ApolloChoiceModelling forum

Hi two changes are needed in your code. As apollo_combineModels is called inside the latent class model, you need P[[paste0("Class_",s)]] = apollo_combineModels(P_temp, apollo_inputs, functionality,components=c(paste0("RP_class_",s),paste0("SP_class_",s)),asList=FALSE) ...

Can you share your code and data with me outside the forum, and I'll try to debug it for you

you can do, if you really think a Normal makes sense (it normally does not)

Hi

yes, I understood you were doing the factor analysis in another package. But you would never use factor scores in an ICLV model. What would do is check what groupings the factor analysis gives you and implement the ICLV model accordingly

Stephane

Hi

apologies for the slow reply, your message had got lost

Did you try with lambda=1 as a starting value. Essentially, you should first see if you can replicate MNL

Stephane

Anat

apologies for the very slow reply.

I would use a lognormal distribution indeed. It's not about whether you have an opt-out or not, really, but about whether it would make sense behaviourally for people to have a diferently signed coefficient

Stephane

Hi many apologies for the slow reply, your messages had been lost. First, "relative function convergence" is the BGW terminology for convergence, so that's fine. Second, in relation to your ASCs, you have: b_ASC*(asc_alt1==0) + b_ASC*(asc_alt1==1) which would mean the ASC is used for both ...

Hi

apologies for the slow reply

Your problem is caused by the effects coding you have implemented in the data. If you have only three levels, then you should also only estimated two parameters, not three. But I would suggest you just use dummy coding

Stephane

Hi many apologies for the slow reply, we had missed your post In relation to your questions: 1. Is classical elasticity analysis suitable for unlabeled cases? (While computations may be possible, I am interested in ascertaining the theoretical relevance of confirming sensitivity to specific attribut...

Hi apologies for the slow reply So what you're finding is that you MMNL overfits the estimation data a bit more than MNL, but these differences are small. No, Apollo does not compute the % correct predicted metric as this is a misleading metric that is contrary to probabilistic choice models. It sho...