I am doing a choice experiment using a random parameter logit model using the mixlogit command "mixlogit choice price no, rand(a b) group(group) id(id)", how do I calculate the marginal effects?
I tried the following command and the result seems to be incorrect, the RPL coefficient is positive, but the marginal effect calculated with this command is negative.Any help will be much appreciated.
preserve
set seed 12345
gen rnd = runiform()
bysort id groupid (rnd): gen alt=_n
replace a = 0 if alt==1
mixlpred p0, nrep(500)
replace a=1 if alt==1
mixlpred p1, nrep(500)
gen p_diff = p1-p0
sum p_diff if alt==1
restore
If I understand you correctly they should be the appropriate regression coefficients in your fitted models.. best wishes David Booth
Thank you very much Professor Booth! I was trying to get the marginal effect, but the model regression results don't show that one. Thanks again for your help! David Eugene Booth
This should help
https://www.peretaberner.eu/how-to-estimate-and-interpret-marginal-effects-from-logit-model-with-stata/
Hello Zongmiao Xu. -mixlogit- is not an official Stata command, it is a user-contributed package. So it could be that the -margins- command will work (see the link Kelvyn Jones posted for examples). But note that if you have a recent enough version of Stata, there two new official commands you might be able to use instead. You can find more info here:
https://www.stata.com/features/overview/mixed-logit-regression/
HTH.
Thank you very much Kelvyn Jones and Bruce Weaver ! I have tried using the -margins,dydx(*) - command, but the marginal effects calculated are exactly the same as the regression coefficients for the random parameter model. I guess that after the regression with the -mixlogit- command I may not be able to use the -margin- command directly to calculate the marginal effects.
Perhaps you should post your problem to Statalist. It may be that some of the members there will have experience with -mixlogit-.
https://www.statalist.org/forums/forum/general-stata-discussion/general
If you do, please see the FAQ for advice about how to make your question more easily answerable (e.g., use -dataex- to generate a small dataset that illustrates the problem, include your code between code tags, state clearly that -mixlogit- is a user-written package, etc.).
https://www.statalist.org/forums/help
HTH.
If you want to get the population average marginal effects on the probability scale you can simply transform the estimated fixed logit and associated predictions. If however you want the subject specific estimates then you need to take account of the random part of the model. See my (long) answer to this question:
https://www.researchgate.net/post/Where_can_I_find_good_material_on_the_difference_between_mixed_models_and_gee_models
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS