Have an idea?

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

Tradional Conjoint Analysis: Can I use a t-test to compare the estimated part-Worths between Groups?

Hello Sawtooth Community,

I did a traditional Conjoint value Analysis (CVA, full-profile method, cards had to be ranked) and  I just estimated the part-worth utilities for the different attributes (organic label) and their levels (EU organic label, organic store label, etc.).

Now I want to test if the subjects which where interviewed in France, Netherlands, and Germany have different parth-worth utilities for the EU organic label. What is the best way of doing this? Can I perform a t-test or it is better to do a ANOVA?

Is it possible that I estimate the subjects┬┤ willingess to pay?  My problem is that I incorporated price as an additional attribute in order to provide WTP estimates. Now I just read the paper by Breidert, Hahsler and Reutterer (2006) and they say that the WTP cannot be calculated when using a traditional conjoint approach. Is there a way arround this problem, so that I can calculate the subjects WTP in the different countries?

Thanks very much for your help.

Best regards from Germany
asked Feb 13, 2014 by anonymous

1 Answer

+1 vote
Hello.  These are good questions.

You can do stat testing on the utilities but I would use a transformation of the utilities to that respondent reliability doesn't confound your test.  A transformation Sawtooth Software recommends is the Zero-Centered Diffs (ZCD) transformation which makes respondents' utilities more comparable by reducing the differences in the scale of the utilities between them (respondents who answer more reliably have better-fitting models with larger utilities than do unreliable respondents).  

Next, if you have 3 groups, ANOVA is the appropriate test.  T-test is specifically a 2-group test and will give biased answers when you have 3 groups (i.e. it will find significance more often than it is really there).  

Finally, WTP is a problematic concept, suspect both computationally and from the standpoint of economic theory.  While I it may be more problematic using traditional ratings-based conjoint , using choice-based conjoint doesn't solve the problem.  I am not aware of a fix for either ratings-based or CBC but some papers in our upcoming 2013 conference proceedings may be discussing this topic in more detail.
answered Feb 13, 2014 by Keith Chrzan Platinum Sawtooth Software, Inc. (50,675 points)