Can multi-stage mortality selection explain a mortality deceleration puzzle?
Elizabeth Wrigley-Field, University of Wisconsin at Madison
Felix Elwert, University of Wisconsin at Madison
A pervasive demographic result is that mortality decelerates: it rises more slowly at very old ages than at younger ones. We extend deceleration analysis by incorporating new dimensions of social stratification, estimating deceleration in U.S. subpopulations defined by baseline health and poverty status, as well as race and sex. Using U.S. Medicare data, we follow 28 million Americans -- nearly the entire elderly population -- from 1993 to 2002, estimating nearly non-parametric mortality hazards. Our results create a puzzle. The traditional explanation of deceleration is mortality selection: populations become increasingly robust as their frailest members die. Conventional interpretations posit that populations with higher mortality should decelerate at younger ages, and more sharply, since they are subject to greater selective pressure. We find the expected pattern along lines of race and sex: higher-mortality African-Americans and men decelerate earlier and more steeply than white Americans and women. For health and poverty, however, the pattern reverses: it is the non-sick, and especially, the non-poor whose mortality decelerates substantially earlier and more sharply than their higher-mortality counterparts. To explain this contradiction with the conventional view, we extend mortality selection theory. Drawing on the demographic literature suggesting mutual causation between poverty and ill health, our multi-stage mortality selection model suggests that the non-poor population may be more heavily selected because the frail tend to become poor. The dynamism of this substantively plausible model, however, comes at a price. Since people enter the poor population from the non-poor, the poor’s frailty composition changes with selection on the non-poor population. We show that this relationship between populations makes deceleration order essentially unpredictable, and discuss implications for using deceleration patterns to understand health inequalities.