The plots become unusual when one crosses the threshold of the 40

th

year of

service. This is the point in the dataset were the data becomes very rare. Of the 2634 data

points in the dataset only nine represent members who have served 40 or more years.

These nine data points are spread among only four subjects. Any data at that extreme

needs to taken with a grain of salt.

The actual values of the coefficients in the regression can be interpreted regarding

their effects on the hazard of retirement. The odds ratio is exp(Beta-hat). For example

when one looks at the variable measuring margin of victory the coefficient is -0.168.

exp(-0.168) = .845

This means that a one unit increase in the natural log of the margin of victory will

reduce the hazard of retirement by 15.5 percent, all other variables being equal. This

interpretation can be shown using survival plots. Again using the margin of victory

variable we can see the effects of a change in the natural log of the margin of victory

from the extremely low value of -4 to the extremely high value of 4. This represents the

change in the hazard when the margin of victory in the previous election moves from

about 0.02 percent to about 55 percent. These are values near the extremes of this dataset.