6 things worth knowing about in empirical economics

In the last couple years, I've learned a few new tricks (which, as an increasingly "old dog," makes me feel good):

1. Falsification/placebo tests.
   
2. The Mundlak regression: recover coefficients on time-fixed variables, test for CRE, and get the within estimates on time-varying regressors.
   
3. Simulation before estimation: Make sure your empirical method actually get the right answers when you generate your own data--and hence know the correct answers. (I credit this idea to Sam Kortum who advocated it at a talk he gave at the 2007 EIIT.)
   
4. Graphical display of regression coefficients (especially for letting the variable of interest enter without parametric restrictions on functional form). A helpful referee suggested this.
   
5. "Strict exogeneity tests" Do leads enter significantly? They shouldn't if effects are causal.
6. Use the linear probability model. It gives marginal effects directly and it doesn't have the Ai/Norton problem for interactions in logit & probit models.