ApolloChoiceModelling forum

apologies for the slow reply. This is very code specific, can you provide your code and data outside the forum and we can investigate

Hi

apologies for the slow reply

yes, this difference looks a bit too large to explain. Did you also look at the medians to see if they are reasonable (though we would want the means of course).

You might want to consider other distributions too with shorter tails

Stephane

Hi

apologies for the slow reply.

Can you be more specific? Did you wait until the process has finished?

Stephane

Hi apologies for the slow reply. Using your numbering: 1) Mixed logit for random taste heterogeneity is fine. For correlation, you would be better with a nested logit or CNL kernel inside your mixture. 2) yes 3) You should not use Halton draws in that case. Try MLHS or Sobol. In relation to the sing...

Hi yes, you're finding correlation. If your model estimates fine and gives you a covariance matrix, then you likely do not face an identification issue. in relation to correlation between the alternatives, I would recommend that you consider a GEV kernel such as Nested Logit instead of MNL inside yo...

Hi

apologies for the slow reply.

You wouldn't use factor loadings inside an ICLV model. Or what did you have in mind?

What you might want to do is to use factor analysis to determine the groupings of indicators for your latent variables

Stephane

Hi

apologies for the slow reply. There is nothing immediately apparent that would be causing an overspecification. Can you tell us more about the data? Maybe there are strange correlations in the data that lead to an empirical identification issue?

Stephane

Hi

can you share your code and data with me outside the forum and I'll investigate

Stephane

Hi first, in relation to your SP model, you could try to refine the specification and see if that helps. Feel free to show us your current results on the forum as we may then have more suggestions. second, in relation to joint "RP"-SP, your datasets look quite different, but you could of c...

Annat

can you show us the model estimates too? It might be that you need to take the absolute values of the sigma parameters as the estimates might have been negative

Also, you're missing the minus sign in front of the exponential

Stephane