Skip to content
Prev 75702 / 398502 Next

two-level poisson, again

On On, 2005-08-17, 08:52, Shige Song skrev:
You're not fitting the same model! `lmer' uses Gaussian random effects and
the default for `xtpois' is gamma random effects.

Also, note that even if you'd specified a Gaussian random effect (through
a `normal' to the right of the `,' in your `xtpois' call) the same fitting
criterion is not used since `xtpois' uses adaptive Gauss-Hermite
quadrature and `lmer' defaults to PQL.

For comparable results, try the following:

m2 <- lmer(.D ~ offset(log(.Y)) + (1|pcid2) + educy + agri, male, poisson,
method = "AGQ")

xtpois _D educy agri, e(_Y) i(pcid2) re normal

You may also want to try G??ran Brostr??m's `glmmML' package.


HTH,
Henric