I am currently estimating spatial spillover effects of cement production on air pollution. I have 400 regions, among which 24 produce cement. I have created weight matrices and plan to apply spatial durbin model. As you know, we add cement production, other control variables and their spatial lags in SDM model. However, I face the problem of how to handle these zeros in the model. I tried to include dummy variable and log of cement production for producers, but in all specifications dummy variable turns to be negative and cement production turns to be positive. The sign of other control variables are quite intutive. What is best practice in such case besides creating dummy variables?