Everyone knows Otto von Bismarck’s supposed opinion about sausages and laws. Butchers and legislators, however, tend not to display much anguish about the production process. Artisans of historical statistics, in my experience, tend to get embarrassed even amongst themselves at the shortcuts and guesswork necessary to manufacture their product. It didn’t use to be this way. Just look at Paul Bairoch’s table notes:
In general, though, economic historians will be refreshingly blunt when it comes to the shortcuts they are forced to take to reconstruct the statistics upon which they rely, and, in that spirit, I’ve decided to write a series of posts on historical GDP reconstruction in historical and geographical contexts where there’s not much quantitative data upon which to rely. Recently, there’s been a bit of a lively back and forth between Jason Hickel, Joe Hassell and Max Roser, Steve Pinker and Branko Milanovic about a graph that purports to show the decrease in extreme poverty from 1820. The controversy circles around the quality of poverty statistics in the 19th and early 20th centuries. Given that no one back then was specifically counting how many people were living below the (price-level adjusted) $1.90 extreme poverty line, we have to make informed guesses, and the reliability of those guesses is essentially what’s disputed here. Most of the guesswork depart from the monumental work that the world’s economic historians have done in the past few decades on reconstructing historical per capita GDP, a collective project that is now incorporated, largely, into the Maddison Project database.
I completely share Branko’s enthusiasm for the Maddison Project: it represents a huge amount of careful work by very smart people. But it’s a work in progress, as its authors & compilers will readily acknowledge, and what I propose to do here is give the historical per capita GDP estimates—and, in particular, those for the non-Western economies—a bit of a poke with a stick. Both of the readers of the blog will know that I briefly examined the new Maddison historical GDP figures last year, and, in particular, their new estimates of African countries’ per capita GDP for 1950. Here, I’ll be a bit more systematic. We’ll go through, step-by-step, the process used to create per capita GDP estimates for countries that (unlike the example of England, which Roser has described in depth) often involve strong theoretical assumptions and non-traditional data sources. One of the funniest economic historians (if that’s saying all that much), Jan de Vries, wrote in a footnote in his book on the Industrious Revolution that “Economists can hope that their accumulated knowledge will make posterity better off than it otherwise would be, but only economic historians can make our ancestors better off than they ever knew they were”. This series hopes to explain exactly how we do that, and to assess, as fairly as I can, how seriously people outside economic history should take us when we try.
The new Maddison income estimates are created using a bespoke, simplified version of what we might call the ‘Malanima shortcut’. (1) In its original form, this ‘shortcut’ was used to derive estimates of per capita GDP for Northern Italy from the fourteenth century. Though you might complain that fourteenth century Italian polities had neither official statistics nor national accounts, the Malanima shortcut is designed to get around the lack of data about production. It consists, roughly speaking, of the following steps. First of all, estimate how much agricultural output is produced per person. Then, figure out how much non-agricultural output is produced, assuming a relationship between agriculture and the rest of the economy. Then, add these two things together, and divide by the number of people in the economy.
The adapted shortcut method (I’ll call it the ‘African shortcut’), adopted by the Maddison Project for the estimates of African colonial GDP in 1950, uses only two bits of information. The first is the real wage in capital cities, measured as the number of families (consisting of two adults and two children) that can be kept alive at a barebones subsistence level, on the wage of one adult male working fulltime. The second is the rate of urbanisation: how many people were living in cities? And that’s it. Everything else is, essentially, an assumption.
The benefits of this approach, if it turns out to be accurate, are of course massive: gathering real wages is time consuming, but often possible even in places without central statistical agencies. Klas Rönnbäck, for example has constructed a series of real wages for what is now Ghana in the 18th century, using the account books of the Royal African Company. Urbanisation can be measured by censuses and other traditional methods of population counting, but can also be estimated with other techniques, including archeological ones. But the potential problems are just as substantial: is it really possibly to describe the level of development in a given economy with two bits of data?
But how good is the shortcut method?
Of course, part of the problem in answering this question is that in order to determine how accurate the method is, you have to have other ways of accurately estimating GDP. We can try to do this the very recent past (basically, post-1960), but even here there are two possibly fatal objections: one is that the quality of African GDP estimates in the twentieth century is highly suspect, and the other is that even if the ‘shortcut’ is accurate in the modern era, there’s no guarantee that it will remain accurate when we use it to estimate income levels in economies that look very different to their contemporary equivalents. We can also compare the results of the shortcut method with other kinds of historical GDP reconstruction.
Probably one of the best test cases is England, for which we have very long and reasonably high quality GDP series. Alessandro Nuvolari and Matteo Ricci have recently used the ‘full’ Malanima shortcut method to re-estimate England’s GDP per capita and compare it to previous estimates, specifically Gregory Clark’s income-side estimates (basically, adding up all forms of income, like wages and rent) and Broadberry et al.’s output-side estimates (essentially, adding up everything produced in the economy). Nuvolari and Ricci’s results track the Broadberry et al. estimates reasonably closely, which, if nothing else, suggests that the full ‘shortcut’ method might be a reasonably good approximation of more traditional output-side methods.
What about the Maddison estimates for Africa? This is a bit harder, since historical GDP contruction using either the output or income approaches are non-existent for sub-Saharan Africa. However, while poking around in the colonial archives in the south of France, I came across an unpublished French estimate of colonial GDPs in West Africa for 1951, which is close enough to 1950 that it seems a useful comparison for the Maddison estimates for French West Africa. In later posts, I’ll dig a little deeper into how the French estimates were constructed, but for now, let’s just look at the results. The French expressed their estimates in 1950 French African francs, which are not easily convertible to the same units as the Maddison figures. The latter are given in ‘international’ dollars—which basically just means US dollars, adjusted for differences in the purchasing power of a dollar across time and space. Without going into excruciating detail on this point, we don’t really have the data required to accurately convert these numbers into equivalent units, so what I’ll do is just express both the Maddison numbers and the French colonial figures as a percentage of Senegalese per capita GDP. (So, for example, Côte d’Ivoire’s Maddison estimate will be divided by Senegal’s Maddison estimate, and its French colonial estimate divided by the French colonial estimate for Senegal). Since the Maddison estimates are meant to be about relative income levels within Africa rather than, necessarily, precise estimates of actual income per capita, this seems like a reasonable approach. When we compare the two series, we find major disagreements:
There are some fairly large disparities here. Mostly, the Maddison method minimises the difference between Senegal and the rest of the colonies of the group compared to the French estimates: in the case of Haute-Volta (what is now Burkina Faso), it estimates GDP per head from less than a quarter of the Senegalese level to somewhat over half. The major standout is Côte d’Ivoire. Maddison estimate per capita GDP there to be about 75% of Senegalese levels; the French administration, on the other hand, estimated that Côte d’Ivoire produced over 125% as much stuff per person as Senegal.
In the rest of this series, we’ll go into the estimation methods in much more (excruciating) detail to understand why these methods come up with such divergent results. And with luck, by the end, we will have come across a way to resolve these differences, and settle on more precise estimates of per capita GDP in French West Africa.
Outline of the series:
- Agricultural production
Estimates, new and old
(1) Malanima’s article on Northern Italy is probably the clearest exposition of the method, but various components of it are quite old: the method of estimating agricultural output goes back to Nick Craft’s seminal critique of the Deane and Cole estimates of English GDP.