Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

From |
Klaus Pforr <kpforr@googlemail.com> |

To |
statalist@hsphsun2.harvard.edu |

Subject |
Re: st: clogit for discrete choice experiment with multiple choice sets |

Date |
Mon, 30 Jan 2012 10:56:44 +0100 |

<> Dear Hadji,

best Klaus Am 30.01.2012 10:36, schrieb Klaus Pforr:

<> Dear Hadji,this seems to be an application for multilevel or panel multinomiallogit. There is a fixed effects model by Chamberlain (1980). The fixedeffects are in your case on the person level. Possible random effectssolutions are discussed in Train (2009). The first model has not beenimplemented yet (cf. Allison 2009, p.44), but I'm am currently workingon an ado for this model(http://www.stata.com/meeting/germany11/desug11_pforr.pdf). The lattermodels are complicated and can be estimated with GLM.There is a also back door solution for the fixed effects estimator forsmall samples and short panel/small clusters (in your case the thenumber of experiments). Börsch-Supan applied the Chamberlain model onhousing choices and rearranged the data in a way so that he could usethe implemented clogit to estimate the model. The data organisation isthe following: In a simplified version of your case you would haveonly 3 experiments (or panel time points in the chamberlain lingo) and3 alternatives.Lets say you have the indiv 1 with this selection (this is example ispurposely simple)xp choice 1 1 2 2 3 3When you look up the equation in the chamberlain model, you find theconditional likelihood of the prob to choose the time series that waschosen conditional ("i.e. divided by") the prob of all permutations ofthe chosen alternatives.You look at all combination of choices, which have the same number of1's, 2's and 3's (or in general all of your outcomes) for the specificindividual. This set of permutation makes your set of alternatives:Permutaion Was it chosen? 123 yes 132 no 213 no 231 no 312 no 321 noAfter this reorganisation you run a clogit on the data with respondentas group, and have the multinomial logit with fixed effects. This isvery cumbersome even your simple application, but it works. You alsohave to think about how to generate you independet variable for thisto get the coefficents that you want.Here is the literature:Börsch-Supan, Axel. 1987. Econometric analysis of discrete choice:With applications on the demand for housing in the U.S. andWest-Germany. Berlin et al.: Springer Verlag.Börsch-Supan, Axel. 1990. Panel data analysis of housing choices.Regional science and urban economics 20: 65–82.Börsch-Supan, Axel, und Henry O. Pollakowski. 1990. Estimating housingconsumption adjustments from panel data. Journal of urban economics27: 131–150.Chamberlain, Gary. 1980. Analysis of Covariance with Qualitative Data.Review of Economic Studies 57: 225–238.Train, Kenneth E. 2009. Discrete choice methods with simulation. 2.ed. Cambridge, MA et al.: Cambridge University Press.I hope this helps best Klaus Am 28.01.2012 08:32, schrieb Hadji Cortez Jalotjot:Hi! Iimplemented a discrete choice experiment to model vehicle choice. Inmy questionnaire, I presented each respondents with 10 choiceexperimentsor choice sets with each choice set having 3 alternatives or choices.The explanatory variables are the characteristics of the vehicles.With this, I am fitting a conditional logit model.In my data set, dummy variables were used to represent the explanatoryvariables. Since each choice experiment has 3 alternative options,each choice experiment corresponds to 3 rows of observations. So 10choiceexperiments per respondent X 3 alternative options per choiceexperiments = 30 rows of observations per respondent. (sample databelow shows only 3 choice experiments withsome of the explanatory variables for respondent 1)respno choice_set choice var1a var1b var1c .. ... none1 1 1 10 0 01 1 0 00 1 01 1 0 0 00 11 2 0 01 0 01 2 1 10 0 01 2 0 00 0 11 3 0 00 1 01 3 0 10 0 01 3 1 00 0 1For clogit to work, I must select a variable that will identify thegrouping for which the software will run the analysis.Now, for this kind of data in which respondents answered multiplechoice sets (10 in my case), which should I used as a group?Is it the respno or choice_set? I am confused because if I use respno, Stata says multiple positve outcomes in a group. And the predicted probabilities is computed for the whole 30 alternative options and not only for the 3 alternative options per choice set. But if I use the choice_set as the grouping and I extend the model to include respondent characteristics (e.g. income), I may have problem with fixed effects because for example choice_set 1 and choice_set 2 is from the same respondent and therefore will have exactly the same income. Any advice is appreciated. Hadji * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/

-- __________________________________ Klaus Pforr MZES AB - A Universität Mannheim D - 68131 Mannheim Tel: +49-621-181 2797 fax: +49-621-181 2803 URL: http://www.mzes.uni-mannheim.de Besucheranschrift: A5, Raum A309 __________________________________ * * For searches and help try: * http://www.stata.com/help.cgi?search * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/

**Follow-Ups**:**Re: st: clogit for discrete choice experiment with multiple choice sets***From:*Hadji Cortez Jalotjot <jalotjot@yahoo.com>

**Re: st: clogit for discrete choice experiment with multiple choice sets***From:*Maarten Buis <maartenlbuis@gmail.com>

**References**:**st: clogit for discrete choice experiment with multiple choice sets***From:*Hadji Cortez Jalotjot <jalotjot@yahoo.com>

**Re: st: clogit for discrete choice experiment with multiple choice sets***From:*Klaus Pforr <kpforr@googlemail.com>

- Prev by Date:
**Re: st: clogit for discrete choice experiment with multiple choice sets** - Next by Date:
**Re: st: clogit for discrete choice experiment with multiple choice sets** - Previous by thread:
**Re: st: clogit for discrete choice experiment with multiple choice sets** - Next by thread:
**Re: st: clogit for discrete choice experiment with multiple choice sets** - Index(es):