My colleague Maxim Pinkovskiy (a former Columbia undergraduate student of mine and currently a PhD student at MIT) and I wrote a paper called “Poverty in Africa is Falling Faster than you Thought”. IN this paper, which is now an NBER working paper published in February 2010 (http://www.columbia.edu/~xs23/papers/pdfs/Africa_Poverty_NBER.pdf) we show that poverty in Africa has been falling rapidly thanks to the positive growth rates experiences after 1995. In fact, we predict that, contrary to the official predictions, Africa may actually meet the Millenium Development Goals on time in 2015 or shortly after.
Right after our paper saw the light, Martin Ravallion, the head economist of the Poverty Division of the World Bank, wrote a criticism of our paper. http://blogs.worldbank.org/africacan/is-african-poverty-falling
It is not the first time that World Bank researchers react negatively to my research. You may remember that Martin Ravallion and the World Bank initially criticized my 2006 paper, where I showed that worldwide poverty was falling faster than previously acknowledged and that, contrary to World Bank claims, global income inequality has fallen since 1970. A few years after the initial fight, Martin Ravallion acknowledged in an interview at The Economist that my results may not have been wrong.
But here we are in 2010 and here we have another (healthy) intellectual fight. People at the World Bank call this “The Clash of the Martins” (MARTIN Ravallion vs Sala-i-MARTIN). As you can imagine, I disagree with Ravallion’s criticism and in this note I explain why.
Reply to Martin Ravallion, “Is African Poverty Falling”
Martin Ravallion makes three criticisms of our results: 1) we look at poverty rates rather than poverty counts, 2) we attribute the difference between GDP and household consumption to consumption, and 3) we are not as cautious about our results as the paucity of our data should suggest. We believe that, on the contrary, looking at income poverty rates (rather than consumption poverty counts) reflects much better both the spirit of what poverty is supposed to mean, and the intention of the Millennium Development Goals, which motivate our question. Moreover, as we document very explicitly through our robustness checks in the paper, the available evidence is conclusive that Africa has been reducing poverty, and doing so sufficiently quickly to halve poverty within a reasonable period of time.
On points 1) and 2), it should be sufficient to recall the exact text of the first Millennium Development Goal, which is to “halve, between 1990 and 2015, the proportion of the population with an income of less than $1 a day” (cite). Hence, insofar as the MDGs are concerned, the relevant measure of the extent of poverty is the poverty rate, and the relevant definition of who is poor is in terms of income, not consumption. We consider that these definitions are useful not just because they are used to formulate the MDG, but also because they make intuitive sense. If country A had 5 million people, 4 million of whom were poor, while country B had 100 million people, of whom 5 million were poor, it would be unreasonable to say that poverty is lower in country A than in country B, and the citizens of country A would feel much more affected by poverty in their midst than the citizens of country B. Moreover, if a person had an income of twice the poverty line, but chose to consume only one-fourth of it (hence, half the poverty line consumption) and invest the rest since she expected the rate of return to be high, it is difficult to consider this person poor using this poverty line, since she had the resources necessary to assure herself a consumption at the poverty line but chose not to use them in this manner. Such an instance might cause us to reevaluate the poverty line downwards, but not to consider this person as poor.
Ravallion’s third argument is that we are insufficiently cautious in reporting our results given the paucity of the data. Most immediately, this contention does not take into account the frankness with which we acknowledge exactly how many surveys we have and what kinds of assumptions were used to recover the poverty and inequality statistics from the data, all of which we do in the first few pages of the paper; nor does it consider that we have performed extensive robustness checks on these assumptions and present the full results. Moreover, the argument ignores the substantial evidence that we present to show that this trend is not spurious, which we proceed to review.
An estimate of poverty can be thought of as consisting of two parts: an estimate of income (or GDP) per capita, and an estimate of inequality. While we do not have survey data to measure the distribution of GDP in each country in each year (which we make extremely clear in our methodological discussion), we do have GDP for nearly the entire sample period. Hence, the only way that our results could be wrong is if our inequality series is somehow masking a strong increase in inequality over the period 1995-2006; the growth in African GDP over that period is extremely well supported with the data we use. However, this is implausible for two reasons, which we present below:
First, our result that poverty is rapidly declining is extremely general. We recomputed our results for special groups of African countries that were supposed to have different growth and poverty reduction performances: for all these groups, poverty fell, and in particular, poverty fell more for those groups that were initially poorer. Moreover, removing any one of the largest 8 African countries from the analysis does not change the results. It is unlikely that we would see such a consistent pattern if our result were driven by the absence of some key surveys in large African countries.
Second, the survey data on inequality are much more informative than the simple number of surveys suggests. Inequality tends to change very continuously and relatively slowly over time, so the survey data that we do have directly gives us information not only about inequality for the countries and years in which the surveys were conducted, but also about inequality for these countries in nearby years. Hence, the inequality in countries during “inter-survey” years is measured very effectively by interpolation. We have computed this fraction for each year in two ways: using all surveys, and just using consumption surveys (so as to account for the potential criticism that our results are due to switching from income to consumption surveys). It turns out that the 1990s, the crucial period in our paper during which poverty begins its sustained decline, is the period that is best covered by the surveys – using all surveys, over 70% of Africans are covered by surveys directly or by interpolation between 1991 and 1999, whereas using only consumption surveys (which is the measure we show in our paper), about 60% of Africans are covered between 1993 and 1999. During this period, not only does poverty begin to fall, but inequality declines markedly and consistently. For our results to be wrong, inequality should have bucked its declining trend and risen substantially in the 2000s, a hypothesis for which the data give no support at all. Even if the declining trend in inequality had ended and inequality were flat during the 2000s, we would observe falling poverty because GDP continued to grow.
Unless Doctor Ravallion wishes to say that we know absolutely nothing about African inequality, we believe that the best interpretation of the evidence is that African inequality declined over the course of the 1990s; hence, that the documented and acknowledged African growth led to a reduction of poverty then. For poverty to have stopped declining subsequently, this decline in inequality would have had to be more than completely offset by a dramatic rise in the last four years of our sample, and there is no basis for believing this to have been the case. Hence, whether one to break the trend in African inequality (as we do in our paper) or to extrapolate it, the result is the same: poverty falls.