The final step was the inclusion of the experience variable and the interaction

between experience and position. Since in a duration model one is not just predicting the

retirement but the time to retirement it is inappropriate to include the amount of time

served in the House as an independent variable in equation. It would be predicting itself.

The inappropriateness of doing this was pointed out by Box-Steffensmeier et al . (2003).

However, in the same paper they also point out the usefulness of time interaction terms in

Cox proportional hazard models. This is exactly what the variable of interest, career

ceilings, represents. It is operationalized as position multiplied by time served in the

House. Box-Steffensmeier et al. endorse this type of interaction as a way around the

major assumption of Cox proportional hazard models, which is that the effects of the

variables are proportional. By interacting a covariate with time one can allow the effect

on the hazard of that covariate to vary monotonically over time. This is exactly what we

want in this case. We are not interested in the effects of one’s position on the likelihood

of retirement but whether the relationship between one’s position in the House and the

likelihood of retirement changes over the course of one’s career. Theriault predicts that as

time increases position will be a greater predictor of retirement.

*III - Results*

These variables are put together in a Cox proportional hazard model. This model

has advantages that other duration models do not. Most importantly it allows the baseline

hazard to vary. It does not assume a particular distribution for the hazard. This is a

significant advantage over more constraining parametric models. The results of the model

are shown below in Table 1. (The exact method was used to deal with ties.)