Syndicate content

New Brookings Study is Overly Optimistic on Progress Against Poverty

Martin Ravallion's picture

A new paper by Laurence Chandy and Geoffrey Gertz at the Brookings Institution reports a remarkable acceleration in the pace of progress against absolute poverty since 2005, as can be seen in Figure 1 of their paper (found here). This would be great news if it could be believed, but there are reasons for doubt. 

In “updating” the World Bank’s estimates using the Bank’s PovcalNet site, Chandy and Gertz have relied heavily on forecasts rather than estimates based on new surveys. Household surveys are the only credible method of measuring poverty. The technology has improved, but naturally the data take time to collect and process. We still do not have sufficiently recent surveys for many countries, especially in Sub-Saharan Africa, the region with the highest overall poverty rate. The next edition of the World Bank’s regular three-yearly updates of its survey-based estimates of global poverty measures is scheduled for release later this year, and will go up to 2008, and revise consistently back to 1980.  (Information on the last update can be found here. The paper documenting the methods and testing their robustness can be found here.)   

So forecasts have their role, and the Bank’s team also produces forecasts of its poverty measures, to both update in lieu of new surveys and to project forward, in assessing likely future progress. But we have to make sure that the forecasts use all the available information and avoid predictable biases.

There are two reasons to question the forecasting methods used by Chandy and Gertz. First, they assume that there is no change in inequality during periods of rapid economic growth. This ignores the fact that the growth processes in the two most populous countries, China and India, as well as a number of other countries, have come with rising inequality (markedly so in China), which has of course dulled the impact of growth on poverty. (For further discussion and references to the literature see my paper here.) Second, Chandy and Gertz assume that all of the economic growth indicated by the national accounts is reflected in the mean consumption of households. This has not been the case in the recent past (as I demonstrated here).

The World Bank’s forecasting methods try to take account of both factors, and we get less optimistic results (as reported in Global Monitoring Report 2010 for example), though we still expect considerable progress against poverty, albeit uneven progress across countries and regions.

************
Also see, Poverty's success story, The Washington Post.

Comments

Submitted by Laurence Chandy... on
We welcome Martin Ravallion’s thoughtful comments on our recent paper, 'Poverty in Numbers: The Changing State of Global Poverty from 2005 to 2015'. (We also acknowledge the Bank’s new open data policy, without which our research would not have been possible.) Ravallion correctly identifies two limitations of our approach, both of which are discussed in detail in our paper’s methodology. With respect to inequality, it is true that if new household surveys reveal shifts in the income distribution such that those living just below the poverty line see their incomes increase less (more) than the average household, our distribution-neutral analysis will understate (overstate) the true extent of poverty. Yet it is worth noting that China’s impressive record in decreasing poverty between 1990 and 2005 coincided with rapidly rising inequality, suggesting that dramatic progress against poverty is possible even when a country’s income distribution is widening. We also recognize the discrepancy between national accounts and survey-based measures of consumption growth. Given the lack of alternatives, however, we believe growth as measured by the national accounts remains the best proxy for growth in the survey mean; we expected Ravallion would be at least somewhat sympathetic to this compromise, as it is the same approach which the Bank adopts in order to line up survey years with a given reference year. In any case, while admitting that these two caveats moderate the accuracy of our findings, we believe the overall shape of our results is robust, and that our primary conclusion – that we are living through the most dramatic decrease in global poverty on record – holds. Indeed, given the fact that developing countries are enjoying their highest levels of sustained growth ever recorded, it would be a surprise if this were not the case. Ultimately, anyone who wishes to measure global poverty faces a trade-off between timeliness and accuracy. We view our analysis and that of the Bank as complementary approaches which weigh this balance slightly differently. Nevertheless, it is valid to ask whether the balance struck by the Bank – whereby policymakers must rely on poverty estimates that are up to 6 years old - is appropriate to today’s rapidly changing world. As implied by the title of our paper, measuring global poverty remains a challenge, despite enormous improvements in recent years (attributable in no small part to the efforts of the Bank’s Development Research Group). We would welcome feedback on two of the policy recommendations in our paper directed to the World Bank, namely (a) how the Bank can (further) improve capacity in domestic statistical agencies and (b) whether the Bank should consider publishing updated global poverty data every year.

The issue at stake here is not whether poverty is falling but by how much. We agree that about 25% of the population of the developing world lived below $1.25 a day (at 2005 prices) in 2005. The World Bank’s calculations imply that this will be down to 15% by 2015, while Chandy and Gertz reckon it will fall to 10%. I wish they were right, but, as I argue in my blog, they are almost certainly being optimistic. Their methods are built on unrealistic assumptions. In their reply, Chandy and Gertz agree that inequality has been rising in some (large) countries, including China, and they acknowledge that they have ignored this fact. Their defense is that “dramatic progress against poverty is possible even when a country’s income distribution is widening.” But that is not at issue. My point is solely that by ignoring the rise in inequality in China and India they overstate the reduction in the global poverty rate. That is also the outcome when one assumes (as they do) that all growth found in the national accounts is passed on fully and immediately to mean household consumption as measured by surveys. That is not consistent with what we know (especially, but not only, for India). They imply that there is no alternative to their method. But there is, and it has been in use for 20 years. The Bank’s poverty forecasts use a more elaborate version of the method used by the Brookings study, but with the difference that the Bank also allows for rising inequality in China and India and for divergence between the national accounts and the surveys. Building the forecasts on more realistic assumptions is feasible, and it is not that much more difficult. They also repeat their claims that the Bank has chosen to only use data that are up to 6 years out of date. That is plainly wrong. Indeed, the only poverty data used by Chandy and Gertz is from the Bank’s “PovcalNet” (http://econ.worldbank.org/povcalnet) site, which includes a great many data points after 2005 (as any reader can readily verify by visiting the site). So the data used in the Brookings study cannot possibly be more “up-to-date” than the Bank’s data. Furthermore, the Bank has been providing yearly updates and forecasts for the global poverty measures (published in it’s Global Economic Prospects and Global Monitoring Reports). These are built on more realistic assumptions, which is why we get a poverty rate of 15% in 2015 rather than 10%.

Submitted by MJ on
I do not have the technical know-how to comment on the methodological issues debated above. However, I think the Brookings report brings up an important point about the timeliness of official global poverty estimates. Speaking as someone who works for a bilateral aid agency, it would be extremely useful to have a sense of the state of global poverty today, rather than the state of global poverty in 2005. Would the World Bank consider publishing draft estimates of today’s global poverty rate? Failing that, is there any reason why after producing the global estimate for 2008 this year, you can’t produce an estimate for 2009 next year, as the authors propose in their paper?

Poverty measurement is based largely on household surveys. Unfortunately, these are not done annually for most developing countries; the average interval between surveys since around 1990 is roughly 3 years. Of course, national poverty analysis always uses the most recently available data. However, we only do the survey-based estimates of global poverty measures at about a 3-yearly interval, based on our judgment of whether we have enough survey data to yield a reliable estimate. The next update is due Fall 2011, which will go up to 2008. The lag has fallen over time (from 5 years in 1990 to 3 years now). But there is still a lag. Recognizing the need for more timely estimates, we do produce regular "forecasts." These use all the surveys we have available at the time, but also rely heavily on projections using national accounts data in lieu of surveys. However, we try to base these on realistic assumptions, as discussed in my reply to the Brookings team.