Have an idea?

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

How to interpret the HB output when using categorical covariates

I have a number of nominal covariates that I want to try (individually) in a CBC using HB. I've read the white paper, "Application of Covariates ... " (2009), but it only illustrates using a continuous variable ("ExpecttoPay"). For a two level variable I get two sets of partworths estimates but not sure what each set is estimating. Thanks for any clarification.
asked Dec 21, 2017 by GaryG

1 Answer

0 votes
Indeed, I frequently have to look back at the CBC/HB Help documentation to remember what the multiple sets of utilities mean in the covariates alpha file (especially the first set, associated with the intercept).

In the online help for CBC/HB, there is a section under Estimation Settings + Advanced Settings + Covariates:

"A set of weights (Theta) associated with the intercept of the population estimates of part-worths as well as adjustments to the population part-worth means due to characteristics of the covariates is written to the studyname_alpha.csv file.  For example, if a 3-level categorical covariate were being used, the first columns of the studyname_alpha.csv file would contain estimates for each of the part-worth utilities associated with the intercept of the coded covariates matrix Z (in the case of categorical coding, the intercept would be the population mean part-worth estimates associated with respondents taking on the final level of the categorical covariate).  Then, the next columns would contain of set of regression weights (Theta) for the adjustments to the population estimates for each of the part-worth utilities associated with respondents taking on the 1st level of the categorical covariate, followed by a set of estimates for adjustments to alpha for respondents taking on the 2nd level of the categorical covariate.  The columns are clearly labeled in the .CSV file.  For example, if an estimate for the level "red" for respondents taking on characteristic 2 of Variable2 was equal to 0.75, then this would indicate that respondents with characteristic 2 of Variable2 had a mean part-worth utility for Red that was 0.75 utiles higher than respondents taking on characteristic 3 of Variable2 (since the final column is coded as the omitted, zero, state)."

I hope this is sufficiently clear.  But, if you still have questions, please let me know.
answered Dec 21, 2017 by Bryan Orme Platinum Sawtooth Software, Inc. (198,315 points)