Re: st: fixed vs random effect model

[David Jacobs <jacobs.184@sociology.osu.edu>]

If the Haussman test rejects random-effects, that test is telling you 
a random-effects approach would not produce consistent results.  And 
this would be a deadly flaw.

A reasonable answer to your question depends on how small your 
over-time (or within) variances actually are (which you don't tell 
us). Given the limited information you've provided, I'd go with 
fixed-effects, but tell the reader the degree to which specific 
explanatory variables approach time invariance.  One source on this 
is the Wooldrige econometrics book on cross-sectional and pooled 
time-series models (sorry but the book's at home, so I don't have a 
complete citation) on about page 270 (I think).

One way to assess this problem is to use the Stata routine -xtsum- as 
it provides over time (or within) standard deviations for variables 
in a pooled time-series model.

Dave Jacobs

At 10:42 AM 7/4/2010, you wrote:
> Good day Stata-listers,
> I'm apoloziging is the question may seems elementary for many of you,
> but i really need to check this before going on in my analysis. i'm
> running a panel data regression and after performing the haussman test
> the conclusion was that my model is a fixed effect one. The problem is
> located on my explanatory variables which display week variations, and
> as it is well known fixed effect model gives weak results in a such
> case. So should 'i use the random effect instead???
>
> Thanks a lot in advance.
>
> Ama.
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/