Dear all,

I am running the following cmclogit (conditional logit / McFadden logit) model (and other models with the same structure) in Stata 16.0:

cmclogit VoteChoice_ i.ident_ envProx_ i.envComp_ c.envCompXCS2simR_, casevars(c.CS2sim_env i.interest_fd age i.gender_f i.income_f i.education_f i.religion_f i.urban_f) altwise basealternative(GPS) vce(cluster id)

I am interested in the interaction between the competence variable (" i.envComp_ " | dummy 0 & 1 | alternative-specific | indicating whether a survey respondent thinks a party is competent in the area of environmental policy) and the issue similarity variable (" c.CS2sim_env " | continuous variable from 0 to 1 | individual-specific | indicating whether the parties considered by a survey respondent offer very similar (or very different) policy options). Unfortunately, the typical coding of the interaction with the " # " did not work as the maximum likelihood function could not deal with the "not concave" areas. To fix this issue, I first created a new "product or interaction term variable" by multiplying the just described direct effects with each other and including this newly created variable in the cmclogit model. This new "interaction term" variable is called " c.envCompXCS2simR_ " (continuous | alternative-specific).

I now want to graphically illustrate the interaction effect via marginal effects to interpret the significant interaction. The problem is, however, that all of the following variations of the same command do not lead to any results:

margins, dydx(*) vce(unconditional)
margins, dydx(i.envComp_ c.CS2sim_env c.envCompXCS2simR_) vce(unconditional)
margins, dydx(i.envComp_ c.CS2sim_env c.envCompXCS2simR_) , predict(outcome(1(1)6)) vce(unconditional)
margins, dydx(i.envComp_ c.CS2sim_env c.envCompXCS2simR_) vce(unconditional) , predict(outcome(1(1)6))

First, I simply tried to calculate all marginal effects so that I have all options to plot afterwards with marginsplot (first line). When I run this line, however, the programme does not stop estimating/calculating. Also after letting it run for many hours (over night), it did not produce any results and I had to stop it with the red break button in the morning. Second, I reduced the model to the three variables of interest (second line) but it did not help (and I assume that my computer is not underpowered, the stacked data contains 28'000 observations). Third, I wondered whether Stata needs to know for which (of the total 7) choice alternatives (including one baseline party alternative) it should calculate the marginal effects, and therefore added the predict command in the way it's done for other margin predictions (third line). Then I receive an error message that the predict command is not valid. Also changing the order of the predict and vce part did not help (line 4).

I have now spent many hours reading all relevant Stata manuals and trying different codings, but it all did not work. Accordingly, I would be extremely grateful if you could tell me how I can obtain a graphic illustration of the interaction in the specified model above, respectively an illustration of how the marginal effect of the competence assessment changes depending on the issue similarity to other parties. This illustration is crucial to my research project and I don't have much time left. A circumstance that might further complicate obtaining this illustration is the fact that the interaction term is introduced as a third variable (and not with the usual " # ").

Thank you very much!! Best,
Michael