A traditional measure of effect size associated with tests for difference b
etween two groups is the variance explained by group membership (R-2). If e
xposure to a disease causes a small but long term deficit in performance, h
owever, R-2 does not capture that cumulating effect. We propose an alternat
ive statistic, gamma, based on the probability of an unexposed person outpe
rforming an exposed person. Although gamma is also a point estimate, it mor
e easily conveys what the cumulating effect of a deficit would be. We discu
ss some of the advantages of this measure. Copyright (C) 2000 John WiIey &
Sons, Ltd.