I’ll explain the cumulative logit model in three different ways. I begin with an intuitive, nonmathematical description. Then I present a formal definition of the model in terms of cumulative probabilities. Finally, I explain how the model can be derived from a latent variable model.

Here’s the nonmathematical explanation. Suppose you wanted to analyze the wallet data but all you had was a binary logit program. One solution to that dilemma is to group two of the three categories together so that you have a dichotomy. To get as even a split as possible on the dependent variable, the best grouping is to combine categories 1 and 2 together and leave 3 by itself. That way, there are 74 cases in the (1, 2) ...

Start Free Trial

No credit card required