Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Using the HB covariate results

I analysed a maxdiff survey using HB and used a bunch of covariates and followed the (excellent) technical paper on covariates in HB. Now I know which covariates have significant impacts on different parameters. I also know whether the effect is positive or negative. But what I would like to know is the magnitude of their effect.
I have zero-centred, transformed and rescaled my HB raw utilities. Would it be correct to do the same for the covariates so that I can compare them? By this I mean add the covariate effect to the average raw score and then follow the same zero-centre-transform-rescale routine?
asked Nov 14, 2013 by anonymous

1 Answer

+2 votes
As long as you recognize that you're dealing with average population utilities, I don't see why you couldn't do this.  

However, a more practical approach would be just to use the covariates as respondent filters and use these as banner points in Sawtooth Software's SMRT or Online Simulators.  Our simulators will normalize the utilities (at the individual level) to zero-centered diffs to allow more correct comparisons across groups (controlling for scale factor).  The alpha vectors found in the .alpha files are population estimates of what should also be reflected quite nicely in the individual-level posterior utility estimates (the final utilities stored in the individual-level utility file).  So, just use our simulators to tabulate and summarize the results by different segments, where the segments are the same variables you applied as covariates in HB estimation.
answered Nov 14, 2013 by Bryan Orme Platinum Sawtooth Software, Inc. (198,815 points)