Let me condense my earlier recommendations into just two for starters. First, raw correlations across countries with no attempt to control for other variables are highly suspect. Take the data on "economic freedom" and growth and add a third or fourth variable. Try, for instance, literacy or average years of education (from World Development Report). What happens to the coefficient on "freedom"? Second, look at the error plot. Which countries seem to be doing better than the trend? Worse? Is there a pattern that comes to mind? This can be a basis for trying new variables. Perhaps the most potent approach -- but also the most difficult -- is to go into the measurement of the "freedom" index. Chances are, the criteria were selected through a trial and error process that generated the best fit. (No one admits to doing this, but I've seen too much evidence of it over the years not to suspect it.) This may make the variable vulnerable. Propose an alternative criterion for "freedom" -- even one that is still based on neoliberal philosophy. Retabulate the sample set on this new measure; what happens? The problem is, unless the original authors have provided a vast amount of raw data, this sort of reconstruction is very labor-intensive. Peter Dorman
