Regression Costs Fall, Mining Ratios Rise, Publication Bias Looms, and Techniques Get Fancier: Reflections on Some Trends in Empirical Macroeconomics

The costs of running a regression are falling rapidly, perhaps by 10-15 percent per year. Considerations of costs and benefits suggest that this will cause an increase in the mining ratio, that is, the ratio of the number of regressions made to the number published. An individual researcher’s full set of the regressions made, and the ratio itself, are private knowledge of that researcher—but informal polls suggest that the average ratio is now well above 25. The wave of meta-studies of published results allows some intuition about the consequences of a rising mining ratio. One consequence would be that publication bias rises as well. I suggest that the rising mining ratio is mitigated somewhat by the rapid rise in the number and sophistication of econometric techniques, even though it appears that the marginal productivity of new techniques is falling.