African inequality & poverty dramatically falling

[Surlethe]

... at least, according to this study (pdf).

The method the authors followed was relatively simple: they assumed that a population's income distribution can be reasonably modeled with a lognormal curve and then least-squares fitted lognormal curves to available GDP per capitas and Gini coefficients for African countries. The data started in the mid-'70s and went until 2006 (so it doesn't account for the recent recession) Then, based on this modeling, they tested various hypotheses about poverty and inequality -- does it depend on agriculture? Mineral resources? Has recent economic growth gone to the upper classes or lifted people out of poverty? Et cetera.

Here's the abstract:

The conventional wisdom that Africa is not reducing poverty is wrong. Using the
methodology of Pinkovskiy and Sala‐i‐Martin (2009), we estimate income distributions, poverty
rates, and inequality and welfare indices for African countries for the period 1970‐2006. We
show that: (1) African poverty is falling and is falling rapidly. (2) If present trends continue, the
poverty Millennium Development Goal of halving the proportion of people with incomes less
than one dollar a day will be achieved on time. (3) The growth spurt that began in 1995
decreased African income inequality instead of increasing it. (4) African poverty reduction is
remarkably general: it cannot be explained by a large country, or even by a single set of
countries possessing some beneficial geographical or historical characteristic. All classes of
countries, including those with disadvantageous geography and history, experience reductions
in poverty. In particular, poverty fell for both landlocked as well as coastal countries; for
mineral‐rich as well as mineral‐poor countries; for countries with favorable or with unfavorable
agriculture; for countries regardless of colonial origin; and for countries with below‐ or above‐
median slave exports per capita during the African slave trade.

The fall of both poverty and inequality from about 1990 on is striking, but I don't know nearly enough about African geopolitics or the African economy to try to isolate the factors that caused both the growth and the drops in poverty and inequality.

[hunter5]

It could also be due to more of the African governments stabalizing. Hard to build up your economy when you change governments every other week.

[Surlethe]

It could also be due to more of the African governments stabalizing. Hard to build up your economy when you change governments every other week.

That's a possibility - they cite the civil war in Congo as a major drag on African poverty reduction, for example (if it gets its act together and poverty keeps dropping like it has the last 15 years, the continent as a whole should exceed the first of the Millennium Development Goals, IIRC) - but like I said, I don't know enough about African geopolitics to tell if that's a major factor or not. Do you have any specific knowledge?

[K. A. Pital]

Damn, I have too little time to go over this. It seems to look solid, on the outline. However, the World Bank has this to say about it's own extreme poverty line ($1 PPP per day) in Sub-Saharan Africa:

Poverty and human development—Extreme poverty declined in Sub-Saharan Africa from 58 percent in 1990 to
50 percent in 2005. Nevertheless, poverty remains the highest among all regions, and the region experienced
the largest increase in the number of people living on less than $1.25 a day, from nearly 300 million in 1990 to
388 million in 2005.

http://siteresources.worldbank.org/DATA ... sa_wdi.pdf
So basically, the extreme poverty reduction looks marginal here - the shift from 58 to 50% and from $1 to $1,25 can't be called a quality change. I don't know what exactly caused the disrepancy.

Moreover, this was done probably based on old WB calculations - IIRC, in 2008 they have re-calculated poverty estimates towards a greater extreme poverty.

Statistics by the World Bank shows that between 1981 and 2005, in Sub-Saharan Africa, the $1.25 a day poverty rate has shown no sustained decline. The rate of poverty started and remained at around 50 percent. “In absolute terms, the number of poor people has nearly doubled, from 200 million in 1981 to 380 million in 2005,” the Bank’s figures revealed.

http://povertyworlddevelopment.suite101 ... _in_africa

So this means that in an absolute term, the number of poor living on $1,25 nearly doubled; whilst their percent has not declined. Probably people were pushed over the $1 boundary, which resulted in the skewed, "too high a reduction" results previously.

In fact, looking at the graphs in the paper, it seems that most people still get concentrated around the $1 line (except now being slightly to the right of it), but still far below $2 and $3 lines.

There's a table of World Bank Poverty "$1 PPP" estimates here from 1990 to 1999 to 2015 (as a projection).
http://www.imf.org/external/np/apd/semi ... angang.pdf

First of all, Sub-Saharan Africa seems to have an absolute and relative poverty increase from 1990 to 1999, and a projected poverty increase by 2015. Meanwhile China has both an absolute and a relative "$1 per day" poverty decrease.

Something is fishy here, but it takes a lot of digging through World Bank statistics to actually see what's wrong.

[Zixinus]

Perhaps we are looking at things stop getting worse rather than getting better?

Sorry if that sounds a bit meaningless, but that makes the most sense. From what I understood of Stas's and the abstract, things are less likely getting worse. Likely, given that several generations have passed in most of the ex-colony countries and people have begun to adjust to the new system, patterns have begun to appear that people can follow, etc.

[K. A. Pital]

I'm sure this is a result of the old World Bank poverty line being used; whilst the 2008 World Bank re-evaluation of the "extreme poverty" line from $1 to $1,25 (which has had it's reasons, namely that the movement from $1 to $1,25 does not represent such a great progress, as well as some PPP measuring changes), as well as some other poverty calculation adjustements, carefully omitted or not mentioned at all? The paper is done post-crisis, and this means they do not have any excuse not to mention the re-evaluation and the consequences of such.

[Alyrium Denryle]

I'm sure this is a result of the old World Bank poverty line being used; whilst the 2008 World Bank re-evaluation of the "extreme poverty" line from $1 to $1,25 (which has had it's reasons, namely that the movement from $1 to $1,25 does not represent such a great progress, as well as some PPP measuring changes), as well as some other poverty calculation adjustements, carefully omitted or not mentioned at all? The paper is done post-crisis, and this means they do not have any excuse not to mention the re-evaluation and the consequences of such.

If you guys want, once I get some spare time (I am browsing and posting a bit while studying) I can go through the paper, read its full methods, and then see about going through my wonderful dataset (For another argument, I went through the CIA factbook, and also the UN's Population Growth Assessment ) and do a bunch of stats in order to see if things are really getting better

[Surlethe]

Something is fishy here, but it takes a lot of digging through World Bank statistics to actually see what's wrong.

I don't have time to do that either, unfortunately. The method they used, as you noted, is straightforward and solid. They didn't use any World Bank statistics, but instead did LS fits of lognormal income distributions to PPP GDP and Gini coefficient parameters. There are thus two broad reasons for the discrepancy (and such exists - e.g., Figure 6, p. 11, has $1/day poverty in sub-Saharan Africa not rising above 43% or so in the lat '80s and then falling to about 32% by 2006). The reasons are: either African income distributions are very much not lognormal, or the World Bank methods of calculating these statistics are screwy. Without digging into the World Bank methods, I cannot say; however, there is some precedent for believing that income distribution might not be lognormal. I recall a paper mentioned recently where rent-seeking among corporation CEOs changed the shape of the high 10% of the income distribution; that might explain why lognormal fits are poor. Or other factors, such as very low mobility, high frictions, and poor institutional structures, might prevent a free-market process from leading to a theoretical distribution.

Edit: Sure, go right ahead, Alyrium. Another thing to do might be to do Gini/GDP fits to other income distributions and see if those better match known poverty statistics (assuming the World Bank has its numbers correct). Hell, we could cowrite a response paper ... .

[K. A. Pital]

They didn't use any World Bank statistics, but instead did LS fits of lognormal income distributions to PPP GDP and Gini coefficient parameters

Yeah (I noticed they only used PWT GDP data and calculated poverty from there), but there's an apparent large disrepancy between their $1 per day poverty estimates and those of the World Bank, which cannot be readily or easily explained. Moreover, the WB PPP revision shows the worst results for Africa even with their basic assumptions (as it is depicted on the graphs), which means the real picture might be somewhat different.

I'm not sure what caused the disrepancy either. I'll look into the WB poverty calculations - they must have laid out at least some methodology in their papers.

Meanwhile, the WB $1 PPP poverty line for Sub-Saharan Africa:

Population living on less than $1 a day (millions) - WB accounts:


Hunger in Africa - prevalence of undernourishment (%):

Based on FAO data, therefore should be reliable. Clearly and vividly contradicts the "poverty reduction" claims in Sub-Saharan Africa.

Sub-Saharan Africa "$1 per day" poverty rates from
http://go.worldbank.org/MMCKV4XRI0

1981 1984 1987 1990 1993 1996 1999 2002 2004
42.26 46.20 47.22 46.73 45.47 47.72 45.77 42.63 41.10

Indicate that there has been only a 6,62% reduction from peak-to-low in the extreme poverty line by WB calculations, which derive from household surveys (not from backwards calculations from GDP), and practically no reduction from 1981 to 2004, which is not the case in the paper's calculations.

Their explanation actually goes as following:

We rely heavily on nationally-representative household surveys for measuring poverty.
This is one of the purposes for which these surveys exist. There is no alternative to using survey
data for measuring the distribution of relative consumptions or incomes — “inequality” for short.
But there is an alternative source of data on average consumption, namely the national accounts
(NAS). (Given certain assumptions, one can derive standard poverty measures from the mean
and a suitable inequality measure.) We use NAS data in some aspects of our estimation
methods, notably in dealing with the fact that different countries do their surveys at different
dates, and we want to line them up in time to a common reference date. However, we do not let
the NAS data override the survey mean when both are known. In other words, we use the survey
at the survey date. In this respect we follow the standard, though not universal, practice in the
literature on poverty measurement.

Advocates of replacing the survey mean by the NAS estimate of national income or
consumption per capita argue that household surveys underestimate mean income or
consumption, due to deliberate under-reporting and selective compliance with random samples.
However, it is not clear that the NAS data can provide a more accurate measure of mean
household welfare than the survey data that were collected for that purpose. And, even
acknowledging the problems of income underreporting and selective survey compliance, there
can be no presumption that the discrepancies between survey means and the NAS aggregates
(such as private consumption per person) are distribution neutral; more plausibly the main
reasons why surveys underestimate consumption or income would also lead them to
underestimate inequality. Furthermore, the NAS-means method is clearly unacceptable when
doing an urban-rural split of global poverty measures, allowing for cost-of-living (COL)
differences, since neither the inequality measures nor the NAS means would then be valid.

Sala‐i‐Martin, one of the authors of the paper in question, is referenced by them as being in favour of discarding survey data in favour of calculations from the NAS:

Examples include Bhalla (2002) and Sala-i-Martin (2006).

Apparently they do not think his method is viable, accurate, or acceptable, as the above explanation says.

[Surlethe]

They didn't use any World Bank statistics, but instead did LS fits of lognormal income distributions to PPP GDP and Gini coefficient parameters

Yeah (I noticed they only used PWT GDP data and calculated poverty from there), but there's an apparent large disrepancy between their $1 per day poverty estimates and those of the World Bank, which cannot be readily or easily explained. Moreover, the WB PPP revision shows the worst results for Africa even with their basic assumptions (as it is depicted on the graphs), which means the real picture might be somewhat different.

The more I think about it, the less I trust the paper's results. Ultimately, the poverty rate is an empirical question, so (assuming the WB did their stats correctly) if the paper doesn't line up with the stats, then the only thing the paper shows is that you can't model African income distributions with lognormal curves. It's ass-backwards to go from unverified model results to pontificating on how statements from observations are wrong.

I'm not sure what caused the disrepancy either. I'll look into the WB poverty calculations - they must have laid out at least some methodology in their papers.

If they didn't, that's shit science.

By the way, here is the paper I referred to in my previous post. The method ought to be immediately generalizable to national incomes, I think.

[K. A. Pital]

I dug some explanations why WB $1 poverty line is so diverging from their calculations, and why WB thinks their methodoly is not acceptable versus survey methodology.

It's ass-backwards to go from unverified model results to pontificating on how statements from observations are wrong.

The survey method seems to be the standard poverty measuring method; what they offer is something critically different. Only by monkeying with the numbers and rejecting survey data in favour of an axiomatic statement about income distribution, can their result be achieved.

It's even more appaling that they seem to genuinely believe one can calculate it backwards despite both WB and FAO survey data showing that poverty has, by large, not been succesfully decreased; and malnutrition has actually increased - malnutrition is a very critical indication that poverty reduction is not significant.

So yeah, quite certainly something is wrong with their results. I have not seen any real examples where malnutrition has increased whereas poverty has significantly decreased. It's not just counter-intuitive, it's wrong.

Apparently they further muddle the issue by looking at all of Africa instead of just Sub-Saharan Africa, which follows different trend patterns, yeah? In that case, double caution is required.

EDIT: Yeah, just like I thought. Sala-i-Martin's methodoly critically examined by Branco Milanovic.

In the "Ricardian vice" part, Milanovic specifically discusses Sala-i-Martin's arbitrary application of distribution to the data, and many other issues. To say the least, if it's not "bunk" methodology, it's pretty close to that. Surveys aren't ideal, but accepting their data, whilst rejecting their means in favour of essentially ass-pulled means is just... man.

To conclude, Sala-i-Martin has succumbed to the temptation of piling one assumption upon another with the result that neither the author, nor the reader can any longer tell which is the part of each assumptions, individually or together, in deriving the final result. Here are, in summary, the Ricardian building blocks used by Sala-i-Martin in his calculations—with (*) signs indicating the assumptions imparting unambiguous downward bias to the results:
1. (*) A strange omission of countries with “disturbing rises” in inequality; then,
2. Use five data points to approximate entire distributions.
3. When these five data points are not available (84 percent of the time), extrapolate backward and forward in time. When only one observation is available; assume distribution stays the same for 30 years; (*) when there is no observation at all, assume everybody in the country has the same income.
4. (*) Treat distributions of household income across households as if they were distributions of per capita income across individuals.
5. Mix National accounts data (GDP per capita) and household survey data.
6. Mix expenditure and income data.

The point (1) is also present in Sala-i-Martin's new work, where he tries not only to soften the picture by taking all of Africa instead of SSA, but also by his attempts to paint rosy pictures with the exclusion of Congo-Zaire, etc.

And this is a very easy explanation on why it's bullshit from Rassilon:

To see why anchoring poverty measures to the national accounts can go so wrong,
consider the following simple example. The true but unobserved distribution of income is (say)
1,2,3 (person 1 has an income of 1, person 2 has income 2, person 3 has 3). The poverty line is
slightly above 1, so the true poverty rate is 1/3. We do a survey, and the three people respond
that their incomes are 1, 1.5 and 2. This also gives the right poverty rate. However, the survey
underestimates the true mean; the survey mean is 1.5. Now let’s assume (for the sake of
argument) that the national accounts do give the right mean of 2. If we assume that the survey
under-estimation is distribution-neutral then we multiply all three incomes by 4/3. The
"corrected" incomes are 1.3, 2 and 2.7 — implying that there is no poverty. We get the mean
right, but the poverty measure is way off the mark.

http://siteresources.worldbank.org/INTP ... atters.pdf

Hahah! That way even I could "multiply" underestimation and get no poverty. Seriously, I'm at loss - can that be considered "serious economics"?

The problem with Sala-i-Martin is that apparently he is a staunch believer in "capitalist convergence" and will do ANYTHING, including the use of dubious methodology above, to prove himself right - from arbitrarily excluding post-Soviet nations to applying distributions when the actual survey data does not fit his picture!

[Surlethe]

I read a rumor that there was a paper in the Journal of Economic Literature a few years ago criticizing their methodology, but I can't seem to find it.

First, a couple of criticisms you made that I think are too harsh.

Apparently they further muddle the issue by looking at all of Africa instead of just Sub-Saharan Africa, which follows different trend patterns, yeah? In that case, double caution is required.

At least they don't confuse between the two, and they include results for individual countries in the tables in the back. Their method is also duplicatable, so it's not like they're being fraudulent.

The point (1) is also present in Sala-i-Martin's new work, where he tries not only to soften the picture by taking all of Africa instead of SSA, but also by his attempts to paint rosy pictures with the exclusion of Congo-Zaire, etc.

They explicitly qualify the exclusion of Congo-Zaire; it's only misleading if you don't actually read the paper.

But I think otherwise the criticisms are correct. Another huge warning flag I should have seen: the complete lack of errors in the work. Interpolating over so few data points should create large errors in the mean and standard deviation of the distributions! Shame on me for not being so skeptical - shame on Sala-i-martin for not including them.

Only by monkeying with the numbers and rejecting survey data in favour of an axiomatic statement about income distribution, can their result be achieved.

The paper does serve a useful function - namely, showing that lognormal distributions, when applied with this methodology, are poor models of the actual African income distributions over time.


More generally, fitting distributions to capital accounts data should be taken seriously as an economic technique; it should just be done with an eye toward predicting survey data, instead of trying to contradict survey data. It would be useful to be able to work with an actual income distribution instead of estimations of poverty lines. For example, I would like to try to fit the income and gini data to the distribution predicted, e.g., in the paper I linked above -- the lower 90% lognormal, the upper 10% pareto (this should be in principle predicted by rent-seeking by those in power).