Coale and Li (1991) concluded that, among all developing countries, the Chinese Han population is probably the only one that has both large numbers of extremely old people and age reporting that is about as reliable as in developed countries. A pilot study of our healthy longevity project reached very similar conclusions. This study focused on age validation of Chinese Han centenarians through rigorous analysis and comparison of the demographic indices of the age reporting with those of Swedish, Japanese, French, and Italian oldest old populations. Our conclusion is that the age reporting of the Chinese Han oldest old population is generally as good as that in Western countries up to the age of 105. The age reporting after age 105 is somewhat questionable (Wang, Zeng, Jeune, and Vaupel, 1998). These “super-centenarians”, who reported ages of 106 and above (156 in total in our sample), will be investigated later case by case and are not included in our analysis in this paper.

How is the age reporting in our 1998 survey, which also included some minority ethnic groups (about 7 percent of the sample) living in the 22 Han-dominated provinces? One of the most effective ways to answer this question is to evaluate the age distribution of the interviewed centenarians. We found the age distribution of centenarians interviewed in our 1998 survey to be very similar to that of Swedish centenarians, especially the curves for the Chinese and Swedish male centenarians, which are almost identical (see Figure 1 on page 7). This fact has led us to believe that age reporting in our 1998 survey is generally good. The Han, Zhuang, Hui, Yao, Manchu, Korean, and Mongolian ethnic groups make up 92.8, 4.4, 1.3, 0.7, 0.3, 0.1, and 0.03 percent of the sample, respectively. These seven ethnic groups account for 99.7 percent of the entire sample, and their Whipple’s Index and Mayer’s Index based on the 1982 and 1990 census data all indicate “very good” quality of age reporting (see Table 1). A comparison of age distributions and sex ratios of Han and Swedish centenarians shows a high degree of similarity. When one compares the centenarians from the other six minority ethnic groups with their Swedish counterparts the differences are somewhat greater, but the data are still generally acceptable (see Table 2). Some other ethnic minority groups living in the other nine provinces that do not report age accurately (e.g. Uygurs, which consist of 47.5 percent of the total population in Xinjiang) are not included in our sample.

Table 1 Ethnic Composition of the Chinese 1998 Survey and the Age Reporting Indices

Table 2. Age Distribution (Males and Females Combined) and Sex Ratio of
Han and Minority Chinese Centenarians Interviewed in the 1998 survey,
as Compared to Swedish Centenarians

The reliability coefficients of the 10 categories of variables are rather good (see Table 3). For example, the ADL reliability coefficient is 0.88 in our 1998 survey, while it was 0.87 in the Duke Older American Resources and Services Program survey (Fillenbaum, 1988) and 0.89 in the Canadian 1991-92 elderly survey (Penning and Strain, 1994). Another way of measuring reliability is to conduct factor analysis to see whether interviewees’ answers to questions of the same category are consistent (Anita et al., 1992). If they are, the answers to questions of the same category should be classified in one component and the values of the coefficients should be close to each other. Our factor analysis demonstrates clearly that the consistency and reliability of the answers are rather good (see Table 4).

Table 3 Reliability Coefficients

Notes: a. RC = Reliability Coefficient. We used Cronbach’s alpha to compute RC as follows:

Table 4 Factor Analysis - Pattern Matrix

The rates of logically inconsistent answers seem reasonably low (see Table 5). The rates of “Don’t know” and “Missing” answers are also relatively small (see Table 6). The above-mentioned indicators have led us to believe that the data quality of our 1998 survey is generally good. However, we also realize that some problems exist in the data set. This is expected given its nature as the first large survey of oldest old people conducted in a developing country. For example, the proportion of oldest old living in institutions might be over-estimated. We noticed that the proportions of those living in institutions based on the 1998 survey had substantially increased as compared with the corresponding figures based on the 1990 census. This increase might be due to two main factors. First, because of rapid economic development and increased governmental and societal attentions to social service for the elderly since 1990, there was a real increase in institutional facilities between 1990 and 1998. Second, as discussed earlier, for each centenarian in the randomly selected half of the counties and cities, we instructed our survey teams to try to interview one close-by octogenarian and one close-by nonagenarian of pre-specified age and sex. If the centenarians resided in or close to an institution, it was likely that the matched octogenarians and nonagenarians were selected from the institution. Therefore, the institutionalized oldest old might be over-sampled. Another example is that we found that a substantially higher proportion of the oldest old, especially centenarians, were unable to answer more of the personality-related questions than other types of questions. This is because some illiterate oldest old, especially centenarians, could not understand the questions about personality, which ask oldest old to provide a comparison of themselves to a typical person with a specified disposition (e.g. always look on the bright side of things). We expect that further evaluation, accompanied by subsequent in-depth analysis, will provide information that will permit a more critical examination of the data quality.