Expected behavior: Not sure that it is a bug, but I can't find the answer anywhere. I am looking at analyzing candidate fixed-effect for the regression below (variable i.candidate). For different reason, I don't want to treat them as dummy variable directly into the model, in order to avoid having a reference category. It eases the interpretation process for what I want to do. Therefore, I put these fixed effect directly in the absorb parenthesis, and I create the variable directly from there (C_FE). From my understanding, this is relatively simple, and it should work quite easily.
Actual behavior: However, I face some issues : each time I relaunch the regression, I obtain different values for my C_FE. The ordering remains somehow the same, but the magnitude differs, which create problem later down the line when I try to interpret the results. From my understanding of the model, I didn't expect the fixed effects coefficient to change each time (especially since I don't observe this behaviour when I put the variable i.candidate directly into the regression - but as explain above, I DON'T want an omitted category). So I guess there are two steps at my question : is this behavior from the model normal ? If not, what could I do ?
Next to that, a few side questions that could help me avoid this issue/improve my results :
Is there a way to produce coefficient WITHOUT an omitted category directly from the regression ?
In case not and if I still have to rely on the little trick I used, is there anyway to get the standard deviation of the fixed effect coefficient C_FE ?
Bug report :
Behavior
Expected behavior: Not sure that it is a bug, but I can't find the answer anywhere. I am looking at analyzing candidate fixed-effect for the regression below (variable i.candidate). For different reason, I don't want to treat them as dummy variable directly into the model, in order to avoid having a reference category. It eases the interpretation process for what I want to do. Therefore, I put these fixed effect directly in the absorb parenthesis, and I create the variable directly from there (C_FE). From my understanding, this is relatively simple, and it should work quite easily.
Actual behavior: However, I face some issues : each time I relaunch the regression, I obtain different values for my C_FE. The ordering remains somehow the same, but the magnitude differs, which create problem later down the line when I try to interpret the results. From my understanding of the model, I didn't expect the fixed effects coefficient to change each time (especially since I don't observe this behaviour when I put the variable i.candidate directly into the regression - but as explain above, I DON'T want an omitted category). So I guess there are two steps at my question : is this behavior from the model normal ? If not, what could I do ?
Next to that, a few side questions that could help me avoid this issue/improve my results :
code : reghdfe relative_value [aw=squ_length], absorb(C_FE=i.candidate i.day) vce(cluster candidate)
Anyway, thank you Sergio Correia for this nice package !