Fixed effects - cross-section, period or both?
Posted: Thu Jun 12, 2014 1:42 am
Hi,
so I familiarized myself a little with the Hausman test and found out how to do it in EViews. But I can't seem to find out whether to use Fixed effects (assume Hausman test p-value =0) for cross-sections (will further call it CS), periods or both.
What I did is: I estimated my model with CS random effects, Period none. Then I ran the Hausman test, p-value was 0, so I concluded CS Fixed effects is appropriate.
Then I re-estimated the same model with Period random effects, CS none. Then I ran the Hausman test, p-value was 0, so I concluded period fixed effects is appropriate.
Does that mean that I should use fixed effects for both when I want to estimate my model or is it not possible to conclude when you test for each effect individually as I did above (because I have not found a way in EViews to test for both effects at the same time/in the same estimation)?
Thanks!
EDIT: another problem is that when I now want to estimate with Period random effects, I get the error message that I have too few cross-sections. But my panel contains over 65000 cross-sections?!?!?! (panel structure: years 2000-2012, annual data; >65000 cross-sections, 445116 observations in total, 13 variables used in the regression mentioned above)
EDIT2: I now get the message "Cross-section test variance is invalid. Hausman statistic set to zero". Is it possible that fixed effects that are possibly inherent in my panel have been taken care of by my control variables? For example, I have an industry control variable. Industry should be a cross-section fixed effect.
Can i just proceed by estimating my model with neither cross-section nor period fixed effects?
so I familiarized myself a little with the Hausman test and found out how to do it in EViews. But I can't seem to find out whether to use Fixed effects (assume Hausman test p-value =0) for cross-sections (will further call it CS), periods or both.
What I did is: I estimated my model with CS random effects, Period none. Then I ran the Hausman test, p-value was 0, so I concluded CS Fixed effects is appropriate.
Then I re-estimated the same model with Period random effects, CS none. Then I ran the Hausman test, p-value was 0, so I concluded period fixed effects is appropriate.
Does that mean that I should use fixed effects for both when I want to estimate my model or is it not possible to conclude when you test for each effect individually as I did above (because I have not found a way in EViews to test for both effects at the same time/in the same estimation)?
Thanks!
EDIT: another problem is that when I now want to estimate with Period random effects, I get the error message that I have too few cross-sections. But my panel contains over 65000 cross-sections?!?!?! (panel structure: years 2000-2012, annual data; >65000 cross-sections, 445116 observations in total, 13 variables used in the regression mentioned above)
EDIT2: I now get the message "Cross-section test variance is invalid. Hausman statistic set to zero". Is it possible that fixed effects that are possibly inherent in my panel have been taken care of by my control variables? For example, I have an industry control variable. Industry should be a cross-section fixed effect.
Can i just proceed by estimating my model with neither cross-section nor period fixed effects?