Fixed, Random or Pool OLS.

[Luigi Lepore]

Hi, I have panel data containing firm observations for two years.

I have to choose Panel model among fixed (FEM), random (REM) and pool OLS.
Results confirm my hypothesis in all models (FEM, REM and OLS), but I don’t know which result I have to report in my paper.
I used Hausam test, BPLM, and Chow test.
I used Hausam test: for choose random effect or fixed effect.
Since the chi2(12) = 0.01; Prob > chibar2 = 1.0000
P-value=1.000, I cannot reject the Null.
Random effect model is more appropriate

I used Breusch-Pagan Lagrange multiplier (LM): for choose Random effects or OLS.
Since the chibar2(01) = 0.00; Prob > chibar2 = 1.0000
P-value is 1, we failed to reject the null, and conclude simple OLS regression is appropriate.

I used Chow test: for choose Fixed or Pooling OLS effect.
The result comes from the fixed effect panel model.
Since F(1, 202) = 0.01; Prob > F = 0.9299
Prob>F is bigger than 0.05.
Since the null is not rejected, I choose the Fixed effect model.

The three tests don’t give a univoque solution.
How can I solve?
Tkanks
Luigi

[Carlo Lazzaro]

Luigi:
p-value=1 in -hausman- is suspect.
Considering the results of the other tests, too, I think you should go pooled OLS.

[Luigi Lepore]

Dear Carlo,
thanks a lot.
Could the problem be generated by the fact that the values of my dependent variable (obtained by a content analysis of firms' non-financial report) for years 2017 and 2018 are very similar for each obseration? Moreover, also the value of the principal independent variable (proportion of independent directors in the firms' board) is very stable over the two years.
many thanks
bye
Luigi

[Carlo Lazzaro]

Luigi:
yes, this may be the reason of what you experienced, that is a negligible within and between variation of regressand and regressors.
If you look at -hausman- table (by the way: in your future posts please use CODE delimiters to share what you typed and what Stata gave you back. See the FAQ on this and other helpful posting-related topics. Thanks) regardless the specification of your regression model, the coefficients are basically the same.

[Luigi Lepore]

Thank you very much Carlo
Best regards
Luigi