Marcel Gerds
2011-Jan-27 21:14 UTC
[R] Modeling Binary x Binary Interactions with mlogit (and interpretation)
Dear R community, I am using the mlogit package to analyze discrete choice data. Apart from a main effects model, I want to estimate interactions between the attributes of the choice set (e.g. the existence of a certain attribute) and some subject-specific data (like gender or income). Studying the mlogit documentation, I found no hint on how to do it. In the literature there is only the case discussed how alternative-specific variables can be combined. In my case, the alternatives are of no interest, meaning I am using a purely generic model. So far, I tried to model these interactions by simple multiplying the variables. Example: mlogit.model <- mlogit(CHOICE ~ ATR1+ATR2+ATR3 + ATR1*GENDER + ATR1*GENDER + ATR1*GENDER| -1, data=data_ml) Here, gender is subject-specific. I get results like the following: --- Coefficients : Estimate ATR1_yes 0.779116 ATR2_ yes 2.257905 ATR3_ yes 1.141625 GENDERfem -14.026649 ATR1_yes :GENDERfem 0.094709 ATR2_ yes:GENDERfem -0.076223 ATR3_ yes:GENDERfem 0.117373 --- I present only the coefficients here. However, when I change the reference level to male, the coefficients of the interactions effects just change sign. I have two questions in this regard: 1.) Is the modeling of such interactions effect feasible in the mlogit setting? 2.) I have some problem understanding the changing sign of the coefficients when I change the reference level. This would imply that females always prefer the opposite of males. Clearly, this cannot be. I imagine that I am misinterpreting this issue and I would be grateful for any help on this. Best regards, Marcel -- Marcel Gerds, M.Sc. University of California Department of Agricultural and Resource Economics 233 Giannini Hall Berkeley, CA 94720 Tel.: +1 510-643-2202 Mobil: +49 176 21302825 E-Mail: marcel.gerds at berkeley.edu web: www.marcel-gerds.de