide spend many pages explaining how they reweight the CPHS data to make it comparable to the NSS series⁠—impressively painstaking work that I can't really do justice to here.

The end result though is cautiously optimistic. The title really lives up to the frequent exhortation to researchers to "just say what you found": Poverty in India Has Declined over the Last Decade But Not As Much As Previously Thought. They make four basic claims:

  1. Extreme poverty fell 12.3 percentage points from 2011 to 2019 using the World Bank’s $1.90 per day line in purchasing-power parity terms.
  2. Poverty stands at 10.2 percent in 2019, considerably higher than earlier projections based on consumption growth observed in national accounts.
  3. Urban poverty rose by 2 percentage points during the 2016 demonetization, and rural poverty reduction stalled by 2019, when the economy slowed.
  4. Inequality has stopped rising since 2011.

The basic story can be seen in the figure below: still good news, but less so than if you just relied on national accounts.

Figure 1. Indian poverty (<$1.90 PPP) since the official data series stopped

Source: Roy & van der Weide 2022, Figure 17

There were good ex ante reasons to think the projection based on the national accounts system (NAS) shown by the dotted line in Figure 1 was too optimistic⁠—more on that below. But consumption growth in the CPHS survey also looks surprisingly high in the latter years, especially 2019-20. That year the index of industrial production for consumer goods registered an absolute decline of 3.8 percent (even more in per capita terms), real rural wages declined, and real urban wages grew less than 2 percent—meanwhile the CPHS showed growth in real per capita consumption of 6.2 percent.

But if the Roy and van der Weide estimate seems mildly optimistic, fasten your seatbelts...

An IMF working paper released the same week says India had eradicated extreme poverty by 2019, and held it near zero through the pandemic

At the same time the World Bank released the Roy and van der Weide estimates, the IMF published an alternative estimate coauthored by Surjit Bhalla, the Indian government's representative at the IMF and a well-published economist who is a veteran of the first great Indian poverty debate. In their paper, Bhalla and coauthors Karan Bhasin and Arvind Virmani eschew new survey data entirely. Instead, they start from the last official 2011/12 survey and shift the distribution in line with the growth of consumption in the national accounts, and then add in explicit allowance for government food subsidies. The conclude that:

  1. India has essentially eliminated extreme poverty. They estimate the proportion of the population below $1.90 in PPP terms at just 0.8 percent before the pandemic.
  2. The pandemic didn't increase poverty. Rather, "food transfers were instrumental in ensuring that it remained at that low [0.8 percent] level in pandemic year 2020.”
  3. Inequality has fallen to the lowest level since 1993/94, with a Gini after transfers of 0.29.

Note that the 0.8 percent poverty rate that Bhalla et al cite here is using a different definition of the underlying consumption aggregate than the one used by the World Bank above (based on the NSS's MMRP series instead of the URP series, for those who follow this stuff). In more apples-to-apples terms, their (URP) number is 1.9 percent—still very low.

Figure 2. An alternative view of Indian poverty, with and without adjusting for an increase in transfers from the Public Distribution System

Source: Bhalla, Bhasin, and Virmani 2022, Figure 1

The novel and interesting claim in the Bhalla et al paper is that previous estimates have overestimated poverty because they rely only on survey measures of actual expenditure on food items, implicitly omitting the government subsidy embodied in India's massive Public Distribution System (PDS). Using administrative data on PDS and various assumptions, Bhalla et al get a significantly lower poverty rate.

It's unclear (to me) how the paper has dealt with leakage from the PDS system, and errors of inclusion and exclusion which we know are rampant. But a recent survey from my CGD colleagues found that the ration system held up quite well during the pandemic. So the question Bhalla et al raise about how much Modi's new welfarist policies may have reduced national poverty is an interesting and important one, even if their impact is at most a couple of percentage points.

[A technical aside: while interesting, it’s not clear that it’s strictly valid to add in the subsidy value to Indian consumption for calculating globally comparable poverty rates using the World Bank’s $1.90 PPP line, unless one does the same for other countries, who also have various food and energy subsidies.]

The deeper challenge with the Bhalla et al "zero poverty" calculation is its reliance on official national accounts data, which various pieces of evidence suggest will exaggerate true poverty reduction.

The optimistic view on Indian poverty requires that GDP growth was exaggerated before 2011, and hasn't been sincetwo claims open to debate

There's been a lot of ink spilled about India's national accounts in the last several years, so it's worth distinguishing two different issues.

First, my CGD colleague Arvind Subramanian has made the case (and here) that the government began, inadvertently at first but systematically, overstating GDP growth after 2011 (pre-Modi). And earlier World Bank estimates for India have included that possibility in their poverty scenarios, though neither Roy and van der Weide nor Bhalla et al do so here. If Subramanian is right, then poverty may be significantly higher than both sets of national accounts projections imply. Indeed, Subramanian’s caution on the recent GDP figures would point in the same direction as Roy and van der Weide’s preferred survey-based poverty estimates, at least through 2017.

Second, in the absence of new survey data, Bhalla et al extrapolate survey consumption from 2011 by assuming nominal consumption growth in national accounts passes through to growth in the survey mean, one for one. They show this was roughly true in India’s most recent pre-2011 survey rounds. But a large literature, starting with Ravallion (2003), shows this doesn’t hold in general. And more recent World Bank research concludes the most plausible rate of pass through for India from real consumption in national accounts to surveys is 0.67. That’s why Roy and van der Weide get a much higher poverty rate than Bhalla et al, even when the former use the same national accounts data.

To compound matters, the historical national accounts data underlying Bhalla and coauthors’ estimates are subject to their own heated debate—totally separate from Subramanian’s critiques of the newer national accounts. In 2018, the National Statistical Commission recommended a backward revision to the GDP series that would have raised the growth rate from 2004-05 to 2011-12 a fraction of a percentage point. But a few months later, the government’s internal think tank, NITI Aayog, decided to go the opposite direction, and published a new backcast GDP series revised in the opposite direction, implying growth in GDP and consumption before 2011 was more than a full percentage point lower. (Critics contended that lowering historical growth helped make the current government look good by comparison.) One side-effect of the downward revision is that it made national accounts and survey data look more consistent, yielding Bhalla et al’s finding of 100 percent pass through. If instead one stuck with the original pre-2011 national accounts, or even more so the National Statistical Commission’s recommendation, you’d have to conclude 100 percent pass through is too high, and hence the implied poverty rate in 2019 is too low.

In sum, there is unlikely to be any resolution to this debate any time soon, but reliance on national accounts only kicks the ball into even more fraught and politicized territory. As it stands now, we have a wide range of estimates for poverty changes since the last (accepted) estimate of 22.5 percent for 2011/12: from an increase (unreleased NSS, 2017/18) to a near-elimination by 2019 (Bhalla and coauthors) with the Roy and van der Weide 2019 estimate falling in-between at about 10 percent.

Until the integrity of the official household surveys is restored, any judgment on the Modi government’s efforts to reduce poverty appears tentative at best, and the great Indian poverty debate will go on.

Thanks to Surjit Bhalla, Alan Gelb, Christoph Lakner, Maria Ana Lugo, Anit Mukherjee, Sutirtha Roy, Arvind Subramanian, and Roy van der Weide for helpful comments and clarifications. Any conclusions (or errors) here are mine alone.


CGD blog posts reflect the views of the authors, drawing on prior research and experience in their areas of expertise. CGD is a nonpartisan, independent organization and does not take institutional positions.