How Long Do We Live?

Period life expectancy is calculated from age-specific death rates using life table methods that are among the oldest and most fundamental tools of demography. These methods are rarely questioned, much less criticized. Yet changing age patterns of adult mortality in contemporary countries with high life expectancy provide a basis for questioning the conventional use of age-specific death rates and life tables. This paper argues that when the mean age at death is rising, period life expectancy at birth as conventionally calculated overestimates life expectancy. Estimates of this upward bias, ranging from 1.6 years for the United States and Sweden to 3.3 years for Japan for 1980–95, are presented. A similar bias in the opposite direction occurs when mean age at death is falling. These biases can also distort trends in life expectancy as conventionally calculated and may affect projections of future trends in period life expectancy. This material may not be reproduced without written permission from the authors. The measurement of human longevity is one of the oldest and most important topics in demography. Estimates of the life expectancy at birth (i.e., the average age at death under current mortality conditions) are now routinely provided by national and international statistical agencies. For example, the United Nations Population Division publishes such estimates for all countries in the world, ranging from a low of 37 years in Sierra Leone to 80 years in Japan for the period 1995–2000 (United Nations 2001). Life expectancy at birth worldwide is estimated at 63 for males and 67 for females. Life expectancy of a birth cohort may be calculated directly if survival in this cohort is observed until the last person dies. Life expectancy at birth is simply the average age at death. This calculation is unsatisfactory for many purposes, however, because it provides a selective summary of mortality over nearly a century, an interval during which mortality conditions are likely to have changed. Most obviously, life expectancy at birth calculated in this way is not useful for studying mortality change over periods of less than a century. The study of mortality change over shorter time spans is generally based on age-specific death rates calculated for a single year or for periods of several consecutive years. Most life expectancy statistics in the demographic literature are calculated from such rates by life table methods that originated with John Graunt's study of the London bills of mortality (Graunt 1662) and have been standard in the field for well over 100 years. They may be referred to as period life expectancies to distinguish them from the cohort or generational life expectancies calculated for groups of persons observed over long time periods. Measures of cohort life expectancy are so rare, however, that life expectancy is generally understood to mean period life expectancy. Methods for the measurement of mortality are regarded by many demographers as an all but closed subject. Age-specific death rates and life tables have been used for so long that their validity as measures of mortality is rarely questioned. In countries experiencing high life expectancy, however, certain empirical regularities are observed that provide a basis for revisiting the use of conventional methods. This paper argues that, for contemporary countries with high life expectancy, the conventional approach to the measurement of longevity is unsatisfactory whenever mortality is changing. We propose an alternative approach and suggest on this basis