JOHN R. WILMOTH
Death-rates at all ages have fallen with an astonishing speed over the past 40 years in Japan. Life expectancy at birth among the Japanese is now the highest in the world for any national population. While most people agree on the advantages of a longer life for the individuals who benefit from it, there is increasing concern, in Japan as elsewhere, about an important side-effect of rapid mortality decline, namely, population ageing. All analysts agree that populations in developed countries like Japan will continue to grow older (whether measured by the median age of the population, the percentage over age 65, or some other indicator), but the extent and speed of this transformation over the next 30-50 years are somewhat less certain and depend, to a large extent, on future trends in death-rates at older ages ( Horiuchi, 1991).
In this chapter, trends in death-rates for total and cause-specific mortality for Japan are examined in an attempt to forecast the future level and pattern of Japanese mortality. In particular, an attempt is made to answer the following questions: 'Will life expectancy continue to rise in Japan, or will it reach an upper limit?' 'How will the proportion of the Japanese population that is older than age 65 (or 85) change in the future?' 'What causes of death will have the greatest influence on trends in total death rates?' In the process of making these evaluations, two important methodological questions arise: first, what difference does it make for our mortality forecasts if we make separate projections by cause of death, as opposed to a single projection for total mortality? And secondly, what difference does it make if we consider trends in Japanese mortality only during the period after 1950, for which relatively accurate and abundant data are available, as opposed to a longer time-frame?
The projection methods employed here are adaptations of the procedure proposed by Ronald Lee and Lawrence Carter. In the words of the authors, their procedure 'combines a rich yet parsimonious demographic model with statistical time series methods' ( Lee and Carter, 1992: 659). This method has been applied to both total and cause-specific mortality, although it was originally proposed for total mortality only. Some modifications to the original