Measures
The dependent variable of interest, the score of depressive symptoms, was measured with 10 questions from the Center for Epidemiologic Studies Depression Scale (CES-D10) [
8,
22,
25]. Respondents were asked how frequently in the last week they: were bothered by things; had trouble concentrating on things; felt depressed; felt everything was an effort; felt hopeful about future; felt fearful; had restless sleep; was happy; felt lonely; and could not get going. After reverse scoring the two positively worded items, we assigned a score from 0 to 3 for each item as follows: 0 for 0 days, 1 for 1–2 days; 2 for 3–4 days; 3 for 5–7 days. Scores of these 10 items were summed to create an additive scale score ranging from 0 to 30, with higher scores indicating more depressive symptoms. The reliability of depression items was tested using Cronbach’s alpha and found to be satisfactory at each wave (alpha = 0.76 in 2013 and alpha = 0.81 in 2011).
The key independent variable, perceived availability of future care (PAFC), was measured by the following question “Suppose that in the future, you needed help with basic daily activities like eating or dressing, do you have relatives or friends (besides your spouse/partner) who would be willing and able to help you over a long period of time? The response option was yes or no. Coding this variable in such a manner allows us to discuss unavailability of support as a risk factor for depression. Additionally, for respondents answering “yes”, a follow-up question was asked about whether the source of expected care support would be children, other relatives, or friends. Examining those with care expectations by source of care (not shown) revealed the importance of off-spring in care availability: 98% of respondents expected care from their children while only 2% expected care from other relatives or friends. This provides evidence to support the existing of traditional filial piety culture in China and the importance to control for the number of children, the frequency of contacts/visits and living arrangement between parents and children into the final regression models.
We chose potential control variables based on previous studies of the determinants of depression in China and other Asian countries [
21,
22,
25,
26,
29,
37‐
39]. These variables were categorized as demographical, socio-economic/finance, health and social factors. Demographics included age and gender. Age was divided into four groups: 45–54, 55–64, 65–74 and 75+. Socio-economic status was represented measured by the highest educational level achieved, perceived living standard, and urban/rural residency. Education was measured as the highest level of education achieved based on three categories: Primary schooling or less; secondary schooling; and college or higher degree. Relative living standard was assessed by responses to the question “Compared to the average living standard of people in your city or county, how would you rate your standard of living?”
2 Response options were
much better, a little better, about the same, a little worse, much worse, collapsed into three categories corresponding to better, same, and worse [
40]. Worse than the average relative living standard is treated as having financial strain/uncertainty.
Urban and rural residency was determined by the most recently published statistical standard by the Chinese National Bureau of Statistics based on an area’s social and economic development [
40].
Health factors included limitations in activities of daily living (ADL), limitations in instrumental activities of daily living (IADL), number of chronic diseases, functional loss, poor memory, level of chronic pain, and childhood health. ADL limitation was indicated if the respondent -reported difficulty performing any of the following basic activities: bathing/showering, eating, dressing, getting into or out of bed, using toilet, or controlling urination and defecation. IADL limitation was indicated if the respondent reported difficulty in any of the following household activities: doing household chores, preparing hot meals, shopping for groceries, managing money, and taking medications. Chronic diseases were assessed as the number of diagnosed health conditions categorized as none, one, two to three, and more than three. Functional loss was indicated by whether respondents reported any of the following disabilities: brain damage/mental retardation, vision problem (blind or half blind), hearing problem (deaf or half deaf) and speech impediment (full or half).” Self-reported memory was assessed with the question: “How would you rate your memory at the present time?”. A dichotomous variable was created differentiating excellent/very good/good (=0) from fair/poor(=1). Level of pain was ascertained by the question “Yesterday, did you feel any pain?”, of pain was assessed as no pain (1), a little pain (2), some pain (3), quite a bit of pain (4), and a lot of pain (5). Childhood health was measured by asking “How would you evaluate your health during childhood, up to and including age 15: excellent (1), very good (2), good (3), fair (4), poor (5). Both level of pain and childhood health were continuous variables controlled in the multivariate regression models.
Social factors included family structure, intergenerational arrangements (contact/co-residence/geographic distance), and social activities. Family structure was measured by marital status and number of children. Marital status was operationalized as currently married or cohabitating, formally married (divorced, separated or widowed), and single or never married. Number of living children was categorized into four groups: no children, one child, two children, and three or more children. Intergenerational contact was measured as the most frequent form of contact (face-to-face visits, phone, email and Internet) with non-coresident adult children. Contact frequency was categorically assessed as daily, weekly, monthly, once per year, or less than once per year. Geographic distance from children was based on the location of the closest child: same/adjacent household/dwelling/courtyard, another household in your village/neighborhood, another village/neighborhood in your city/county/district, or outside your city/county/district. Three categories were generated from these responses: Having at least one child living in the same or adjacent household, having a child living in the same village/neighborhood but not co-residing or living adjacent, and having all children living beyond the village/neighborhood. Participation in social activities was assessed by whether or not respondents participated in each of three types of activities in the last month: leisure activities (e.g. interacting with friends; playing Ma-jong or cards, participating in a club or community-related organization); helping activities (e.g. providing help or care for family, friends, neighbors or others who do not live with you and did not pay you); and educational activities (e.g. attending an educational or training course for stock investment; using the Internet).
The CHARLS interviewed both the husband and the wife in a same household as long as they were both aged over 45, as one of them being the “main respondent”. We note that the CHARLS study randomly selected one respondent per household (58% are main respondents and 41% are the spouses of the main respondents) to report family-level information, including number and location of children, as well as contact and exchanges of money and support with children. This method resulted in about half of our sample having no directly reported data about children. We empirically handled this issue by generating a category for “non-response” in order to retain these cases in our analysis. As a robustness check, we also estimated models borrowing values from reporting respondents and found very similar results (not reported).
Analytic approach
Previous studies on risk factors of depressive symptoms among older adults in China generally find that depression correlates with being older, female, retired, physically disabled, chronically ill, financially stressed, and low educated, and having weak social and family support systems [
6,
8,
21,
22,
41]. We account for these factors in our predictive models of depressive symptoms as a function of expected future care.
Emotional distress deriving from uncertainty in meeting future care needs may be assuaged by financial resources (allowing the purchase of private services), good functional health (rendering its impact less consequential), and living in an urban area (where a relatively strong service infrastructure and relatively weak filial norms shift the burden away from families).
In order to select the most relevant control variables, we used univariate OLS regressions predicting depressive symptoms to identify plausible variables with coefficients significant at <.10 and r-squares higher than 0.01. Variables that did not meet these criteria were excluded from our analysis. Selected variables are presented in Table
1. Multivariate OLS (Ordinary Least Squares) regression was then used to explore the association between perceived future care availability and depressive symptoms, first with only demographical variables controlled, then sequentially adding health, socioeconomic, and social variables until all control variables were entered to the final model. We used this hierarchical estimation approach to assess the unique contribution of each variable grouping toward explaining the association between perceived future care and depressive symptoms.
Table 1Perceived availability of future care and mean depression scores by sample characteristics
Perceived availability of future care |
Expected future care support (Yes) | 11,886 | 70.7 | 100.0 | 7.1 |
Expected future care support (No) | 5010 | 29.3 | 0.0 | 9.1 * |
Predisposing factors |
Aged 45–54 (Ref.) | 5920 | 33.4 | 71.5 | 7.3 |
Aged 55–64 | 6694 | 35.2 | 69.0 * | 8.0* |
Aged 65–74 | 3736 | 19.7 | 70.1 | 8.2* |
Aged 75+ | 1855 | 11.7 | 74.6* | 7.8* |
Male (Ref.) | 8806 | 48.4 | 69.1 | 6.9 |
Female | 9429 | 51.6 | 72.3* | 8.6* |
Financial factors |
Rural (Ref.) | 10,881 | 58.7 | 71.6 | 8.4 |
Urban | 7361 | 41.3 | 69.4* | 6.8* |
Under primary (Ref.) | 8182 | 44.9 | 71.5 | 8.9 |
School without degree | 9633 | 52.9 | 70.4 | 7.0* |
College and above degree | 414 | 2.2 | 63.5* | 5.2* |
Better living standard (Ref.) | 503 | 3.6 | 69.7 | 6.0 |
About same living standard | 3595 | 28.1 | 72.3 | 6.4 |
Worse living standard | 8891 | 68.4 | 69.0 | 8.5* |
Living standard not reported | 5634 | ~ | 72.6 | 7.5 |
Health factors |
With ADLs (Yes) | 1114 | 9.4 | 69.0 | 12.3* |
With ADLs (No) (Ref.) | 11,360 | 90.6 | 69.3 | 8.8 |
ADLs not reported | 5621 | ~ | 73.9 | 5.2 |
With IADLs (Yes) | 2628 | 14.8 | 70.2 | 11.4* |
With IADLs (No) (Ref.) | 15,402 | 85.2 | 70.8 | 7.3 |
No disease (Ref.) | 4846 | 33.7 | 72.8 | 6.5 |
One disease | 4430 | 29.9 | 71.6 | 7.8* |
Two diseases | 2844 | 18.6 | 69.9* | 8.5* |
Three and more diseases | 2728 | 17.8 | 66.6* | 9.8* |
Diseases not reported | 3394 | ~ | 70.4 | 7.5 |
With functional loss (Yes) | 16,030 | 10.9 | 68.9 | 10.2* |
With functional loss (No) (Ref.) | 2047 | 89.1 | 70.9 | 7.5 |
With bad memory (Yes) | 2751 | 82.4 | 68.7* | 8.3* |
With bad memory (No) (Ref.) | 13,520 | 17.6 | 75.6 | 5.2 |
Memory not reported | 2352 | ~ | 76.8 | 9.8 |
No pain (=1) (Ref.) | 10,545 | 65.2 | 72.0 | 6.4 |
A little pain (=2) | 2915 | 17.6 | 69.2* | 8.9* |
Some pain (=3) | 1383 | 8.3 | 65.1* | 10.8* |
Quite a bit pain (=4) | 1051 | 6.0 | 65.1* | 12.6* |
A lot of pain (=5) | 519 | 2.9 | 55.6* | 14.4* |
Pain not reported | 2210 | ~ | 76.4 | 11.4 |
Excellent childhood health (=1) (Ref.) | 1883 | 10.7 | 71.4 | 6.8 |
Very good childhood health (=2) | 6774 | 38.1 | 72.0 | 7.5* |
Good childhood health (=3) | 4753 | 26.8 | 70.3 | 7.9* |
Fair childhood health (=4) | 3226 | 17.4 | 69.3 | 8.3* |
Poor childhood health (=5) | 1244 | 7.0 | 66.1* | 9.0* |
Social factors |
Married (Ref.) | 15,799 | 84.8 | 70.2 | 7.5 |
Separated/divorced/widowed | 2273 | 14.2 | 75.8* | 9.3* |
Single | 155 | 1.0 | 42.0* | 10.1* |
No child (Ref.) | 378 | 4.2 | 48.6 | 8.6 |
One child | 1878 | 18.0 | 67.4* | 7.3* |
Two children | 3616 | 31.6 | 70.8* | 7.9 |
Three or more children | 4930 | 46.3 | 73.1* | 8.5 |
Child number not reported | 7821 | ~ | 71.1 | 7.2 |
Children visiting daily or weekly (Ref.) | 4652 | 54 | 74.3 | 8.0 |
Children visiting every month | 1510 | 18.1 | 72.1 | 7.7 |
Children visiting every year | 2236 | 24.9 | 65.0* | 8.8* |
Children visiting less than once per year | 216 | 2.7 | 45.7* | 9.3 |
Children visiting not reported | 10,009 | ~ | 70.6 | 7.3 |
Children contacting daily or weekly (Ref.) | 3568 | 50 | 70.4 | 7.9 |
Children contacting every month | 2083 | 30.0 | 68.8 | 8.9* |
Children contacting every year | 421 | 6.1 | 67.2 | 9.9* |
Children contacting less than once per year | 878 | 14.1 | 70.2 | 8.8* |
Children contacting not reported | 11,673 | ~ | 71.3 | 7.3 |
With co-resident children | 6144 | 56.9 | 75.4* | 8.1 |
With children in a same neighborhood | 1616 | 15.0 | 67.2* | 7.8 |
With children in another neighborhood (Ref.) | 3042 | 28.2 | 61.25 | 8.2 |
Living arrangement not reported | 7830 | ~ | 71.1 | 7.2 |
Social activities leisure (Yes) | 8747 | 47.9 | 71.8* | 7.1* |
Social activities: leisure (No) (Ref.) | 9348 | 52.1 | 69.7 | 8.5 |
Social activities: helping others (Yes) | 2495 | 13.8 | 71.7 | 7.1* |
Social activities: helping others (No) (Ref.) | 15,600 | 86.2 | 70.6 | 7.9 |
Social activities: learning (Yes) | 999 | 6.3 | 61.1* | 5.4* |
Social activities: learning (No) (Ref.) | 17,096 | 93.7 | 71.4 | 7.9 |
Next we tested stress-buffering/double jeopardy hypotheses by adding interactions between perceived future care availability and financial insecurity, functional disability, and urban/rural residence. We estimate the confounding effect of rural-urban residents, financial conditions, and ADL when they are controlled parallelly with perceived care support, while estimate the moderating effect when they are interacted with perceived care support in the model.
Since we were alert to the possibility that perceived availability of future care is endogenous to depressive symptoms--a condition that would exist if depressed individuals were less capable of mobilizing an effective support network—we also estimated regression models controlling for a lagged measure of depressive symptoms taken in 2011.
Finally, as mentioned above, we have treated respondents with missing information on family support into a “not reported” group as it is missed by random so that we can run the model with full sample size. However, robustness check has been done by assigning the reported values to other respondents within a same household. Only very small and insignificant differences were found between using the initial data and imputed data (See Table
8,
9 and
10 in Appendix 3).