I am setting up a MaxDiff survey for feature prioritization. The market I work with is by its nature difficult to recruit from and I have to plan on having smaller-than-ideal samples for any study I run. I know the answer to these kinds of problems is often "Just run interviews instead," but in my case I and my stakeholders would like the scale of 50+ survey responses (100 would be amazing) and the depth of running a handful of interviews.
I'm aware that typical guidance is to get 300 respondents for a MaxDiff to be reliable for hierarchical Bayesian modeling, but that is simply infeasible in my situation. I am considering using a Bradley-Terry model and/or Elo ratings to analyze the data resulting from the MaxDiff exercise, as I've read they (might) have lower sampling requirements.
My questions for you all:
Does anyone have insights into whether this is a sound approach (both in terms of methodology and implied power analysis/sampling requirements)?
Has anyone taken a similar approach? If so, was it successful?
Thanks as always to this community. I always learn something new here.