Have an idea?

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

Are covariates an acceptable workaround for different treatment groups?


I have a research design with three treatment groups (10, 20, and 30 exposures to a certain phenomenon) plus one control group (no exposure to the phenomenon). In order to measure the preferences of the members of each group and compare them for the different treatments, I plan to conduct a CBC HB analysis. I know that HB assumes one underlying population-wide distribution and only "adjusts" this distribution to get individual-level estimates. A clean study design would therefore include three CBCs (four including the control group), that look alike but are estimated separately. This in turn requires a high number of respondents because I have to run multiple analyses.

My problem is that it will get very hard to get enough respondents for the specific matter that I am interested in. My idea was to conduct only one analysis including all four groups at once, save 75% of respondents and enter the number of exposures to the phenomenon (my treatment groups) as a metric covariate. As far as I know, this will produce estimates that do not come from one underlying distribution but from several, which would be pretty much the same as having different treatment groups. Is this an acceptable workaround or is there a flaw in this argumentation?  

Thanks a lot!
asked Mar 13, 2017 by Ulf

1 Answer

0 votes
Hello, Ulf.

This seems reasonable to me - I use covariates in this way, too.  You are allowing (but not forcing) your treatment groups to be different.
answered Mar 13, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (53,875 points)