Longevity advances in high-income countries, 1955-96

摘要

Lee and Carter (1992) have argued that the best historical predictor of change in American mortality from 1933 to 1987 is a function that generates linear projections of logged age-specific death rates. While their model allowed for a variable rateof mortality change over time, they found that the best fit involved an almost perfectly linear model. As Manton, Stallard, and Tolley (1991) have shown, many expert predictions of future change made in the course of the twentieth century were later proven to be wildly wrong, but 92.7 percent of the variance in age-specific log death rates over time can be explained by straight-line projections (Lee and Carter 1992; Lee and Miller 2001). Since Lee and Carter's article, advocates of historical projectionhave rallied to the use of logged age-specific death rates to predict future trends (Preston 1993; Wilmoth 1998a, 1998b; Tuljapurkar, Li, and Boe 2000). This article expands on this empirical work by asking: Is linear change in logged age-specific deathrates the empirical model that best describes change over the more recent period, 1955-96? If not, what simple trend has been the best historical predictor, and hence a good basis for future projections? Does the pattern found in the United States holdfor other countries whose economic and social development has been most similar? And what does the experience of the group of developed countries imply for the future of individual states?