Why growth regressions are so scary

A lot of empirical research on growth and development has been done with cross-country regressions. This has been in particular favored by the availability of the Penn World Tables, which are a careful exercise of providing macroeconomic data that is comparable across countries. But it is obvious that for many countries, the data remains of poor quality, and the makers of the Penn World Tables label each data point with a quality grade. Unfortunately, nobody takes those labels into accounts.

Why would they matter? Because data of poor quality may be subject to major revisions, for example. And major revisions of some data points (which have a tendency to be outliers) can lead to changes in the empirical results. This is what Simon Johnson, William Larson, Chris Papageorgiou and Arvind Subramanian explore by trying to replicate dozens of studies using the Penn World Tables, using various versions of the database. The results are sobering. Essentially, the results of every study using the panel data can be reversed. Studies only using long-run averages are much safer, though. Do not use the Penn World Tables blindly, and do not believe results using them before scrutinizing them yourself.