In the fifth post of a six part series on the estimation of poverty in India, Martin Ravallion provides a non-technical summary of Roy and van der Weide's working paper. He discusses some of their main findings, which show that not only has the poverty rate continued to decline, but rural and urban poverty measures have been converging. He notes that although fewer people live at or near the consumption floor, the benchmark has not changed much since 2011.
India has had a long and (mostly) distinguished record in monitoring poverty, based on the National Sample Surveys (NSS), done by the Government of India’s National Sample Survey Office (NSSO).1 Prior to the early 1990s, researchers had to rely on NSSO’s published tabulations from the surveys. This changed when the micro data became publicly available. There have been debates about how poverty should be measured using the NSS, and how the results should be interpreted. But these debates were at least well-informed, given public access to the data and an active research community within and outside India’s borders.
That changed with the 75th NSS round for 2017-18, which the government chose not to release to the public in any form (a press release from the Ministry of Statistics pointed to concerns about the quality of the data, though with few details). The prior round of the large (five-yearly) sample surveys that have been most relied upon in measuring poverty was in 2011-12. That means that it is now over ten years since we have been able to update the estimates of poverty in the country that has historically held more of the absolute poor than any other (judged by a global poverty line that aims to have purchasing power parity (PPP) across countries). In the World Bank’s much-used interactive PovcalNet2 website, this became a gaping hole in the global picture of poverty.3
There were some efforts to figure out what the missing survey data would have told us about how India’s poor were doing. Some observers argued that the poverty rate had risen after 2011, others claimed that it had fallen, though by different amounts depending on who you listened too (see Sandefur 2022, for further discussion). Some observers have tried to provide an answer largely avoiding the need for household survey data (as in, for example, Bhalla, Bhasin and Virmani 2022). However, it has long been agreed that household surveys are necessary for credibly measuring poverty. Here my focus is on the problem of getting an internally consistent survey-based series of poverty measures for India in the context of measuring global poverty.
Until early April 2022 it seems that we were still pretty much in the dark about how India’s poor have been doing over the last ten years. That changed with the release of a World Bank working paper by Sutirtha Sinha Roy and Roy van der Weide (2022) (henceforth referred to as RW). I will briefly review what this paper has done, and then discuss what their results tell us.
Roy and van der Weide’s methods and findings
There has been a major private sector effort since 2014 to provide new survey data that could fill the gap left by the government’s decision to not release the results of the 2017-18 NSS. I refer to the Consumer Pyramid Household Survey (CPHS). While there was much hope that this new survey would do the job, it soon became clear that there were serious questions about its reliability for the purpose of measuring poverty, and (most importantly) its comparability with the NSS (see, for example, Drèze and Somanchi 2021). There are differences in how consumption was being measured, and questions about the representativeness of the CPHS, notably in how the sampling was done. The CPHS on its own was not going to be good enough to fill the gap.
What RW did was figure out a reasonable way to re-shape the CPHS to clone the NSS. Their paper provides a lot of technical detail in how they did this and various robustness checks (the Working Paper is over 90 pages most of which is devoted to technical/methodological issues). Here I will only provide a non-technical summary.
In addressing the concerns that have been raised that the sampling methods used in the CPHS are likely to have under-counted poor households, RW undertake a systematic re-weighting of the CPHS, such that its socio-economic composition (as measured by a set of observed household attributes) matches other nationally representative surveys. They then apply the new sampling weights in their estimations from the CPHS. They show that the likely biases in the CPHS are reduced.
They consider two approaches to dealing with the differences in the consumption measures obtained by the CPHS and NSS. The first uses a host of predictors of consumption, found in both the CPHS and the NSS, to impute NSS consumption into the CPHS. The second method models the relationship between CPHS and NSS consumption, which they then use to re-calculate CPHS consumption to make it look more like NSS consumption (as indicated by the first three moments4 of the NSS consumption distribution, exploiting the fact that the third moment of NSS consumption turns out to be highly stable over time). The two approaches differ in the assumptions they make but (reassuringly) they give similar results. The authors prefer the second approach, and I have no reason to disagree, so I will also focus on that here. Of course, it would be better to have the data from the 2017-18 NSS round, but that is still not an option. However, one advantage of using the CPHS in this way is that we get more high-frequency estimates of the poverty measures for years in which there was no NSS.
The ‘headline’ finding from the RW paper is that India’s poverty rate has continued to decline since 2011. Using their re-shaped CPHS consumption distributions, they find that the poverty rate (at $1.90 a day) fell from 19% in 2015 to 10% in 2019, as compared to the estimate using NSS data, of 22% in 2011. Rural poverty incidence also followed a downward trend, but RW find a brief reversal in the trend in urban poverty, which rose by two percentage points in 2016 (at the time of the demonetisation). Among other findings, RW confirm the finding of Datt et al. (2020) that urban and rural poverty measures have been converging over time, such that there is probably very little difference today. Table 1 puts together the PovcalNet numbers (both nationally and split between urban and rural areas) with those from RW’s study.
Table 1. Poverty measures for India
Sources: World Bank’s PovcalNet, Roy and van der Weide (2022), and author’s correspondence with Roy and van der Weide.
Notes: i) The measures use the $1.90 a day poverty line, in 2011 prices. ii) Poverty gap is the average distance of an individual from the poverty line.
Thoughts on consumption floor and other measures
This is, of course, good news. To put the RW results in historical context, India’s progress against poverty was undeniably slow prior to the 1990s, but has picked up its pace since then. Eyeballing the RW results in the context of the series for India in PovcalNet, as well as past work in the literature, I would not interpret their findings as either a marked acceleration or deceleration in India’s progress in eradicating poverty since the early 1990s5, but it is indicative of continuing progress. Simple end-point comparisons show that the annual rate of decline in the poverty rate was slightly higher post-2011 than post-1993 (1.5 percentage points per year in 1993-2019, as compared to 1.6 in 2011-19), but I would not make too much of this difference.
RW only provide the headcount index, but they kindly provided their estimates of the poverty gap (PG) and squared poverty gap (SPG) measures (Foster et al. 1984), which are also provided in Table 1. Unlike the headcount index, which only changes when households cross the poverty line, these measures are responsive to changes in consumption below the poverty line as well. The results generally follow a similar pattern over time to the headcount index, also echoing past findings in the literature related to earlier periods.6
Another perspective on progress against poverty is provided by estimating the consumption floor, as defined by Ravallion (2016). Ravallion proposes that the consumption floor should be measured by taking a weighted mean of all consumptions below the poverty line, with highest weight on the lowest observed consumption, and with weights declining linearly as consumption rises, reaching zero at the poverty line. This can be thought of as a Rawlsian measure of poverty7, focussing on the level of living of the poorest. Of course, there is likely to be considerable measurement error if one were to use the (literally) lowest observed level of consumption in any sample survey as the measure of the 'floor', so some averaging is clearly called for.
Table 2 provides this measure of the consumption floor, again for both PovcalNet and the Roy and van der Weide series; Figure 1 plots the series. The floor has generally risen over time – though rather slowly – from $1.09 a day in 1983, to $1.35 in 2019. Progress in lifting the floor has clearly been slow since 2011, when it was $1.32 a day. There has been convergence between the urban and rural consumption floors up to 2011. The Roy and van der Weide series suggest a fall in the urban consumption floor since then, and a reversal in the prior pattern of a higher floor in urban than rural areas. The signs of a drop in the urban floor could well reflect changes in the type of migrants moving from rural to urban areas over the last ten years.
Table 2. Consumption floor in India
Source: Author’s calculations based on World Bank’s PovcalNet, Roy and van der Weide (2022), and his correspondence with Roy and van der Weide.
Figure 1. Consumption floor in India over the years
Source: Based on NSS (up to 2011) and Roy and van der Weide (2022) (from 2015 onwards).
These findings on the consumption floor provide a qualification to the evidence pointing to India’s progress in reducing the proportion of the population living in poverty in the last decade (and prior to that) – namely that this progress appears to have come with only limited gains for the poorest, who are probably living at a similar standard of living today to that found in the last NSS round in 2011-12. Fewer people live at or near the floor, but the floor has changed rather little.
Next steps
We still need to see what India’s poverty measures over the last ten years look like when using India’s national poverty lines, applied to the careful re-shaping of the CPHS consumption data by RW. The World Bank’s international $1.90 a day line in 2011 prices has been anchored to the poverty lines found in low-income countries (updating the $1.25 line for 2005 proposed by Ravallion et al. 2009, based on a survey of national poverty lines). Other lines can also be defended, as PovcalNet readily allows.
Relative poverty8 measures are also of interest, to better reflect changing perceptions of what ’poverty’ means in growing developing countries such as India. The ‘strongly relative’ measures that have been popular in Western Europe are hard to defend on theoretical grounds, especially in developing countries, but ‘weakly relative’ measures offer a better compromise (Ravallion and Chen 2011).
Of course, the best next step would be for the Government of India to return to its policy of providing public access to the NSS data. Until then, we at least have something credible to work with, to fill that gaping hole in our knowledge about global poverty.
A version of this article previously appeared on the Center for Global Development (CGDev) blog.
Notes:
- The surveys have had many applications, but here the focus is on measuring poverty.
- PovcalNet is a product of the World Bank’s Global Poverty Monitoring project. It has an interactive computational front end that allows users to replicate the calculations made by the World Bank's researchers in estimating the extent of absolute poverty in the world.
- You can see the hole here; notice how the South Asia rows for recent years are labelled “Survey coverage is too low” – that is entirely due to the missing NSS data for India.
- The shape of any distribution can be described by its various ‘moments’. The first three moments are: the mean, which indicates the central tendency of a distribution; the variance, which indicates the width or deviation; and the skewness, which indicates any asymmetric leaning to the left or right.
- Going back to the earliest year in PovcalNet, the corresponding poverty rate was 56% in 1983; it was 48% in 1993 – the first post-reform survey round used in PovcalNet. For a longer-term perspective on India’s progress in reducing poverty and how it differed between the ‘pre-reform’ and ‘post-reform’ periods, see Datt et al. 2020.
- Over the period as a whole, the correlation coefficients between the headcount index and PG and SPG are 0.99 and 0.98 respectively; they are highly correlated. Taking the first differences, the correlation coefficients are 0.85 and 0.69
- Under the Difference Principle in his Theory of Justice, John Rawls favoured maximising the improvement of the ‘least-advantaged’ group in society. Similarly, Ravallion weights all consumption below the poverty line, with the highest weight for the lowest observed consumption.
- Measures of relative poverty remain unchanged when all incomes grow or contract by the same proportion. ‘Strongly relative’ poverty lines are set at a constant proportion of the current mean or median income.
Further Reading
- Bhalla S, K Bhasin and A Virmani (2022), ‘Pandemic, Poverty, and Inequality: Evidence from India’, International Monetary Fund, IMF Working Paper No. 2022/069.
- Datt, Gaurav, Martin Ravallion and Rinku Murgai (2020), “Poverty and Growth in India Over Six Decades”, American Journal of Agricultural Economics, 102(1): 4-27. Available here.
- Drèze, Jean, and A. Somanchi, 2021, “View: New Barometer of India’s Economy Fails to Reflect Deprivations of Poor Households,” The Economic Times June 21.
- Foster, James, J. Greer, and Erik Thorbecke, 1984, “A Class of Decomposable Poverty Measures,” Econometrica 52: 761-765.
- Ravallion, Martin, 2016, “Are the World’s Poorest Being Left Behind?” Journal of Economic Growth 21(2): 139–164.
- Ravallion, Martin, and Shaohua Chen, 2011, “Weakly Relative Poverty,” Review of Economics and Statistics 93(4): 1251-1261.
- Ravallion, Martin, Shaohua Chen and Prem Sangraula, 2009. “Dollar a Day Revisited,” World Bank Economic Review 23(2):163-184.
- Roy, SS and R van der Weide (2022), ‘Poverty in India Has Declined over the Last Decade But Not As Much As Previously Thought’, World Bank, Policy Research Working Paper 9994.
- Sandefur, J (2022), ‘The Great Indian Poverty Debate, 2.0’ Ideas for India, 12 October.
Comments will be held for moderation. Your contact information will not be made public.