In an article on a Brookings website, Laurence Chandy and Homi Kharas chide the World Bank for three so-called “contradictions” in its global poverty numbers, including the Bank’s latest update. Let me look more closely at these “contradictions” in turn.
First, Chandy and Kharas chide the Bank’s team for assuming that North Korea has the same poverty rate as China. I wish Chandy and Kharas good luck in trying to measure poverty in a place like North Korea, with almost no credible data of any sort to work with. I could offer a guess that 80% of North Korea’s population is poor today—roughly the same as China before it embarked on its reform effort in 1978. This would add slightly less than 1 percentage point to our estimate of the “$1.25 a day” poverty rate for East Asia in 2008.
North Korea is only an example, of course. The more general point made by Chandy and Kharas refers to the fact that we assume that the countries with the survey data needed to measure poverty are representative of all developing countries, including those without such data. The key issue is whether there is a bias in our estimates due to the possibility that the countries without survey data are poorer. In fact the first estimates we did, for the 1990 World Development Report, included corrections for this potential bias. (We used regressions where the predictors include variables that are measured for the countries without household surveys. This method is described in the relevant background paper to the 1990 WDR, which was published in a scholarly journal, and is cited in the latest paper, downloadable from the main website.)
A correction for this source of bias was essential in 1990, because we only had survey data for 22 countries. Now our coverage is much more complete—130 countries representing 90% of the population of the developing world. Granted coverage is not uniform, and in the latest update we signal the years and places when there is a problem of poor coverage.
We have also returned at times to the old method of correcting for bias, to see if it makes any difference. The effect appears to be small. I grant that it is not easy to be sure given that the countries without survey data these days don’t have much of any sort of reliable data.
However, Chandy and Kharas are mistaken when they claim that “those countries that remain without a survey…are unsurprisingly amongst those where one would suspect poverty levels are especially high.” They seem to have come to this conclusion by eyeballing the missing data. But if they had done a more careful check—and it is quite easy to do—they would have found out that the probability of having a survey is actually higher for poorer countries, as measured by GDP per capita. But (North Korea aside) the main reason why countries don’t have surveys these days seems to be that they have small populations, not that they are unusually poor. The bias in our poverty measures is likely to be small. So we abandoned our earlier corrections.
Their second claimed “contradiction” concerns our use of surveys for measuring poverty. Here they point out that the survey-based mean consumption in India has been growing at a much lower rate than the private consumption component of the national accounts. On this basis they claim we are under-estimating India’s progress against poverty.
We have addressed this concern quite fully in the background paper, cited in the footnote to the latest update. We cannot rule out the possibility that India’s national sample survey is underestimating mean consumption, though I would not rush to assume that the national accounts are getting it right, given how consumption is measured there.
But that is not the point. We are not estimating mean consumption, we are estimating the incidence of poverty. Chandy and Kharas are wrong to assume that underestimation of the mean in India will imply a corresponding over-estimation of the poverty measure. Everything we know about this problem suggests that the surveys in India as elsewhere are probably also underestimating inequality. It is not the poor who are refusing to participate in surveys or under-reporting their incomes. So “correcting” solely for the error in the mean could actually give you a worse estimate of poverty incidence. There is no basis for this second claim of bias by Chandy and Kharas.
Their third “contradiction” is about the Purchasing Power Parity rates we use. Here they are also greatly overstating their case, when they say that “for some countries, most notably China, the PPP conversions have little credibility.” They correctly point out that there was a sampling bias in the 2005 price surveys for China. But they ignore the fact that we implement corrections for this bias. These are noted in the background paper, which refers to another published paper that outlines the problem (we were in fact the first to point it out) and explains our corrections, using supplementary price data for rural areas.
Chandy and Kharas also question our claim that projecting backwards implies that China in 1980 would have been as poor as the poorest few countries today. But they do not present or cite any evidence to support their belief that this is implausible. There is a large literature on poverty in China, how poor the country was around 1980, and how much poverty has fallen since the economic reforms that started in 1978. Somehow all that literature is just swept aside by Chandy and Kharas.
The common theme in all this is that Chandy and Kharas should have done a more thorough job of reading the background papers, and the literature more generally, before they jumped to criticize the Bank’s poverty estimates. Those who would like to be better informed than Chandy and Kharas about the methodology used by the Bank’s researchers should read the paper written by myself and Shaohua Chen published in the Quarterly Journal of Economics 2010. We refer to this paper in the summary of the latest update, available from the website.