Dear Statalisters
Can someone put me straight on the interpretation of the following Hausman
test statistic (Fixed v Random Effects).
Test: Ho: difference in coefficients not systematic
chi2( 10) = (b-B)'[S^(-1)](b-B),
S = (S_fe - S_re)
= 165.97
Prob>chi2 = 0.0000
I take this to mean that fixed effects are not appropriate, since there are
quite large discrepancies between Fixed and Random coefficient estimates.
Questions
1. Is random effects just a test of specification rather than a 'model'
in its own right? i.e. is it indicating the pesence of
group heteroscedasticity/autocorrelation/both?
2. How do I proceed? [I've thought of trying xtgls -specifying i() and
t()]
If trying xtgls is reasonable, I need some advice on the t() variable.
The data I'm using is sports match data
i(team) = 1 to 18 represents all the teams in the league
However, I have six years of (home) match result data (points difference)
for each team
I think it is reasonable to treat (for example) the possible autocorrelation
element as different season by season because in cricket squad changes can
only be made between (rather than during) seasons in this sport.
3. Is it possible to stratify t() in this way, while maintaining the
i(team) effect?
Many thanks
Ron Dorsey
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
