ApolloChoiceModelling forum

I am combining several DCE datasets, one of which ranks alternatives. Is it possible to combine MNL and EL specifications in a single model?

Hi

apologies for the slow reply.

Yes, this is fairly easy, just look at the section in the manual on joint models, and e.g. the hybrid choice examples

Stephane

Thanks. The RP/SP example showed how to multiply the likelihood functions from the separate datasets. I got it to work for a data-fusion model combining 5 DCE datasets. However, one of the studies used a two-question format to get a full ranking of 3 alternatives. I set up the data as exploded logit, so there really should be a scale control between the first question and the second question. It was clear how to control for scales among the various datasets, but I'm not sure how to add a scale adjustment within the exploded-logit sub-model.

Hi

scale differences across stages in exploded logit are part of the apollo_el function, please consult the help file or the manual for details

Stephane

Reasonable answer. I looked, but obviously not hard enough. I suspect you have a button you push to generate that text with a blank space for "exploded logit" or corresponding term. I don't know how you find time to respond to all these postings, but it is a huge service to the research community.

There's also an example that does this, see http://apollochoicemodelling.com/files/ ... ating/EL.r

I did see that example. I should have used the term "sequential logit", since the survey did not elicit a ranking as in the Case-3 BWS, but elicited a choice over 3 alternatives, followed by a second question to obtain a choice between the remaining two alternatives. So the data is set up as two choices over 3 and 2 alternatives, respectively. It is possible that obtaining the ranking in this way could result in different scale effects than Case-3 BWS or a direct ranking question.

Hi,

possibly, but that's an empirical question you could only answer if you had data from the other formats too. But what you can do for now is to allow for scale differences across stages

Stephane