[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: fixed effects and random effects
Sunil W wrote:
> One, when I run a random effects model on my data, I
> get sigma_u =0. How do I interpret this? Does it mean
> that the random effects model is not appropriate for
> this data? The fixed effect model seems to work fine.
I can't answer your second question, but I'll tackle this one. The
-sigma_u- estimate is zero, which implies that your model contains no
time-invariant errors. It's difficult to say whether this is plausible or
not, since you don't tell us what kind of models you're fitting to what
kind of data you have.
To understand whether or not fitting the random-effects model is sensible,
ask yourself this question: "Are my observations sampled _at random_ from
the larger population to which it belongs?" If it is, then you should have
good reason to expect your model's u_i and X-variables to be uncorrelated.
If you find that they _are_ correlated (perhaps because your observations
were _not_ chosen randomly), then the RE model is unlikely to be valid. At
that point, you'll almost certainly have to switch to FE.
In Stata, there is a quick and easy way to compare your RE and FE
formulations of the same model (and data):
(1) run your FE model;
(2) then run your RE model; and
(3) run -xthausman- to conduct a null hypothesis test of the difference in
If H_0 is rejected, the test suggests (but does not _prove_) that the use
of RE is invalid. However, I think it's important for you to understand
why running such a quick and easy test like this is useful, rather than
just running it willy-nilly and accepting the result as gospel truth.
I hope that helps. :)
CLIVE NICHOLAS |t: 0(044)7903 397793
Politics |e: firstname.lastname@example.org
Newcastle University |http://www.ncl.ac.uk/geps
* For searches and help try: