Interpretation of significant treatment effects: Empirical bias and improved estimators

The two main outputs of an empirical analysis of a treatment effect are often the estimated effect size and its statistical significance. When considering a single treatment, these measures are well-motivated. For example, the estimated effect size is unbiased. However, when we consider multiple treatment effects, the effect size estimates of the significant effects are not unbiased. This is often ignored in empirical research. This research program will solve some statistical problems from selective reporting of significant effects. It will yield new insights on the statistical interpretation of significant effects in published research as well as new econometric tools that can be used by applied researchers to mitigate systematic bias in their significant estimation results.