sergiocorreia/reghdfe

reghdfe omitting dummies that reg keeps

Opened this issue · 2 comments

mariaventura95 commented

Hi Sergio,

Using version 5.7 (2019). Not sure this is a bug but I thought I asked. I am running a regression with a lot of fixed effects and interactions, single variable clustering. I have tried to run the same specifications with reg and reghdfe (when I can) and it turns out that reghdfe omits (apparently, because of collinearity) one dummy variable that reg instead would keep. Now the issue is that I am trying to predict fitted values for different values of this dummy, and when this is omitted by reghdfe the predicted values are the same, as if there was no variation across that dummy (which is not the case).

Any idea of what is going on?

Thanks!

oscargarn12 commented

Hi Maria! Did you find a way to solve this issue? I am having the same problem :(

mariaventura95 commented

Hi Oscar,

No, I am afraid this remains a little but of a blackbox for me. It may have to do with the fact that reghdfe drops 'singletons' (according to its own definition, you should find it in the documentation), but I do not recall being able to fix it by specifying the option not to drop them...