Logit Models for Sets of Ranked Items

Methods are presented for analyzing data generated by asking respondents to rank a set of items. Based on a conditional logit model, these methods allow us to estimate and test for differences among items in respondents' preferences for them; to test for differences in item preferences across subpopulations; and to incorporate predictor variables describing respondents, items, or both. The models can be easily estimated with programs for proportional hazards models, and they can be generalized to allow, for ties in the rankings. Detailed examples are given.