R help - Jul 2009 - clogit comparison between Stata and R

Hello all

I'm moving back and forth between stata and R at the moment - of course,
using R whenever possible :-)

I'm running conditional logits on some panel data and I get slightly
different results and different N in the two programs.

In R I run
clogit(trans.dem ~ I(avg.gle_rgdp.500/gle_rgdp) + log(gle_rgdp) +
timesince.dem + I(timesince.dem^2) + timesince.dict + I(timesince.dict^2) +
p_polity2 + I(p_polity2^2)  + strata(ccodecow) + cluster(ccodecow),
method="approximate", data=univ)

and I get an n of 3747.

In Stata, I run
clogit trans_dem avg_gle_rgdp_ratio loggle_rgdp timesince_dem
timesince_demsq timesince_dict timesince_dictsq p_polity2 pol2sq,
group(ccodecow) vce(cluster ccodecow)

which I hope is the same model. I get a message "29 groups (935 obs)
dropped
because of all positive or all negative outcomes", and an n of 2812. Also,
the coefficients are slightly different.

I understand why Stata is dropping the groups with all outcomes the same...
this is inevitable in a conditional logit, right? Is R doing the same? And
what might be the cause of the difference in coefficients?

Cheers
David Hugh-Jones
Post-doctoral Researcher
Max Planck Institute of Economics, Jena
http://davidhughjones.googlepages.com

	[[alternative HTML version deleted]]

On Wed, 8 Jul 2009, David Hugh-Jones wrote:
> Hello all
>
> I'm moving back and forth between stata and R at the moment - of
course,
> using R whenever possible :-)
>
> I'm running conditional logits on some panel data and I get slightly
> different results and different N in the two programs.
That's probably because you are using method="approximate" in R.
>
> I understand why Stata is dropping the groups with all outcomes the same...
> this is inevitable in a conditional logit, right?
Yes.
> Is R doing the same?
Yes.

 	-thomas