Page 22 - IJPS-2-2

Assessments of mortality at oldest-old ages by province in China's 2000 and 2010 censuses

lieve that these crossovers are due to significant undercounting of death or age misreporting at ages
90 and older. Considering the relatively smaller discrepancies between the observed and fitted death
rates, underestimations were also prevalent at ages 60 to 70. As a result, the Kannisto curve was fit
at a very low level, creating a crossover in death rates with Japanese females. Also considering that
Japanese females have the world's highest life expectancy, it is highly implausible that these Chinese
provinces (relatively undeveloped areas in China) would have life expectancies at older ages com-
parable to older Japanese females. Of course, it is possible that very low mortality in Hainan Prov-
ince (in South China Sea) may be because it has attracted many older migrants from other provinces
seeking the favorable climate during winter months (Xia, 2016). In turn, these seasonal mi-
grants/retirees may have been enumerated as a part of de facto population counts in Hainan Province
at the time of the censuses (November 1). However, these older adults may ultimately die in their
residence of origin — which brought down death rates in the censuses for Hainan Province. Never-
theless, more empirical research is needed to validate these arguments.
Our findings on age-trajectories of mortality are consistent with previous research on Chinese
oldest-old showing mild to substantially lower age-trajectories of mortality after age 97 compared
with the Kannisto curve among Han Chinese (Zeng and Vaupel, 2003). Considering that Han Chi-
nese have more accurate age-reporting than ethnic minorities and that discrepancies from the classi-
cal age-trajectory remains prevalent after 20 years, we conclude that age exaggerations exist not only
among ethnic minorities, but also among Han Chinese, especially after age 90. The rates of underes-
timation for mortality at ages 70 and older are also generally consistent with recent findings by
Wang (2013).
Underestimations of mortality — in terms of the level and shape of age trajectories — in the 2000
and 2010 Chinese censuses may be attributable to age exaggeration/misreporting, underreports of
deaths, and/or incomplete death registration. In China, the prestige associated with old age in tradi-
tional Confucian culture may motivate individuals to overstate their age (Chou, Ju, and Huang,
2013). Indeed, several studies have documented age exaggerations in enumerated population counts
in the Chinese censuses (Liu, 1991; Wang, 2012; Yang, 1988). In cases of age exaggeration, death
rates computed from death records will be biased downwards relative to true death rates; and the
discrepancy with true data would progressively widen with advancing age (Andreev and Gu, 2017).
Furthermore, in Chinese censuses, death counts are primarily collected from reports gathered from
household family members — which can lead to possible misreporting and/or underreporting for
purposes of receiving social benefits.
We recognize the need for caution when interpreting our findings. First, we did not provide exact
levels of underestimation in mortality. Instead, we only explored possible rates of underestimation
from regression-based models (strict criterion) and from confidence ellipses (lenient criterion). The-
se two scenarios can be interpreted as the high and low boundaries, respectively, for underestimation.
Furthermore, because data were not available for calculating age-sex-province-specific underestima-
tions, we assumed that rates of underestimation were the same for China and its provinces at ages
60–70 (or at ages 70–80) when we estimated rates of underestimation at ages 70–95 (or at ages
80–95). However, these assumptions may not be true (Wang, Wang, Cai et al., 2011). Alternatively,
our findings would be more robust if province-age-sex-specific rates of underestimation are applied
when data become available. Therefore, we encourage more research on the levels of mortality un-
derestimation at oldest-old ages in the 2000 and 2010 censuses.
Second, although some provinces had age trajectories that matched the Kannisto curve, it does not
mean that there were no (or low) rates of underestimation in these provinces. The Kannisto model
does not produce adjusted estimates for mortality at advanced ages (Andreev and Gu, 2017). Instead,
it only smoothes the age trajectory of mortality and extrapolates death rates at very old ages (e.g.,
age 110 or beyond). All ages (80 or older) may be systematically or proportionally underreported
even though the observed death rates match the Kannisto model. Nevertheless, if age-specific death
rates do not follow the Kannisto function, these age rates are likely very distorted and proper ad-

16 International Journal of Population Studies | 2016, Volume 2, Issue 2