2
$\begingroup$

We are new here, and have recently gotten a question that we have very much been struggling to answer. It is concerning a question regarding a conjoint analysis in which we have to incorporate an individual-specific intercept. We are working with the a regression model that includes:

  • 15 Consumers (subjects)
  • 22 Profiles 
  • 1 - 5 attributes
  • amount of Levels varying from 2 levels to 3 levels

We already have all dummies needed to conduct a general conjoint analysis in SPSS. This question however is as follows:

while consumers’ preferences over different attributes and levels are more or less the same, they do have different baseline level of utility (constant terms). As a consequence, to derive more accurate estimation of consumers’ part-worth values, researchers need to make the intercept individual-specific (but the part-worth values are the same across consumers). In your analysis, try to think of a way to incorporate individual-specific intercept in the regression.

We have tried many different approaches but haven't fully figures out how to solve this problem. Can anybody maybe give us some tips? We highly appreciate it!

$\endgroup$
  • $\begingroup$ Conjoint analysis (and the basic one, full-profile conjoint in available SPSS) is by default done separately for each individual. It is ANOVA modeling performed for each of $n$ respondents. You thus get $n$ equations with their utilities and intercepts. You can request also to output omnibus model which combines all respondents. $\endgroup$ – ttnphns Nov 21 '15 at 16:41
  • $\begingroup$ You seem to want to try an in-between approach: omnibus utility terms but individual intercept levels. One would ask why you think this is warranted. Why, really, individual reactions should be identical on one model's parameter and differ on the other parameter? Well, if you have reasons to insist - you may do such analysis, but I doubt much that it is possible via the Conjoint command. That might be possible, I suppose, via mixed models or GEE; right now I can't recommend you a specific way, it has to be overhauled. $\endgroup$ – ttnphns Nov 21 '15 at 16:47
  • 1
    $\begingroup$ Apart from the Hierarchical Bayes logit model, one may also use the mixed logit model under the classical approach. The mlogit package or the gmnl package will do the trick. $\endgroup$ – Fischer Nov 30 '16 at 16:35
0
$\begingroup$

How many scenarios did each respondent see? I typically use hierarchical Bayes to estimate individual random effect discrete choice models, especially if I have a small sample. You can use covariates in the upper-level model (like country, respondent type, etc.) to further pool the data which adds more stability and statistical support to the individual-level random effects. The upper-level models can be used directly as well for even more stability. There are several programs that do this including R that you can call from SPSS. The package is called ChoiceModelR. I would also check out Elea Feit's book called "R for Marketing Research and Analytics" by Springer. A link with code examples and data below.

Hope this helps.

http://r-marketing.r-forge.r-project.org/

$\endgroup$
  • $\begingroup$ Thank you very much for your answer and thanks for the link! We are currently trying to solve our problem with it, we will let you know whether we succeeded! $\endgroup$ – StatsTeamNL Nov 23 '15 at 14:32

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.