Aggregating unit effects is common practice in Political Science. For example, scholars in American Politics often add in state fixed effects when dealing with individual- or county-level data, and Comparativists include region fixed effects when studying panel data with multiple countries. We show that aggregating unit effects is only appropriate when there are strong theoretical reasons to believe that the individual effects are indeed highly similar within a given group. Since unit effects are intended to control for unobservable factors, all these unobservables per unit must be highly similar within a group. Aggregating up and including group fixed effects otherwise leads to biased parameter estimates. We illustrate this using Monte Carlo simulations and an application with real-world data.