I'm midway through analysis of a repeat measures conjoint. To look at the stability of the logit coefficients I converted the t Ratio to a p value using Excels tdist function (=tdist(abs(tvalue),df,2)) for the p value for a two tailed test. This allowed a first view of the difference from zero of each coefficient.
To test the differences between the logit coefficients, to check for significant differences, I used the coefficients and standard errors to bootstrap the mean differences in the standard errors between the two waves to calculate the t value (based on this article*) for the t-test (mean difference in means/ mean difference in standard errors). I'm assuming that the correct df is again the sample size.
I've not been able to find guidance as to whether or not this is a legitimate test to perform on logit differences so would appreciate any views.
Can this be extended to the output from the HB? At this point I'm layering inferential statistical test on baysian statistical model so think it is probably a step too far to think there is a rigorous statistical test. Is there an accepted method of analysing differences between conjoint waves?
This of course is a stricter version of 'matched sample' https://www.sawtoothsoftware.com/forum/2293/stability-of-cbc-hb-estimates-for-matched-sample. I can safely say that there appears to have been big shifts in attitudes between the two waves (there was a big life event for everyone involved that could cause this shift). I analysed the waves separately with no reference to each other. The client is interested in the differences between the waves.
*http://my.ilstu.edu/~wjschne/138/Psychology138Lab19.html