Economic growth and poverty reduction in Africa

A perennial question in development economics is whether economic growth, by itself, is enough to reduce poverty.

The question came up in the most recent edition of Development Drums. Claire Melamed argued that the fact that so many of the world’s poor now live in middle income countries (which, by definition, have experienced a reasonable amount of economic growth) suggests that growth by itself is not enough to reduce poverty. Andy Sumner, in the same programme, said that there is some evidence that economic growth tends to increase inequality in societies that are already unequal, whereas the benefits will be more broad based in societies in which the starting point is more equal.

This graph by Maxim Pinkovskiy and Xavier Sala-i-Martin is very interesting. It shows the growth rate and the number of people living on less than a dollar a day in sub-Saharan Africa. The data are notoriously incomplete, but on the basis of these estimates, as the authors say (apologies for the econ-speak): “Poverty seems to co-move with GDP almost perfectly.”

Graph by Maxim Pinkovskiy and Xavier Sala-i-Martin

This graph implies pretty strongly that if you want to reduce poverty in Africa, you should concentrate on economic growth.

The entire article is well worth reading for its upbeat assessment about both growth and poverty reduction over the last fifteen years.  They say:

The sustained African growth of the last 15 years has engendered a steady decline in poverty that puts Africa on track to meet the Goals by 2017. If peace is established in the Democratic Republic of Congo, and it returns to the African trend (which is what happened to other African nations that were formerly at war), Africa will halve its $1/day income poverty rate by 2013, two years ahead of the 2015 target.

Moreover, African poverty reduction has been extremely general. Poverty fell for both landlocked and coastal countries, for mineral-rich and mineral-poor countries, for countries with favourable and unfavourable agriculture, for countries with different colonisers, and for countries with varying degrees of exposure to the African slave trade. The benefits of growth were so widely distributed that African inequality actually fell substantially.

21 comments on “Economic growth and poverty reduction in Africa”

  1. Great news if true. On that point, Martin Ravallion has some fairly serious concerns about the paucity of the data used to construct those trends (http://blogs.worldbank.org/africacan/is-african-poverty-falling), though he still thinks the overall trends are very positive.

    But leaving those aside, aren’t we still left with a situation in which economic growth *and* a drop in inequality have produced large falls in poverty? So it doesn’t tell us how much growth ‘by itself’ would have reduced poverty or really get at the classic ‘dilemma’ (probably a false one) of whether to focus on growth *or* reducing inequality. I haven’t read the full NBER paper, but it would be interesting to see if they ran any simulations assuming similar growth but static or increasing inequality.

  2. Andy Sumner is surely right to suggest that we cannot credit rising growth with rising equality. It’s nevertheless encouraging to see both phenomena in Africa, assuming more complete data would verify the Pinkovskiy/Sala-i-Martin results. But I’m (happily) at a loss to account for the rising equality. Can somebody propose an explanation?

  3. The paper combines relatively complete data on GDP with data on income inequality for country surveys. In their voxeu article, Pinkovskyi and Sala i Martin write that:

    “For countries and years with inequality data, we compute the distribution of income by fitting a lognormal distribution to the inequality data, whereas for countries and years without inequality data, we interpolate inequality on the basis of neighbouring years. If a country has no inequality data for the sample period, we interpolate on the basis of the average inequality of countries with inequality data.”

    On page 4 of their NBER paper, they declare they use 118 surveys to monitor 48 African countries. Notice that the paper aims at estimating income distribution and poverty rates for the 36 years period between 1970 and 2006.

    So with an average of roughly 2.46 surveys per country, the authors construct annual estimates of the income distribution for 36 years for each country. Put it differently, from 118 observations, they interpolate a dataset with 1728 observations.

    Even more heroically, some countries in their dataset actually have no surveys and the authors simply assume average inequality rates for these countries.

    I am very skeptical that we can understand any relevant facts on poverty trends and their drivers using data constructed out of statistical assumptions and not out of observation.

    1. Stefano

      You are right that the missing data are constructed. But almost all statistics are constructed that way. The question is whether there are enough observations here to enable the authors to construct meaningful statistics. That depends on the sample size, as you imply, but also on the distribution of the population from which those samples are being drawn. (For example, opinion polls in the UK often use one or two thousand people only: nobody says that they are meaningless because not every citizen has been asked.

      Take a look at the Ethiopia graph – figure 7 in the NBER paper. Ethiopia does have good survey data, and it mirrors the findings for Africa as a whole. So you can’t dismiss all the findings on the basis that the assumptions are too heroic. The margin of error may be large, but unless you know otherwise, this is the best estimate we’ve got.

      Owen

  4. Hey Owen, the paper’s dodgy I’m afraid.

    Here’s a letter from Lawrence Haddad and myself that went in the Guardian:
    http://www.guardian.co.uk/society/2010/mar/09/africa-aid-economic-development-bbc

    In sum we say –

    The paper does three things that suggest extreme caution. First, it manufactures 1,800 data points on inequality from surveys that cover only 118 data points: in other words, 94% of the inequality numbers are extrapolations from other countries and other years. Second, the poverty estimates rely heavily on government-reported GDP, when we know that GDP data from national income accounts do not match income levels recorded from household surveys. Third, using GDP per capita and the manufactured inequality data, the authors construct poverty rates for 48 African countries for each year between 1970 and 2006. The authors find few correlations between their manufactured poverty rates and structural features of the countries in the sample. This insensitivity to structural features either means that poverty has been reduced in every single location (unlikely) or that the data do not reflect reality.

    Lawrence also blogged at:
    http://www.developmenthorizons.com/2010/03/can-africa-make-poverty-history.html

    A better reference point is Ravallion’s ‘Inequality is Bad for the Poor’:
    http://www.eldis.org/assets/Docs/20092.html

    Here’s my summary –

    One of the main factors that determines how effectively growth will reduce poverty is ‘initial inequality’. As Martin Ravallion, head of research at the World Bank, wrote in his 2005 paper, Inequality is Bad for the Poor, ‘the higher the initial inequality in a country, the less the poor will share in the gains from growth’. Put another way, growth reduces poverty faster in countries with more favourable income distributions.

    Ravallion illustrates this by way of example, using two hypothetical countries, one with low initial inequality and one with high inequality. If it takes a low-inequality country 10 years to halve poverty, a high-inequality country growing at the same rate and with the same initial poverty level, would need 57 years to halve the poverty rate. In short, poverty responds slowly to growth in high-inequality countries. Or, to put the same point slightly differently, high-inequality countries will need unusually high growth rates to achieve rapid poverty reduction. The upshot is that devoting attention to equity could drastically reduce the overall cost of ending world poverty by making growth more effective in poverty reduction.

    Various international agencies are heeding this call and have argued in favour of taking equity more seriously – most recently the IMF and the WEF. There is a sense that the growth agenda or ‘growth-only delusion’ that emerged from Paul Collier’s work, which states that growth is expected to be the primary driver of poverty reduction, simply takes too long for some countries at high levels of initial inequality. Growth is fine when there is no rush. Further, the fact that gross domestic product (GDP) growth often happens without social, economic or political transformation might begin to explain the continuing levels of absolute poverty in new middle-income countries (and in the remaining low-income countries).

    This all points to the fact that in many countries, and especially in middle-income countries, poverty may be increasingly turning from an international to a national distribution issue.

    1. Andy

      I don’t buy your argument on this.

      The data are incomplete: that is certainly true. That means we should acknowledge that there is a wide margin of error around these estimates. But that does not means that it isn’t the best estimate we have. Your comment implies that it is systematically wrong – that it overstates the decline in inequality. But you have no basis for asserting this: you do not point to a source of bias in the estimation, or to alternative, better statistics which show something different.

      Where we do have data – such as in the case of Ethiopia, Africa’s second most populous country – the story of rising growth and falling inequality is exactly reproduced. That suggests that the paper is at least partly right.

      So let’s acknowledge the uncertainty and margin of error, but if you want to make an allegation of statistical bias I’m afraid you are going to have to stand it up with some evidence. Until you do, this is the best estimate we’ve got.

      Nothing in the paper contradicts the point you make about Ravallion’s paper – that countries with lower rates of inequality will, as a matter of arithmetic, be more able to translate economic growth into poverty reduction. Other things being equal, we should prefer more equality, both for its own sake and because it means that growth will translate more quickly and easily into poverty reduction.

      But what if more equal societies grow slower (e.g. because the amount of state intervention needed to create a more equal society also has the effect of stifling business or disincentivizing investment)? Then we’ve got a real decision to take: how much growth would we want to give up to achieve greater equality, and hence a stronger relationship between growth and poverty reduction? My guess: not much.

      Owen

  5. Is there any strong evidence that more equal societies do grow more slowly? I was under the impression that there is an emerging consensus around high inequality being bad for growth – as well as poverty reduction.

    1. Jim – Not that I know of. Cross country statistical work on the determinants of growth is notoriously hard to do – that’s why many econometricians retain a healthy scepticism about the aid-growth regressions, for example. I’m just saying that while Andy focuses on the equality part, we need to keep our eye on the growth part too.

      Owen

  6. Thanks for your reply Owen.

    Opinions polls are designed so that their results will be statistically representative of the population from which the sample is extracted. There are statical techniques to ensure correct representation. In the paper in question:

    (1) the few observations used are simply those available, not a subset that can be considered as representative. It may well be that big household surveys are carried out in years with certain characteristics (no war, no disasters, no macroeconomic shocks), which will bias the results. In Ethiopia, for example, the agricultural sample survey (it’s not an HH survey, I know, but it’s just an example) was not carried out in 2002/2003, ie during the worst drought of the decade. Similarly, and more to the point here, the HICE was carried out in (I believe) 2005 and 2009, missing for example the price crisis of 2008 and the floods of 2006/7.

    (2) for some countries there are simply no observations on inequality and horizontal extrapolation methods are used. If an opinion poll in the UK doesn’t sample, say, the political preferences of any South Asian immigrant, on what basis can we say anything about the preferences (or the trends in preferences) of this subset of the population on the basis of average preferences in other groups? Again, we can imagine that the countries missing are characterised by particularly bad outcomes and/ or bad trends. Having no survey in 36 years surely signals something about the institutions of a country, doesn’t it?

    In the robustness section of their paper, the authors discuss two types of sample selection (few surveys and a higher percentage of consumption surveys towards the end of the survey). But they do not discuss the crucial issue of whether they are over-sampling more equal (or more inequality reducing) countries and hence whether there will be a systematic downward bias to their inequality intrapolations and extrapolations. They do not give details of the 118 surveys they use either, hence it’s difficult to say. But this is surely a point they should acknowledge.

    Finally, on Ethiopia, I wasn’t aware of the falling inequality issue. The paper below by Dercon and Hoddinot (up to the 2004/5 HICE)equally doesn’t mention decreases in inequality. And a WB study presented last year documented that urban Ethiopia in the last 10 years actually saw one of the highest increases in inequality in history.

    http://papers.ssrn.com/sol3/papers.cfm?abstract_id=1259341

  7. Hi Owen. The Sala-i-Martin paper is deeply flawed for the 3 reasons we gave in the letter to the Guardian. I just don’t see how it can be defended.

    On the bigger point, economic growth and poverty, when I learned economics, I was taught that growth is a necessary but not a sufficient condition for poverty reduction. This seemed radical to those who said growth is sufficient. But I think the so called radical position is too benign. I think we should regard growth as we do technology–something that could be good or bad for poverty reduction, depending on the governance of it. At the extremes we have resource curses (growth in pure GDP terms that is corrosive), but the variation in the impacts of growth on poverty (by sector, location, initial inequality, policy environment, natural environment, market rules) suggests to me that the governance of growth is vital. Growth is not only good or neutral for poverty reduction, it can also be bad. We need to be much more deliberate about setting up growth to reduce poverty.

    1. Lawrence

      Thanks. With respect, I thought the paper was rather better than your comment on it.

      Your so-called three points are really just one point: that we don’t have sufficiently complete and reliable data. With that I completely agree. But it is nonsense to suggest that without 1800 data points the analysis of the paper is “deeply flawed”. If you have inequality data for a country for two years, and you interpolate missing data points in between, you are not going to get nonsense or “deeply flawed” estimates of inequality in the intermediate years. It is true that if you have less data, or doubts about its reliability, then you have a greater margin of error; but acknowledging that there is a large margin of error is quite different from saying that there is some reason to think that the results are biased.

      Indeed, the only basis you give in your Guardian letter for thinking that the results might not be right (as opposed to merely being subject to a significant margin of error) is that they don’t accord with your priors:

      The authors find few correlations between their manufactured poverty rates and structural features of the countries in the sample. This insensitivity to structural features either means that poverty has been reduced in every single location (unlikely) or that the data do not reflect reality.

      Unfortunately, you have got this wrong: the authors don’t claim that poverty has been reduced in every single location (look at figure 10, for example). That would indeed be a finding that would take some believing. The authors claim something rather different: they say that there is no systematic difference in the extent of poverty reduction between different types of country according to whether they are large or small, landlocked or not, and whether or not they are rich in natural resources. This is an interesting finding, and the fact that it does not confirm your priors does not make it flawed. (It might turn out to be wrong, of course; but saying you don’t believe it doesn’t make it so.)

      Though your reasoning is faulty, your conclusion is right: the governance of growth matters. We probably don’t know as much as we would like about what constitutes good governance of growth, but it does not seem likely to be controversial to say (for example) that the way that Ghana handles its forthcoming status as an oil exporter will have a significant impact on how far its economic growth reduces poverty in Ghana.

      Accepting that the data leave a lot to be desired (and what a shocking indictment of the development community that we have invested so little in statistics), such evidence as we have suggests that Africa is growing rapidly, and that this growth is in fact reducing both inequality and poverty. Nothing you’ve said in your comment here, nor in your Guardian letter, seems to me to be grounds to dispute this.

      kind regards
      Owen

  8. One more point on this interesting debate.

    “the governance of growth matters” seems to be a point of consensus.

    It seems to me, a better angle to look at the whole issue is not focus so much on the poverty reduction impact of growth, but on that of government expenditure, which is the real variable policy makers can control in order to affect the real economy.

    There should be more research into the poverty reduction elasticity of different types of government expenditure and fiscal policy. Preliminary work from IFPRI for Ethiopia, for example, shows some real trade offs: investing in rural infrastructure has higher poverty reduction elasticities but lower effects on growth compared to investment in urban areas.

    Maybe, as Sala i Martin argues, we are on track on achieving the poverty MDG and growth is responsible for a fair share of this positive record (although, as I said in the above posts, I am really not convinced by their empirical methodology). But stopping there gives misleading messages to government such as Ethiopia’s, where, if IFPRI’s analysis is correct, more poverty reduction would be achieved if public investment is diverted towards sectors that don’t exhibit high growth multipliers.

  9. Missing in this conversation is the (lack of) reliability of the dollar a day line itself (which is actually 1.25 in 2005 USD now, and will surely be revised again when the next round if ICP data come in). The 1.25 line is deeply flawed. There are a number of reasons, but perhaps the clearest is that the line is insensitive to the costs faced by poor people in the goods and services they consume. This occurs twice: first in the conversion of household survey data to US dollars, which relies on a conception of purchasing power parity that reflects the price of all goods and services, rather than those consumed by the poor, second in the use of consumer price indices to convert to the base year for purposes of international comparison.

    There are many, many other problems with the line. The most sustained critique is available here: http://www.socialanalysis.org. Deaton’s work on these questions is also quite relevant. Stiglitz et al edited a recent volume on global poverty measurement which is also quite useful.

    It is worth noting that Sala-i-Martin has been proclaiming quite massive poverty reductions for some time. In 2002, he argued that the dollar a day poverty rate had fallen to 5 percent as of 1998. http://www.nber.org/papers/w8904 . I know this is a bit ad hominem, but it is at least a reason to give pause and consider why his work consistently shows reductions of poverty and inequality.

  10. Perhaps worth looking at other evidence on growth and poverty to broaden the debate out a bit:

    Martin Ravallion’s recent work at the World Bank comparing growth and poverty reductin in China, India and Brazil finds quite strikingly different rates of poverty reduction per percentage increase in GDP in each country. So while growth is associated with poverty reduction in each case, it has very different effects depending on intial starting points, changes in inequality etc.

    A recent summary of the evidence on growth and MDG outcomes, produced by ODI, found that while there is a consistent positive correlation between GDP growth and improvements in MDG 1 (dollar a day poverty and hunger), there was no evidence of any correlation between GDP growth and the other MDGs, which depend more on how governments use growth to fund public servcies etc. It’s available at: http://www.odi.org.uk/resources/details.asp?id=4892&title=millennium-development-goals-equitable-growth-policy-brief

    Of course we’ve been here before – Dollar and Kraay’s 2001 paper ‘growth is good for the poor’ sparked off a huge debate on this theme, which converged on the question of how the benefits of growth are distributed and how the impact of growth varies hugely by country. That has left me rather sceptical of the value of regional aggregates for answering the growth and poverty question – whatever the quality of the data, the averages tend to hide as much as they reveal.

  11. Dear Owen,
    Thank you for bringing the Pinkovskiy and Sala-i-Martin paper to our attention.
    You bring about an interesting point: that this paper gives us empirical evidence on the relation between growth and poverty. It doesn’t seem to bother if the evidence is on a claimed set of 1728 observations more than tenfold the actual number (118). Here I fully agree with Stefano’s caveats.
    There is other structural problem in the paper, though.
    Quoting the article:
    “The crux of the methodology is to assume that the distribution of income in each country and each year has the same functional form, with changes in GDP and inequality manifesting themselves through changes in the the parameters of this form only.”
    They infer 1728 income distributions based on theoretical stochastic distribution functions. Because there is no statistical certainty on which function (lognormal, Gamma, Weibull, or even other) better fits the data, so the authors choose one and test for the others. The choice they do is important, because those 3 distribution functions only require 2 parameters to specify their form, and that’s handy: GDP per capita works as the central measure parameter, Gini and interquantile differences for the dispersion parameter. The problem I see with it it this process only adds to the artificiality of the “dataset”.
    Finally, because they don’t have much data on dispersion (one tenth of all “observations” they use), most of the variation in the income distribution happens because of changes in the other parameter: GDP per capita (GDPpc). Changes in GDPpc (GDPpc growth) are commanding the observed changes in the income distribution in the story Pinkovskiy and Sala-i-Martin are telling.
    If we look at the final result we seek, i.e., changes in poverty, if, because of lack of data, inequality does not seem to change much, almost all changes in poverty must mirror changes in GDP.
    It stands to me clear than, that the result Pinkovskiy and Sala-i-Martin reach is commanded much more by the method they use that from the (weak) data they have.
    As a research paper this is not a problem. There is a method (a rather elegant one, I might add) that is proposed, there are serious problems with the data, but from that methodology and the weak data some provisional results can be observed. Nothing wrong with it. What concerns me is if we overlook the weaknesses and assume that the provisional results are good enough to be taken into consideration when we discuss whether extreme poverty in Africa is in a sure path of being overcome. If the method and not the data makes it so that “poverty seems to co-move with GDP almost perfectly” one should use caution and begin by questioning the method and the data, before risk overoptimistic stances on such a serious issue.

  12. Hi Owen, maybe we will both end up reviewing the Xala-i-Martin paper for some journal…I suspect one could spend a whole lifetime arguing over this paper.

    By the way I noticed a typo in the Guardian letter (in the para you highlight) which should read “The authors find few correlations between their manufactured poverty rate declines and structural features of the countries in the sample.” I had missed out the word “declines” in the letter.

    And they really are 3 points: (1) complete data on national income accounts, yes, but we know there is weak correspondence with income from survey data (Deaton, Robinson etc), (2) generating 94% of the inequality data — poverty rates are very sensitive to small errors here — just look at the way FAO generates its inaccurate undenourishment data in much the same way. Its true that the errors may not be systematic in any given direction, but this low ratio of signal to noise is worrying and (3) the declines in manufactured poverty rates cannot be explained by any structural features of the countries involved. For me that means that the data and assumptions do violence to reality.

    But, as Stefano points out, the much bigger issue is the weak governance of growth.

  13. Owen,

    I read this post, and subsequently the dialogue, with much interest.

    As a statistics wonk, I have a question for you. I went to the UN-WIDER inequality database they used (http://www.wider.unu.edu/research/Database/en_GB/wiid/), and checked out the country-years for which they had information. Here’s a few of the poorly represented countries, with the years for which they have data:
    Algeria – 1988, 1995
    Benin – 2003
    Burundi – 1992, 1998
    CAR – 1992
    Congo – 1958
    Guinea-Bissau – 1991, 1994

    By contrast, countries like Kenya and South Africa are well-represented.

    If, say, 80% of the data comes from countries like those, is it even a valid approach to interpolate to other countries across decades of no data? Or should this simply be taken into account as something that increases the error bars?

    Also, you challenged Mr. Sumner to point out a source of statistical bias to back his claims. Might the recent research on the strong correlation between GDP and publications (http://blogs.worldbank.org/africacan/too-little-knowledge-is-a-dangerous-thing) be one source of bias? That is, countries that are doing better are more likely to have shown up in this survey?

    I don’t mean to attack the post or the paper. I’m just asking the extent to which these considerations are relevant, understanding that I don’t really have the expertise to criticize the methodology. I’d love to hear your thoughts on either of these issues I raised.

    Thanks for the thought-provoking post! I thought it was entertaining and engaging even before reading the lively dialogue.

    Take care,
    Nathan

  14. Hi Owen, I love your blog and the discussion here has been important and well intentioned.

    Here is a paper by Antonio Ciccone and Marek Jarocinski that blows the bottom out of the entire field. http://www.fedea.es/pub/papers/2009/dt2009-36.pdf

    I am inclined not to believe in any macroeconomics concerning anything more than very loose general approximation of trends. Especially ones doen by Sala-i-Martin, who I find is one of the worst offenders of fuzzy statistical extrapolation.

    To paraphrase Chris Blattman, “Have you ever seen the sausage factory that is GDP calculation in Chad?”

  15. Lawrence & Owen

    The statistical data presented in the Sala-i-Martin report as well as their analyses are important, but what is more important to actual poverty reduction as you both agree, is governance. The same countries with high poverty rates suffer from incredibly high and devastating corruption at all levels. While there might be economic growth as shown in numbers, they don’t always translate into poverty reduction for majority of the people. Therefore, conducting actual observations and bring poverty research closer to poor people numbers will be great. It will be wrong to assume that poverty is on its way to being halved by 2013 solely on the basis of Sala-i-Martin’s study.

    The researchers might think of a follow-up study.

Leave a Reply

Your email address will not be published. Required fields are marked *

Published by

Owen Barder

Owen is Senior Fellow and Director for Europe at the Center for Global Development and a Visiting Professor in Practice at the London School of Economics. Owen was a civil servant for a quarter of a century, working in Number 10, the Treasury and the Department for International Development. Owen hosts the Development Drums podcast, and is the author Running for Fitness, the book and website. Owen is on Twitter and