3.4. Consider the 2 x 3 x 2 contingency table relating a binary x\ to an ordinal Y, by row, of (10, 10, 10; 0, 0, 30) at the first level of x2 and (10, 10, 10;

10, 10, 10) at the second level of x2. Explain why a cumulative logit model with main effects has finite estimates for the effects of x\ and x2 but a model that also has an interaction term does not. Observe what happens when you fit these models with software.