Unbalanced panel data: How to chose between fixed effects or random effects

[niam]

Hello;

I have an unbalanced panel data set and would like to know which method to use for estimation. In PROC PANEL, the hausman test is not working, since I have an unbalanced data set, I get a warning which says: "

Hausman statistic cannot be calculated because different variables were dropped in the  random effects model than in the fixed effects model". I tried Breusch-Pagan tests, but again since my panel data set is unbalanced, the test results are not valid. I get this warning :"The Breusch-Pagan Test is not valid in the presence of an unbalanced panel, use the statistic with caution."

Do you know of any statistical method that can help me with choosing,  between the fixed effects and random effects estimation methods? If not, how shall I present the results in an academic journal, specially when the results of two methods are very different?

Thanks for reading this post

[Ksharp]

I remember PROC MIXED has an option which can tell you whether has the random effects .

Sorry, I forgot. Maybe you should post it at Statistical Forum , there are some experts can help you more.

[niam · Accepted solution]

I think the problem was due to multicollinearity in the independant variables. I fixed it. Thanks

[raqthesolid]

HI how can we fix for multicollinearity problem ??