Children often declare a game over when they happen to be on a winning streak, just to be able to claim victory. That seems to be what Surjit Bhalla is trying to do in a recent article, co-authored with Karan Bhasin (IE, November 4, ‘Poverty is down, period‘). In that article, the authors (B&B) make startling claims of runaway poverty decline under the Narendra Modi government, and proceed to state without false modesty that this “unambiguous and strong conclusion” allows them to “close this debate”. Problem is, their figures are very misleading.
B&B’s central claim is that poverty decline accelerated under Modi across the full range of 10 deprivations used in the calculation of “multidimensional poverty” (MP). The claim is based on a set of figures presented by B&B as deprivation-specific “headcount ratios” (HCRs) — the percentage of deprived persons in the population. The deprivations of interest are those commonly used in MP calculations, and relate to three distinct domains — health, education and household amenities (also called “living standards”). HCR estimates for 2005-06, 2015-16 and 2019-21 are presented, based on successive National Family Health Surveys. The authors also insert, in this series, 2011-12 estimates based on the second India Human Development Survey (IHDS-2).
B&B then compare the annual rate of decline of deprivation-specific HCRs before and after 2011-12. The results are startling. Across the board, the rate of decline of HCRs was higher after 2011-12 — even “eight times higher” on average. Wow!
But here is the rub. The 2011-12 figures are plain wrong, as B&B graciously admitted by email. They are actually 2005-06 figures from an outdated MP series, copy-pasted by mistake. So, the alleged comparison between 2005-06 and “2011-12” is actually a comparison between two different sets of estimates for 2005-06 (indeed, the respective figures are very close to each other), and the comparison between “2011-12” and 2019-21 is actually a comparison between 2005-06 and 2019-21. No wonder B&B found that HCR decline was so slow before “2011-12”, and so rapid after that!
This goof-up cannot be repaired, because IHDS data are not comparable with NFHS in any case. We can try to salvage something from this wreck by discarding the 2011-12 figures and focusing on the comparable NFHS figures for 2005-06, 2015-16 and 2019-21. Alas, those figures are wrong too, as anyone familiar with NFHS data would immediately notice. The figures projected by B&B as indicator-specific HCRs (“the percent poor/deprived according to that indicator”, as they put it) are not, in fact, headcount ratios. They are “censored headcount ratios”, very misleading for the authors’ purpose as I shall explain shortly.
Meanwhile, the correct HCRs are presented in the accompanying table, along with the corresponding rates of decline before and after 2015-16. Contrary to B&B’s analysis, the rate of HCR decline was not uniformly faster in the second period — far from it. It was faster for most household amenities, but slower for most other indicators.
How we judge overall progress would depend on how much “weight” we attach to different indicators. In MP work, the convention is to give equal weight (one third each) to health, education and amenities, and then equal weight to individual indicators within each domain. Based on the conventional MP weights, we find that the overall rate of decline of deprivation was the same in both periods.
Another way of judging overall progress is to look at the “multidimensional poverty” HCR. Briefly and approximately, this is the proportion of persons who are “multidimensionally poor” in the sense of experiencing more than a threshold number of deprivations. The multidimensional-poverty HCR, it turns out, declined faster in the second period (last row in the table). This is not necessarily surprising, since a given percentage decline is easier to achieve from a lower base.
The fast decline of multidimensional poverty in the second period is largely driven by the rapid improvement of amenities. The latter is also driving the rapid decline of B&B’s “censored HCRs”. For a given deprivation, the censored HCR is the proportion, in the total population, of people who are deprived and multidimensionally poor. Even if there is no improvement in, say, nutrition, the censored HCR for nutrition deprivation would decline over time simply because multidimensional poverty is declining. That is why B&B’s use of censored HCRs to evaluate deprivation-specific poverty decline is patently inappropriate.
It is important to note that none of these MP-related indicators capture short-term purchasing power. As B&B rightly point out, amenity improvements in the second period are more a reflection of public policy and subsidies than rising incomes. The impact of public policy in this domain deserves appreciation. But trends in household purchasing power would also be important to look at. On this, the first period was almost certainly more favourable: The rate of growth of per-capita consumption, according to national accounts figures presented by B&B themselves, crashed in the second period (after the demonetisation self-goal, much applauded by Surjit Bhalla). And remember, all this pertains to the pre-Covid years — most sample households in the 2019-21 NFHS survey were interviewed before the Covid crisis began.
That, roughly, is the picture as things stand, with all the qualifications that apply to these indicators and methods. None of them are perfect, and the debate must certainly continue. Hopefully, the next NSSO consumer expenditure survey and third India Human Development Survey will soon shed further light on these matters.
This is my third analysis — in three years — of Surjit Bhalla’s legendary spinning skills. I promise to leave it at that for now — time to quit!
The writer is Visiting Professor at the Department of Economics, Ranchi University