[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: ordered probit and panels
What exactly do you mean by "direction of correlation", and how is
that important? -cluster- provides a correction assuming that there is
a population of individuals who respond in similar ways -- that's the
main assumption behind the design-consistent estimators of variance.
You don't need to have any special correlation structure for -cluster-
correction to work. -gllamm-, in the model like
gllamm response explanatory, i(id) link(oprobit)
imposes a more specific structure of a single random effect,
underlying all observations for a single individual, and thus assumes
constant factor loading and homoskedasticity of the remaining error,
just like all -xt* , re- commands do, so it has a very specific
correlation structure in mind... unless you override it with your own.
Of course -gllamm- is substantially more flexible, so that you can
estimate models for thresholds with it, if that is of any relevance in
your research -- say you suspect perceptions change over time.
I would personally start with -oprobit , cluster- and see how that
goes, and then migrate to -gllamm- providing -oprobit- results as
starting values. I'd be surprised if they changed a lot -- say more
than a couple of standard errors -- in any of the parameters though.
On Tue, 26 Oct 2004 09:14:17 +0200, firstname.lastname@example.org
> I would like to estimate an ordered probit with a balanced panel of 3-4
> waves. When I understand the handbook correctly, there is no xt-command for
> that purpose, and svyoprobit applies to a cross section only . Oprobit with
> clustering of individuals is no alternative either as the cluster option
> does not take into account the 'direction' of the 'correlation'. Besides
> gllamm, is there another alternative ?
> P.S: I know that I have posted a similar question one year ago, but there
> were only 2 waves available at that time. So I could apply oprobit,
> cluster(id) in that case.
* For searches and help try: