Have an idea?

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

How to best code categorical variables with missing values for use as covariates in CBC-HB

Dear Sawtooth users,

I have a doubt concerning the codification of the categorical variable "income" for usage as covariate in CBC-HB.
My income variable currently takes on values from 1 to 9, where 1 corresponds to the lowest income class, 8 to the highest and 9 characterizes those respondents that prefer not to answer the income question.

For the purpose of using this demographic variable as covariate in the CBC-HB analysis, I was wondering whether I should assign another class to those respondents scoring 9, delete those respondents, or leave the coding as it is.

Thank you!
asked Jul 14, 2017 by lotika (455 points)

1 Answer

+1 vote
Good question.

First, I would recommend you collapse your coding into fewer categorical categories for the use of categorical covariates.  8 is too many in my book and also experts I've consulted with on this have recommended very few categories for categorical covariates in the upper-level model of HB (such as just 2 or 3).  

If you have a "refused/Don't Know" category and if these people make up very few of the respondents (such as 15% or less), I'd do an imputation for them and assign them to the most likely income category.  So, if you recoded to three levels of income and the most likely income assignment was level 1, then I'd assign these folks to level 1.

If there are other variables that exist in your study that are predictive of income level, then I'd use those other variables to predict the most likely income category for your missing respondents.
answered Jul 14, 2017 by Bryan Orme Platinum Sawtooth Software, Inc. (198,315 points)