Have an idea?

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

Comparison between groups using MANOVA

We've got a 2x2 between-subjects-design and would like to test a significant difference between four groups.

We used relative importances and analysed a chi2-test. The difference between the groups is not significant.

Furthermore, we used zero-centered-diffs (from SSI web) for MANOVA (independent variables: part worth utilities, dependent variables: group) to test the difference between the groups.

My question is: should zero-centered-diffs from every seperate group used for analysis?
Otherwise data sets from the four groups could be brought together, analysed via sawtooth and those zero-centered-diffs could be used for MANOVA.

Which pf the procedures is better to minimize a potential error?
asked Nov 4, 2014 by anonymous

1 Answer

+1 vote
I'm a little unsure about your question, but I THINK you're asking whether you should run your HB analysis on all four groups of respondents combined or run HB separately for each of the four groups and then combine the utilities afterwards.  I might suggest a compromise between the two extremes you describe and run the four groups together in a single analysis but use a 4-level categorical variable indicating group membership as a covariate in your HB analysis.  The covariate allows (but does not force) the four groups to have utilities that come from different distributions, so it seems like a happy medium sort of approach.    

If instead your question was about using MANOVA to reduce Type 1 error, then I think MANOVA is a good first step that will need to be followed by additional steps to prevent multiple comparison error.
answered Nov 4, 2014 by Keith Chrzan Platinum Sawtooth Software, Inc. (95,675 points)
Perfect, thanks for this answer!
Pooled VS split sample VS covariates