Pii: S0531-5565(00)00131-5

As A gets larger the inflection point (when it exists) gets smaller. The parameter we (Drapeau et al., 2000) call a “breakpoint” is the day at which a Gompertz equation plus a flat line fits the data better than the Gompertz alone. The actual values of the breakday occur before the data have settled down to a flat line. This happens because well before an asymptote is reached the rate of increase in mortality rates slows and the observed mortality rates start to depart from the Gompertz predictions. Consequently, we would expect that the parameter breakday would be highly correlated with the inflection point. If the more robust lines have lower values of A then they should have greater breakdays, which they do not. Eq. (1) can be used to estimate the parameters of A, b, and s 2 for the variance-in-A model. When this is done for large cohorts the resulting model predicts the timing of mortality fairly well except for a prediction of a small class of very old individuals that are not observed (Rose and Mueller, 2000). The difference between the observed number of very old individuals and the expected is large enough to reject the variance-in-A model. If the variance-in-A model described by Service is to be taken seriously then there should be some empirical support for its underlying mechanisms. Eq. (1) can be used to derive estimates of s 2 directly. For many Drosophila populations, we typically get estimates in the range of 1. If we assume the variance of frailty is 1 and has a gamma L.D. Mueller et al. / Experimental Gerontology 35 (2000) 1089–1091 1089