Efficient designs for ranking

[miq]

I am considering using ranking instead of simple best-choice for each choice set to increase the amount of information from each respondent. Each respondent would pick the best and the worse alternative, for more than 3 alternatives in a choice set this can be done consecutively on the remaining alternatives to get full ranking of all the alternatives in the original choice set.

I was wondering if anyone has any ideas on how to make Ngene generate a more efficient design for this. As far as I understand, currently it is efficient for a best-choice situation.

Thank you in advance for any ideas / suggestions.

[johnr]

Hi Miq

Currently we only have capabilities for generating first pick type designs. We have recently developed the theory for generating rank exploded designs and are no looking at programming and testing. It involves having to calculate the Fisher matricies for all possible rank explosions and then probability wieghting these by the number of respondents who are likely to choose a particilar rank structure. For MNL models, this should be fairly easy to implement. For the panel MMNL model however, I am not sure this is going to be feasible. Unfortunately, this not available in Ngene 1.0.

John

[miq]

Thanks for the answer John. I'll bee on the lookout for the new add-ons to Ngene then. Meanwhile I guess I'll go for a Bayesian efficient design for a first-pick type of experiment with priors from MNL model based on a pilot that we've just done. It's still better than nothing.