On the Measurement of Importance
35 Pages Posted: 17 May 2019
Date Written: May 10, 2019
When cost-benefit analysis is infeasible, many empirical papers use standardized beta coefficients, Shapley values, or partial R² to demonstrate that their relationship of interest is important. I show that these measures of importance are flawed, and therefore propose two new approaches to measure importance. While the ceteris paribus approach captures the independent effect of a regressor on the dependent variable, the non-ceteris paribus approach explicitly considers whether this effect is reinforcing or going against the effects of other correlated variables. I apply these complementary methods to assess the importance of the determinants of long-run growth. I find that legal origins, malaria ecology, and distance to coast are the three most important factors explaining contemporary development, ceteris paribus. Religion, genetic diversity, and slave trade intensity are also relatively important. In comparison, ruggedness, the share of the population of European descent, and the timing of the Neolithic transition appear to be unimportant. The effects of malaria ecology, of the slave trade, and of the share of the population of European descent are mutually reinforcing.
Keywords: Importance, Effect size, Regression, Long-run growth
JEL Classification: B4, C18, O4
Suggested Citation: Suggested Citation