1 min readMay 30, 2018
Controls for race are picking up a whole of different effects — everything from geography to discrimination. The point of a control variable is to remove these effects from influencing the coefficient on the variable of interest, not to be directly interpretable.
It is true that correlation != causation, and that you have to be caution in interpreting observational studies even when they have controls.
This same logic applies to gun ownership rate.