Hi all,
I have a statistical question as well its application in Sawtooth. I have Best-Worst choice data, where people indicated a best and a worst option out of three alternatives. I estimated part-worth using Sawtooth HB CBC module for Best-Worst data. I guess the uniform prior is fine. A colleague is now concerned about differences in error variances and argue that me results highly depend on the error variance and that I can't compare different results (even within subject) as long as I do not control for it.
A friend suggested to use standardized variables (dependent and predictor variables). But I do have binary variables (dummy coded) only and the dependent is coded as 1, 0, or -1 (1: best, -1: worst). For me it doesn't make sense to standardize them and the estimation might not work if I have other than 1, 0, or -1 as input. Another way, suggested by a friend as well, is to to put a restriction such that variances in the var-cov matrix are identical (which is a strong assumption by the way). But I don't know how to do this.
Does anybody know how to deal with this issue? Every help is highly appreciated.
Best regards,
Michael