Have an idea?

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

Different result when using CBC/HB and SMRT for subgroup


I estimate my CBC survey using HB estimation in SSI Web by divided my samples into two subgroups. However, when I use SMRT to estimate the utilities by putting my subgroup categories in banner I have got a different result. The preference scores change quite a lot and I am not sure which one I should use.

The next question is, by using SSI, I can see the std. deviation and be able to calculate the t-statistics to see whether which levels are significance. However, in SMRT because there is no statistical significance reported there, I don't know how to draw a conclusion from the result. Is there any way I can tell whether the utilities and preference scores are significant or not?

Thank you very much,
asked May 20, 2014 by anonymous

1 Answer

0 votes
SMRT only reads the HB utilities estimated within the SSI Web platform (using CBC's HB utility estimation routine), rescales them, and displays them on the screen.  SSI Web does not "estimate" the utilities.  But, the rescaling of the utilities is probably what is throwing you off.

When SMRT uses utilities to estimate shares of preference in the market simulator, behind the scenes it is using the raw utilities to do the Randomized First Choice or Share of Preference math.  But, to normalize the utilities to allow you better to make comparisons between groups of respondents, it uses a transform (zero-centered diffs) to expand the utilities to a larger (and common) scale across respondents.  Those are the utilities it is showing on the screen.

Zero-centered diffs rescaling comes up with a multiplier for each respondent such that the average difference between best and worst levels within attributes per respondent is 100 points.

SMRT will report the standard error of the shares of preference it reports.  But, it doesn't report the standard errors (or standard deviations) of the utilities themselves (the zero-centered diffs transformed utilities).  However, you can export the zero-centered diffs utilities out of SMRT for quick computation of those standard errors or standard deviations using an outside software package (such as Excel or SPSS, for example) by (within SMRT) clicking Analysis + Run Manager + Export... and then choosing to export the Zero-centered Diffs for your HB utility run.
answered May 20, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (198,315 points)