Thread Rating:
  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
How do you calculate utility scores in conjoint analysis
We had a masters student working on a survey for us that offered respondents a series of choices between A and B retirement plans in the form of a conjoint analysis. We have the data back and now we don't know how to do the analysis. How do you get to utility charts from the choices?

- Jan
Conjoint analysis depends to a large extent on getting the design right in the first place so that the attributes and levels shown can be analysed statistically. And secondly the choice of type of analysis depends what type or flavour of conjoint you're doing.

If your design is OK, you'll need to code it into dummy variables for analysis - see here for an illustrative example:

Then different designs use different analysis strategies. For simple ranking or rating type conjoints then, in simple terms, linear regression (OLS) can be used performing a regression for the dummy variables against the rank or rating. With choices (and ranking as an alternative), you can use Logistical Regression for categories, or go more advanced with Hierarchical Bayes techniques.

If you'd like to send me your design and data I can provide some advice on what's most appropriate and check your design looks sensible.


Saul this is an area of my interest also. I've made some basic research on the subject also, Wikipedia and this article: I am planning to use conjoint also. If I have 5 types of parameters, with 5 choices (categories), where's the line between using Logistical Regression and Bayes Tecniques. Can I use the same data to make the analysis and will the results differ from each other. Have in mind that my experience is only in simple linear regression models.

[Mod: Careful on adding links please - you're trying to promote an agency without saying you're working for them]
Hierarchical Bayes can be seen as an extension of Logistic Regression - both are used for categorical decisions (ie choices instead of ratings). In the original form of choice-based conjoint, data is analysed at a sample-level, so bringing all the data from all respondents into one pot for analysis, which would then use logistic regression. For marketing purposes this has the problem that it treats the dataset as homogeneous and so potentially masking different subgroups in the data. A variety of methods were used to try to identify different groups (eg latent class) from choice-based data and in current times, HB is the now preferred method to get back to individual level utilities from choice-based data.

Thanks for this wonderful information.

Forum Jump:

Users browsing this thread: 1 Guest(s)