Notes (Japan)

• Adjustments to Census and Death Counts
• Estimating January 1 Population
• Construction of the Japanese Cause-specific Mortality Data, 1951-1990
• The 40-category Causes of Death for Japan 1951-1990

Published death counts contained an "unknown age" category for every year from 1950 to1996. Deaths of unknown age were distributed proportionately across the entire age range by sex for the entire year and for October, November, and December.

For years 1950, 1955, 1975, 1980, 1985, 1990, and 1995, census counts contained an "age unreported" category. These counts were distributed proportionately across the age range by sex.

January 1 population estimates for ages 0-99 are constructed using the methods outlined in Vallin (1973). In brief, a cohort is "aged" from census to census, or from birth to census, by subtracting out recorded deaths. The difference between the actual and predicted cohort size at the time of the second census is assumed to be migration and is distributed evenly over the time period in question. The procedure is complicated by the fact that censuses refer to October 1, although we desire population estimates on January 1. Since deaths are recorded by month, however, we have an exact record of cohort deaths for each intercensal period. For cohorts aged 94.75+ at the time of the first census, there are missing values reflecting the fact that the cohorts involved would have been aged 99.75+ at the time of the second census. January 1 population totals for these cohorts at advanced ages were later reconstructed using either the method of extinct generations, the survival ratio method, or a combination of the survival ratio method and the intercensal survival method (see below).

For cohorts that have reached age 125 by 1996, January 1 population estimates are constructed using the method of extinct generations. Next, the survival ratio method is used to calculate January 1 population estimates for cohorts that have reached ages 106 to 125 by 1997. For cohorts aged 100 to 105 in 1997, the survival ratio method is used to obtain January 1, 1997 population estimates. The intercensal survival method is then applied from the 1990 census to the survival ratio estimate of the 1997 population. 

The above etimates are used for ages 80 and above and merged with the intercensal estimates to form the final January 1 population estimates. Where January 1 population estimates are available using both methods, I use the estimates derived by the method of extinct generations. January 1 population estimates for 1996 and 1997 (after the 1995 census) and 1950 (before the 1950) census are estimated by subtracting or adding deaths and assuming no migration.

  illustration

This report documents the construction of the Japanese cause-specific mortality data for the time period from 1951 to 1990. This project is done by Long Wang and Lijing Yan under the guidance and supervision of Prof. John R. Wilmoth at the Department of Demography, University of California, Berkeley. The main challenge is to establish a comprehensive yet concise cause classification system that includes major causes of death and is compatible across four revisions of the International Classification of Diseases, Injuries, and Causes of Death (ICD). The final product is a set of death counts in Japan by sex, by 5-year age group, and by 40 causes for each year from 1951 to 1990. The 5-year age groups are: total, 0, 1-4, 5-9, ... , 95-99, 100+, and unknown ages.

Raw data were converted into a 40-category cause classification system. The entire time period from 1951 to 1990 is divided naturally into four periods by their corresponding ICD revisions: 1951-57, 6^th; 1958-67, 7^th; 1968-78, 8^th; and 1979-90, 9^th. The appendix shows the 40 categories and their corresponding ICD codes by time period.

Internal consistency was checked by comparing the "total" column with the sum of all other columns and by looking at the continuities of the data from year to year. The data were found to be completely consistent internally. Whenever possible, the converted data were also checked against some external sources. All detected data entry errors were corrected as well as a few obvious inconsistencies in the original data sources.

As of January 1998, there is a problem in the original electronic data for years 1979 to 1990. The data for age group 95-99 were omitted and deaths in 95-99 age group and unknown ages were lumped together. The current solution is to distribute the total "unknown" deaths obtained from another source proportionally to each cause of death according to the share of this cause in the total number of deaths. Then the 95-99 age group is the difference between the distributed unknown deaths and the total of 95-99 and unknown, which is available in the raw data. When the difference is negative, the distributed unknown was forced to zero and the 95-99 was equal to the original total of 95-99 and unknown. The final computed 95-99 and unknown deaths were rounded to integers. This procedure was done for each year from 1979 to 1990. Unfortunately, the total number of deaths in 95-99 or unknown age group after such manipulation do not add up to the real totals known from another source. At this point in time, we are still waiting for the correct data to arrive from Japan to solve this problem for good.

The sixth through ninth revisions of "The International Classification of Diseases, Injuries, and Causes of Death" adopted for use in Japan are the ideal sources for documenting the ICD codes and the corresponding disease names. However, they are published in English in the Vital Statistics Japan for two time periods only (7^th 1958-67 and 8^th 1968-78). For years 1951-57 (6^th), the World Health Organization publication of the Manual of ICD is used; and for years 1979-90 (9^th), the United States Department of Health and Human Services publication of the ICD (clinical modification). In addition, they are supplemented by and checked against other sources available at the time the project was undertaken.

* The B-list number, when existing, is in the parentheses for 6-8^th ICD (1951-78); and for 9^th (1979-90), the 120 list number.

** ICD 421.1 should be in Cause 23 instead of in Cause 25. However, 421.1 was not tabulated for this time period in Vital Statistics Japan and thus not added to Cause 23. Judging from the secular trend of death counts in this cause, there seems to be some discontinuity due to the changes in ICD revisions even if this problem could be avoided.