Page 85 - IJPS-2-2

Haiyan Zhu, Qiushi Feng, and Danan Gu

these two questions, we coded his/her family/social relation as good (vs. poor). Self-rated life satis-
faction was a dummy variable, with ‘satisfied’ coded as 1 and ‘unsatisfied’ coded as 0. Due to great
regional differences in economic development in China, we additionally included the geographic
area of the respondent as a control variable with five categories: northern, northeastern, east-
ern/southeastern, central, and northwestern.

2.3 Analytic Strategy

We chose gender as the leading stratifying variable to examine relative hazard risks of mortali-
ty based on SRH and IRH. For each gender, we further separated the subpopulation by age, marital
status, urban/rural residence, years of schooling, and family economic condition. We used this strat-
egy because gender differentials are one of the major sources of health and mortality disparity in old
age, as discussed in the Introduction. Another reason that we stratified our analyses by these sub-
groups is that there were statistically significant differences in the associations between SRH/IRH
and mortality for each set of the subgroups (results not shown). To compare the predictive powers of
SRH and IRH for mortality, we calculated relative hazards of mortality risk based on three paramet-
ric Weibull hazard models for each of the subgroups: SRH only, IRH only, and both SRH and IRH.
This model design tests whether SRH and IRH have independent associations with mortality and
shows how those associations change in the presence of the other. All stratifying variables and co-
variates are from the baseline 2005 wave of the CLHLS, while mortality status and exposure incor-
porate data from the 2008 wave. Since the proportion of missing data was only about 2%, it was
not considered to be a serious source of bias.

3. Results
Table 1 describes the sample characteristics by gender. The majority of respondents were in the 80+
age group, more than half lived in rural areas, and most of the elders had poor family economic con-
ditions. There was a significant gender disparity for education and martial status. About 64.2% of
men received 1+ years of schooling, while the rate was only 7.8% for women. In terms of marital
status, women were more likely to be unmarried (widowed) in old age than men (81.4% vs. 49.2%).
Among covariates, it is noteworthy that women had worse health conditions than men with higher
scores of IADL, ADL, MMSE, and number of chronic diseases; men were more likely to smoke and
drink alcohol than women.
The sex-specific distributions of SRH and IRH in the baseline 2005 wave and deaths from 2005 to
2008 are shown in Table 2. Among older men, 11.7% of the sample reported their health as very
good, 37.8% as good, 34.6% as fair, and 15.9% as poor/very poor. In comparison, older women
tended to report poorer health status with 8.1% reporting health as very good, 36.1% as good, 36.1%
as fair, and 19.7% as poor/very poor. In terms of IRH, the proportions of reporting healthy, fairly
healthy, slightly ill, and moderately/severely ill were 32.1%, 54.0%, 12.2%, and 1.7%, respectively
for men; the corresponding rates among women were 22.3%, 57.4%, 17.6%, and 2.7%, respectively.
Overall, IRH tended to be better than SRH for both women and men, and IRH was rated lower for
women than for men. Table 2 also shows that the death rate was generally higher for women than for
men across the SRH and IRH categories, except within the healthy and fairly healthy categories of IRH.
Figure 1 shows Kaplan-Meier survival curves for SRH and IRH for men and women. It reveals
that respondents in very good or good categories of SRH had a better survival trajectory than those
in fair and poor/very poor categories of SRH. The same pattern applied to survival curves of IRH. In
other words, respondents with better ratings of SRH or IRH had a lower mortality risk.
Table 3 presents three-year mortality risks for men and women predicted by three parametric
Weibull hazard regression models: SRH only, IRH only, and both. SRH was not a significant predic-
tor of mortality in any models. In contrast, the differentiation of mortality risks by IRH was robust
with a good gradient for both men and women, regardless of inclusion of SRH and controlling for

International Journal of Population Studies | 2016, Volume 2, Issue 2 79