I would like to compare Latent Class Conjoint Analysis with other Conjoint-approaches using artificial Monte-Carlo Data generated with SAS. Is it possible to use MPlus for this purpose? Are there any problems with SAS data?
In this version of Mplus, the Monte Carlo facility cannot read data generated from other programs. This will be available in Version 3. In the meantime, you can use the Monte Carlo Utility which you can find on our homepage. This utility will read externally generated data, analyze them, and save the results from each replication.
Thank you very much for your quick response. Unfortunetly, I don't know Mplus too well. I think about using Mplus or Sawtooth instead. Do I have to have a good programming knowledge to implement a (Choice-Based) Latent Class Conjoint Analysis with Mplus. Is there anything about Latent Class Conjoint in the documentation? Thank you! Thomas
bmuthen posted on Wednesday, September 03, 2003 - 5:02 pm
Can you briefly describe what the model is for the choice-based latent class conjoint analysis?
With conjoint analysis I mean the alternatives (products) are defined entirely, i.e. via a cerntain number of dimensions. The respondents choose one or none product of a given set of alternatives (choice based). Several alternative-sets are presented to each respondent. For generating the choice designs for the other Conjoint-variants I have used SAS. The model itself: 1)The number of clusters and the clusters themselfs must be set (to intial values). 2)The logit model is used to see to which class each respondent fits best. 3)A logit solution is estimated for each group. A log-likelihood is computed given the data. 4)Steps 2 through 3 are repeated. Then log-likelihood is computed, assuming that the respondent belongs to the first class.The individuals likelihood for each group is weighted with size of the group. By summing the weighted likelihood over the different classes the individual likelihood is obtained. Then the logs of these likelihoods are cumulated over individuals. For more (non-technical) details: www.skim.nl/software/Papers/cbclctech.pdf
bmuthen posted on Friday, September 05, 2003 - 4:59 pm
I looked at the Sawtooth document you pointed to and it lists conjoint analysis under the heading Latent Class analysis. Since Mplus does LCA in very general forms, I assume conjoint analysis can be done in Mplus, but I would have to read the marketing literature indicated to be sure. The LCA features discussed in the document all seem to be within the Mplus framework. Perhaps there are Mplus Discussion readers who can chime in here. For an overview of Mplus latent class modeling capabilities, see "paper #86" on the Mplus home page at www.statmodel.com, "Latent variable mixture modeling."
Conjoint analysis is main-effects only modelling of not completely crossed stimuli (by orthogonal design) for a series of subjects. This would require doing at least type = general meanstructure in combination with latent classes. The model would then generate parameters for each of the levels of each main effect, and do one seperately for each latent class. I'm interested in developing this but I'm not sure about the steps to follow to input the orthogonal matrix - although I've done some learning on that part of MANOVA I've not rehearsed it for decades!
bmuthen posted on Saturday, October 04, 2003 - 2:08 pm
If the response variables are continuous, the orthogonal matrix would be entered in Lambda or Beta to use language of the Mplus User's Guide Technical Appendix. In Mplus terms, Lambda is specified using BY statements and Beta using ON statements. Not knowing conjoint modeling, I can't say more than that. I do however have books on it and one day I hope to get time to read them.