Comments on Barro and Lee

Abstract

Beginning with Barro’s (1991) paper, we have seen a large number of cross-country growth regressions. In the present paper Barro and Lee make use of their recently assembled data set (Barr0 and Lee, 1993) on educational attainment, and they focus on the extent to which a few variables can distinguish between fastand slow-growing countries. The new data set is a valuable contribution and the empirical results are sensible (with one exception, which I will come back to). A weakness of the regression framework, however, is that it can say nothing about the channels through which various factors operate. I will begin my remarks by relating growth regressions to growth accounting, which allows one to speculate in a disciplined way about those channels, suggests some alternative interpretations of the current empirical results beyond those that the authors stress, and also suggests factors that may be useful to include in future work. The only peculiar empirical results in Barro and Lee’s work are the coefficients on the female schooling variables. Those variables behave oddly in almost all of the regressions, and I will conclude by offering an explanation for that behavior. On a purely mechanical level, growth in output must be generated by growth in inputs or by improvements in technology or by some combination of the two. Other factors that “cause” or “influence” growth government policy, market distortions, foreign trade, and so forth must work through these channels. Consider the aggregate production function