Nested alternatives in Ngene

[alex.mitrani]

Hello

I am using Ngene to design choice exercises with 5 alternatives. Because of the nature of the alternatives, it is very likely that there will be 2 pairs of nested alternatives with the 5th alternative not being in any nest. Please could someone advise on the best way to handle this type of situation in Ngene? I could not find any reference to nested logit in the manual, but have read the paper "Constructing efficient stated choice experiments for estimating nested logit models" by Bliemer, Rose and Hensher that addresses this topic, so I am hoping that Ngene might be able to help.

Best regards

Alex Mitrani

[Michiel Bliemer]

I am afraid that the current release version of Ngene cannot handle the nested logit model. Indeed, in an earlier research version of Ngene we added the nested logit model (restricted to 2 levels). In the next release version of Ngene we plan to include the nested logit model in its general form. However, this will still take some time.

[alex.mitrani]

OK, thanks very much for clarifying the situation.

Alex

[asahoo]

Are there any updates on Ngene's ability to generate designs for the nested logit model? I'm using Version 1.1.1, and its manual does not mention the nested logit model.

I'm assuming that this means that Ngene 1.1.1 does not have these capabilities. Are there any general guidelines one can follow when using Ngene with the intention of estimating a nested logit model? My concern, as pointed out in "Efficient stated choice experiments for estimating nested logit models", is the large difference between the NL and MNL AVC matrices. I would appreciate any pointers.

[Michiel Bliemer]

I am afraid there is no progress yet on the nested logit model, as we have focussed on other additions to Ngene due to a lot of request for certain features, and the nested logit model is not used very often. At this stage, nested logit is on the list for version 2.0 that we are working on right now, which includes also other logit type models, and many more features. We do not expect to release version 2.0 before the end of 2013.

Regarding your concerns, yes, the AVC matrix could look quite a bit different if the scale parameters in the nested logit model would be far away from 1. One way of approximating the nested logit model would be to scale the parameters in according to the scale parameters in each nest. This would only work well if there are no generic parameters across nests.

Please note that a design that is optimised for the MNL model will also work for the nested logit model, it will just not be as efficient. In order to increase the robustness of the design, it is important to add Bayesian parameters, which can also approximate the uncertainty in the scale parameter in different nests. So using Bayesian priors, I would not be afraid at all to use an MNL optimised design for estimating a nested logit model later on. The only thing that you will be most uncertain about, is about the reliability of estimating the scale parameter, as it does not enter the AVC matrix of the MNL model.

Michiel

[asahoo]

Thanks for your prompt response, Michiel. I'll use our pilot data to pursue the Bayesian approach.