How do I design and analyse an Adaptive Max Diff experiment?

Question

Hi - I am trying to determine how to design an adaptive MaxDiff experiment and analyse it. I read through your white paper related to the AMDiff design process.
a) Since items are randomly selected on Stage II (other than the best and worst selected on Stage I); is it possible that across all sets on Stage II we dont end up showing a particular item at all (for e.g. if the random selection of the Set 1 item say ends up being the same item on 2 sets in Stage 3?).
b) Do I need to have a HB software to run the analysis for this or can I do it with a Latent Class software? Do you have any detailed documentation on how I would analyse this data?
Any help is greatly appreciated.
Thank you.

Answer 1

The way I've described the process, all items are shown once in Stage 1.  Then, in stage 2, losing items are not shown in stage 2.  So, yes, in stage 2 not all items are shown.  But, each item will appear at least once, because all items are present in Stage 1.

You can estimate the models in Latent Class as well.  Each best-worst task is coded as two tasks.  In the first task, the dummy codes are all 1s and 0s.  But, in the worst task those codes are all multiplied by -1.

You should use "User-Specified" Coding as described in the Latent Class manual.  And, you'll need to build your data file by yourself to manage the proper coding.