ApolloChoiceModelling forum

Essentially, you started the estimation at a degenerate solution, where two classes had exactly the same parameters, and you fixed the weight for the two classes to be the same. once you fix that issue, things work. The reason for the order of starting values messing things up in this case is that t...

This is likely due to numerical issues. Have you tried with a simpler model specification first, and using a subset of the data?

Hi

again, you would find answers to this in standard textbooks. In choice modelling, our emphasis is much more on relative fit by comparing different models on the same data, rather than on fit per se. Base models often have a low rho2

Stephane

Hi

you would find answers to these queries in standard textbooks on choice modelling. I would use the robust covariance matrix. 0.6 is not an issue - inspecting the covariance matrix is useful to see if there are any nearly perfectly correlated parameters, but 0.6 is way off that

Stephane

Hi

your specification is correct but your results are worrying as the effect in the reference group is positive and as the income effect is not monotonic

Stephane

Okay, in that case, Normals might be okay. In terms of computing error measures, you can have a look at this paper (DOI: 10.1016/j.trb.2013.09.010) and think about how to set this up in Apollo, but it is not precoded

What are your attributes?

yes, having all the starting values at zero also led to an issue with the initial check of the gradients. It's always good practice to use different starting values in different classes

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