Page 84 - IJPS-2-2

Self-rated health and interviewer-rated health: differentials in predictive power for mortality among subgroups of Chinese elders

gories: very good, good, fair, poor, and very poor. Following previous research (e.g. Feng, Zhu, Zhen
et al., 2016), we combined both very poor and poor into one category, “poor/very poor”, due to the
very low prevalence of the very poor category. In addition, there was a response category “unable to
answer” that accounted for about 8% of the sample. We performed a number of analyses to evaluate
how this group affected our study under different scenarios, such as excluding these cases, including
these cases with the assumption that they were in very poor health, or imputing these cases. Since all
of the methods showed similar results, we chose to impute these cases in the analysis. We assumed
that those who were not able to answer the question had the same answer as those who answered the
question if the attributes of the former group were the same as the latter group in terms of de-
mographics, socioeconomic status, family/social connections, health behaviors, and health.
2.2.3 Interviewer-Rated Health (IRH)

IRH was measured by the question “How do you rate the respondent’s overall health?” with four
response categories: “healthy,” “fairly healthy,” “slightly ill,” or “moderately or severely ill.” This
question was answered by the interviewer after the interview was completed.
2.2.4 Stratifying Variables
The sample was stratified into different subgroups for analysis based on demographics and socioec-
onomic status (SES). Demographic variables included chronological age group (65–79 vs. 80+), sex
(men vs. women), marital status (married vs. single), and residence (urban vs. rural). SES was meas-
ured by years of schooling (1+ years of schooling vs. none) and family economic condition com-
pared to others (good vs. not good). These variables are the basic factors significantly associated
with health and mortality at late ages (Feng, Zhu, Zhen, et al., 2016; McFadden, Luben, Bingham et
al., 2009; Zimmer, Hidajat, and Saito, 2015).

2.2.5 Covariates
Covariates included health conditions, health practices, health care coverage, social connection,
self-rated life satisfaction, and geographic area. Previous studies have shown that these variables are
either components of SRH or factors associated with the process of self-rating health (e.g., Feng,
Zhu, Zhen et al., 2016; Anson, Shteingrad, and Paran, 2011). By controlling for these variables in
models, we clarified how SRH and IRH could contribute differently to mortality prediction beyond
these established predictors. Health conditions included instrumental activities of daily living (IADL)
limitations, activities of daily living (ADL) limitations, cognitive impairment, and chronic conditions.
IADL was measured by whether a respondent needed assistance with the following eight activities:
visiting neighbors/friends, shopping, cooking, washing clothes, walking 1 km, lifting 5 kg, crouching
and standing up three times, and taking public transportation. For each of the eight items, no need for
assistance was coded as 0, and 1 otherwise. We summed these items to create an IADL index ranging
from 0 (no limitations) to 8 (limitations in all activities). ADL was measured by six items: eating,
dressing, indoor transferring, using the toilet, bathing, and continence. Scoring was similar to that for
IADL, producing a summed index ranging from 0 (no disability) to 6 (most severe disability). Cog-
nitive impairment was measured by the Chinese version of the mini-mental state examination
(MMSE), which tested respondents’ orientation, registration, copy and design, calculation, recall,
naming, and language. A score of 23 or less out of 30 was considered as cognitively impaired. Al-
ternative criteria for cognitive impairment were tested and the results were similar (not shown).
Chronic conditions were measured by the self-reported number of chronic diseases in a given list,
ranging from 0 to 11.
Health practices included whether the respondent smoked (yes vs. no), consumed alcohol (yes vs.
no), and regularly exercised (yes vs. no) at the time of survey. Healthcare coverage was measured by
whether the respondent was covered by public medical service (yes vs. no). Family/social connec-
tion was measured by two proxies: whether the participant has family members, neighbors or friends
to talk with when in need (yes vs. no) and whether the participant has family members, neighbors or
friends to ask for help when in need (yes vs. no). If a respondent gave a positive answer to either of

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