A reminder to my quant social science friends: if you estimate a regression with the goal of causal inference, your coefficients for control variables are not interpretable in the same way as your main coefficient.
Please consider this next time your referee complains about a "wrong sign" on your control variables.
