choice-metrics.com

Hi,

I am hoping to perform a generalized multinomial logit (G-MNL) after I collect my data and am trying to figure out which model to use in the ;eff option.

I don't think the pure 'mnl' option would be appropriate but I'm trying to decide between 'rp' and 'rpec'--in short, the issue is whether the error should be specified or not.

I had read that the GMNL is an extension of the MMNL, but I wasn't sure if it extends it by creating an additional error component.

Thanks for any help!

The GMNL model is indeed an extension of the MMNL model. Note that data collected in choice experiments requires the panel version of the MMNL model, so that would be rppanel or rpecpanel. However, optimising for panel MMNL models is extremely difficult due to the large computation times such that in practice optimising for these models is often not possible. I always recommend everyone to optimise the design for the MNL model. The designs optimised for MNL are also largely efficient for estimating MMNL models, and therefore will also be fine for estimating GMNL models. The main thing to take into account is that the number of parameters in a GMNL model is much larger than in an MNL model, so you will need to increase the variation in your data by increasing the number of rows in the design.

Michiel

Many thanks Dr. Bliemer!