Footnote 79

A more refined analysis would use a logistic regression model to estimate survival probabilities for various possible combinations of claims - e.g., sex alone, sex and retaliation, sex and race, age and race, age with race and retaliation, etc. For a discussion of logistic regression, see Michael O. Finkelstein & Bruce Levin, Statistics for Lawyers, § 14.7 (2d ed. 2001). We made this analysis, combining some types of claims so that there were only 15 groups. But the coefficients estimated for the groups were not statistically significant and, in some instances, were perverse (e.g., sex plus retaliation had a lower probability of surviving than sex alone). In addition, the probabilities of survival for some combinations were unreasonably high. It seems clear that we did not have sufficient data for this more detailed analysis. We are indebted to Bruce Levin for suggesting this approach, which may be useful in future studies.