I am interested in clogit, not multinomial logit, and the overall effect of the predictor on the choice of n alternatives that can only be identified by their attributes (i.e., the choice sets are rarely repeated across cases).

In choice sets with only two alternatives, the effect of a binary predictor such as sex, can be estimated from cases in which there is variation across alternatives (e.g. one is female, the other male). I understand that cases in which there is no variation will not contribute to the estimation of that coefficient. What happens when there are three alternatives in the same choice set, and two of them adopt the same value for that binary predictor, and one is different? The same question applies to nominal predictors with >2 categories.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.