Morten Jerven on Measuring African Poverty and Progress
What this covers
Morten Jerven of Simon Fraser University, author of Poor Numbers, discusses with Russ Roberts the statistical foundations of what we know about African economic growth and poverty. The core argument is that GDP estimates for sub-Saharan Africa rest on such shaky methodological ground—varying wildly depending on which data set and assumptions are used—that the econometric studies built on them offer little real insight. Jerven traces the problem to structural gaps: African statistical offices lack reliable data on domestic food production and the informal sector, forcing them to rely on crude proxies and guesswork that can leave level estimates off by 50 to 100 percent. Beyond this, the methods themselves shift over time within single countries. Tanzania's GDP series, for instance, switched methodologies in the 1990s regarding how the unrecorded economy moves relative to the recorded one, producing a sudden apparent growth recovery that settled academic disputes at the statistical office rather than at the data.
The conversation ranges across the practical and political costs of these breakdowns. When ranking sub-Saharan economies by GDP per capita in year 2000, different data sources agreed on only one country's position out of 45, with discrepancies so large that Liberia appeared among both the ten richest and the ten poorest depending on the source. Jerven distinguishes between validity (whether the level is correct) and reliability (whether the method stays consistent), arguing that neither holds reliably enough to support policy-relevant conclusions. The broader concern is institutional: because statistical agendas in poor countries are often set by foreign donors rather than domestic priorities, states lose the ability to know themselves and hold themselves accountable. The discussion closes with what might actually improve the situation—stronger, independent national statistical offices answerable to their own governments rather than external funders.
Jerven argues that GDP and growth statistics for sub-Saharan Africa are so unreliable—varying wildly across data sets, methodologies, and time—that the empirical studies built on them are essentially meaningless, and that fixing this requires strengthening national statistical offices for the sake of domestic accountability rather than global data sheets.
- African statistical offices lack data on the informal sector and rely on crude proxies, producing GDP estimates that can be off by 50-100%
- Methodological assumptions change over time (e.g., Tanzania), so growth rates and cross-country rankings are not comparable
- Improving data matters because statistics are a state knowing itself and the basis for political accountability
Economists treat measurable data as valuable regardless of accuracy, ignoring whether measurement reveals truth or mere correlation.
- Against the common argument that having some numbers is better than none (the Taleb anecdote of a Paris map being used to navigate Washington to New York), Roberts argues it is better to say something true than something unknowably true or possibly false, questioning the value of bad statistics as a guide.
“it's better to say something that's true than something that's unknowable true, or maybe false.”
- Roberts invokes the Einstein-attributed aphorism that not everything that counts can be measured and not everything that can be measured counts, to express his broader skepticism that econometricians take whatever data they can get, find correlations, and tell ex-post stories.
“Not everything that counts can be measured; not everything we can measure counts.”
Projections based on past relationships presented as measured effects mislead by concealing their methodological assumptions.
- Roberts argues the CBO's estimate that the 2009 stimulus created 500,000 to 3 million jobs was not a measurement of actual jobs created but a projection that ran regressions applying past relationships between government dollars and employment, holding constant structural conditions that no longer applied, so calling it a measured effect of the stimulus is intellectually bankrupt even though the CBO confessed the method in a buried paragraph.
“they had run a set of regressions and basically done a set of projections based on the amount of money being spent by the stimulus package, and holding constant the impact of dollars of government employment that it held in the past.”
Assuming unmeasured economic sectors remain constant is false even in developed economies, far more problematic in developing ones.
- Unmeasured household production in the US changes only slightly year-to-year but radically decade-to-decade, and the assumption that the unmeasured sector is roughly constant is false even in the United States and far more so in Africa where the size of the unmeasured sector changes radically.
“Year-to-year the changes are small. Decade-to-decade the changes are big even in the United States. Certainly in Africa they are enormous.”