Sunday Morning Development Economics Experiment
In the course of my podcast listening this month, I came across this unknown-to-me economist being interviewed on the FT’s Economics Show - Lant Pritchett. One of his key contributions on his Wikipedia page is a “Pritchett Test”, which is a “four-part "smell test" for pro-development policies.”
In a cross-sectional comparison of levels, do countries that are more developed have more X?
In cross-sectional comparison of growth rates, do countries that have rapid growth in X also tend to experience a rapid increase in standards of living?
When we look at the few countries for which we have long historical records, do the ones that become much more developed also acquire much more X?
If we look for countries that switch from a regime of slow economic development to a regime of rapid development, do we see a parallel shift in the rate of growth of change in X?
Having a Sunday morning to spare, and wanting to test out the newest AI tools (Claude Sonnet/MCP and ChatGPT o1), I wanted to run a gamut of the World Bank’s Development indicators to see what such an explanatory baseline would yield.
Here are the results:
Nothing groundbreaking perhaps. But some interesting bits:
R&D indicators (researchers per million, journal articles, patents) all score well (2-3 tests out of 4), suggesting the paramount importance of managing R&D and knowledge production well.
Manufacturing value added scores only 1 out of 4 on development tests: perhaps successful countries have moved beyond manufacturing to services/knowledge economy
Financial Development Matters: Private sector credit scores highly, (domestic) financial deepening may be more fundamental to development than commonly emphasized. This is something emphasized by Mr Joe Studwell’s book How Asia Works a decade ago.
Energy use per capita scores well.
However, the Pritchett tests look for linear relationships - they set a baseline explanatory bar, which is not to say that there cannot be more non-linear targets. But I accept that if my favourite indicator doesn’t make it on the leaderboard, any claims about more complex non-linear relationships will need to explain why they're more compelling than these straightforward patterns.
Data and code available at: https://github.com/kennethtiong/pritchett-analyzer . Run it yourself and come to your own conclusions!