Page 9 - IJPS-2-2

Danan Gu, Runlong Huang, Kirill Andreev, et al.

function fit the data best from ages 65 to 95 and that the mortality decline at older ages is real rather
than an artifact of biases. Wang (2013) applied the Brass logit system to the 1990, 2000, and 2010
Chinese censuses by using the mortality scheme in the 1982 census as a standard. He found that
there was an approximately 20-30% rate of underreporting in female deaths at ages after 65 in the
1990 census and 10–12% rate of underreporting in male deaths in the 2010 census. Huang and
Poston (2000) applied linear regressions between the logit-transformed death rates and age from ag-
es 50 to 89 in the 1981 and 1990 censuses and identified biases between fitted and observed rates for
these ages. They found that the observed rates in most ages from 50 to 89 had a slight to mild (±
3%) bias with fitted rates in terms of their overestimation in even ages and underestimation in odd ages.
Among these Chinese applications, however, almost all of the studies suffered from some design
or methodological issues, with the exception of that from Zeng and Vaupel (2003). Duan and Shi’s
(2015) conclusion of actual mortality deceleration at old ages is likely true; however, the degree of
deceleration is presumably incorrect because many of these methods have been shown to be inade-
quate at such age ranges — and thus cannot capture the actual underestimation of mortality at these
ages in the censuses. Likewise, the findings by Wang (2013) are likely biased due to reported inac-
curacies in the mortality scheme in the 1982 census (Banister and Hill, 2004). In the case of Huang
and Poston (2000), as the linear relationship between the logit-transformed mortality and age is an-
other form of the Kannisto model (Thatcher, Kannisto and Vaupel, 1998), the approach is only valid
when mortality after age 50 follows the Kannisto model. However, the HMD data show that death
rates from age 50 to 89 do not follow the Kannisto model. Therefore, the linearity is presumably due
to an underestimation of mortality at later ages.
In examining the accuracy of death rates and age aggregation at oldest-old ages, Coale and Kisker
(1986) investigated the ratio of the population aged 95 or older versus the population aged 70 or
older — in 23 countries/areas with relatively accurate data — and found that this proportion was less
than 6 per thousand. However, they found that the proportion ranged from 1–10% in 28 countries
with poor data. Zeng and Vaupel (2003) found that this ratio was 1 per thousand for male Han Chi-
nese and 2 per thousand for female Han Chinese in the 1990 census, almost exactly the same as the
values for Sweden in the period 1985–1994. However, because such a ratio increases with reductions
in mortality, the absolute value of the ratio has a limited meaning.
Numerous recent studies have documented subnational variations of mortality in China and have
shown substantial provincial differences in life expectancy at birth and overall mortality for the gen-
eral population (Bignami-Van Assche, 2005; Fang, 1993; Hao, Arriage and Banister, 1988; Li,
Bateman and Liu, 2015; You and Zheng, 2005; Zhou, Wang, Zhu et al., 2016). For example, Zhou et
al. (2016) reported significant heterogeneity in life expectancy at birth and in mortality rates at ages
0–14, ages 15–49, and ages 50–74 across provinces — with provinces in eastern-coastal areas exhib-
iting the highest life expectancy; whereas provinces in western China exhibited the lowest life ex-
pectancy. Moreover, these provincial patterns have remained consistent since the early 1980s (Hao,
Arriage and Banister, 1988; Ren, You, Zheng et al., 2004; You and Zheng, 2006; Zhou, Wang, Zhu
et al., 2016). The provincial/regional variations may be attributable to differences in socioeconomic
development (Zhang, Wu and Zhang, 2004), incompleteness of registration systems (Song, 2000;
Wang, Wang, Cai, et al., 2011), and age misreporting in censuses/surveys (Guo and Che, 2008; Liu,
1991).
With one exception (Huang and Poston, 2000), however, no existing study has focused on provin-
cial variations in mortality after age 80 in China. The lack of research on mortality at oldest-old ages
in China is largely because of concerns about the low-quality of mortality data in the census and the
inaccuracy of age reporting, especially among non-Han minorities. For example, Coale and Li (1991)
found that death rates at very-old ages in China exhibited serious distortions from age exaggeration
in the Xinjiang province. In Xinjiang, Han Chinese account for only about 40% of the total popula-
tion in 2010 (National Bureau of Statistics of China, 2012). The majority of Xinjiang inhabit-

International Journal of Population Studies | 2016, Volume 2, Issue 2 3