[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Multinomial selection, but Selmlog/Svyselmlog not feasible
Thanks for the references. Ok, the user-written command -ssm- does not
allow the link mlogit and I don't know why. I speculate that there was
no simple good solution. For some related articles, see the whole SJ
issue 6-2, and the recent -gllamm- article at
Rabe-Hesketh, S. and Skrondal, A. (2007). Multilevel and latent
variable modelling with composite links and exploded likelihoods.
Psychometrika, in press.
On 5/22/07, Guido Heineck <email@example.com> wrote:
thank you for your answer. Apologies for not providing references of the
mentioned papers. Here they are:
* Bourguignon, F., Fournier, M., and M. Gurgand, 2007, Selection bias
corrections based on the multinomial logit model: Monte Carlo
Comparisons, Journal of Economic Surveys, 21(1), 174-205.
* Lee, L.F., 1983, Generalized econometric models with selectivity.
Econometrica, 51, 507–512.
Also, thank you for pointing me towards ssm. I only had a quick look at
it yet, but it apparently does not handle 'mlogit' for the selection
equation. Since one of the authors, Sophia Rabe-Hesketh, is also one of
the 'gllamm'-authors, I suppose multinomial selection cannot be
estimated by gllamm or why should they otherwise not include this option
in their 'ssm'-package?
Any further ideas?
> A possible alternative is the "gllamm wrapper" ssm. See -findit ssm-
> and the associated article in Stata Journal.
> Guido Heineck <firstname.lastname@example.org> wrote:
> I want to examine the relationship between alcohol and wages. Taking
> into account that smoking and drinking are related I estimate separate
> models for smokers and nonsmokers.
> My question: Assuming that selection into employment and smoking are
> not random, how can I correct for selection bias? I first thought of
> using selmlog/svyselmlog by generating a multinomial selection var
> (empl. smoker; empl. nonsmoker, not empl. smoker; not empl.
> nonsmoker), but these progs require observed values in y for only one
> of these categories/choices.
> Given that I observe wages for employees irrespective of smoking
> behavior I cannot employ selmlog/svyselmlog. I'm thus somewhat stuck
> since the authors of selmlog (Bourguignon et al.) show in their recent
> J. of Economic Surveys article that the widely used procedure proposed
> by Lee may not be appropriate.
> Any insights or ideas for alternatives?
* For searches and help try: