On determining the importance of a regressor with small and undersized samples

A problem encountered in, for instance, growth empirics is that the number of explanatory variables is large compared to the number of observations. This makes it infeasible to condition on all variables in order to determine the importance of a variable of interest. We prove identifying assumptions under which the problem is not ill-posed. Under these assumptions, we derive properties of the most commonly used methods: Extreme bounds analysis, Sala-i-Martin’s method, BACE, generalto- specific, minimum t-statistics, BIC and AIC. We propose a new method and show that it has good finite sample properties.