SLIDE 1
Causal Analyses of Chinese Elders' Living Arrangement and Subjective - - PDF document
Causal Analyses of Chinese Elders' Living Arrangement and Subjective - - PDF document
Causal Analyses of Chinese Elders' Living Arrangement and Subjective Well-being Jinyuan Qi ABSTRACT This study employs propensity score matching, weighted regression and fixed effects models to examine the causal relations between living
SLIDE 2
SLIDE 3
3
paper will present matching results and regression results using weights from propensity matching score, and then compare the cross-sectional results in 2011 with the results from panel data analyses. The last part is the discussion, limitation and implication of this study. Literature Review From late 1970s to now, the Chinese family household has been aging rapidly and dramatically, and transformed to a smaller size due to the fertility decline, increasing life expectancy and dramatic changes in social attitudes and economic mobility which is related to living arrangements between elderly Chinese and their children (Yi, et al, 2008). Under a medium fertility and mortality scenarios, Yi, et al (2008)’s population projection shows that by the year 2030 and 2050, the proportion of the elderly aged 65 or above living in empty-nest households without co-residing children among the total population will be 2.9 and 4.6 times of that in 2000, and the percentage of the oldest-old aged 80+ living in empty-nest households will be even more dramatic: 3.4 and 11.1 times in 2030 and 2050 as in the year 2000. Only a few studies investigate what factors are related to Chinese elders’ living arrangements. Sereny (2011) has reported the association between living arrangements and ethnicity and SES, living arrangement preferences, self-reported health and disability. Zimmer (2005) has also found the oldest-old living arrangements are strongly associated with their function limitation. Chinese elder’s SWB is related with living arrangements, socio-economic status, social support, self- rated health, and age-related functioning (Deng, et al., 2010; Gao, Jin & Unverzagt, et al., 2009; Yip, et al., 2007; Zhang, & Liu, 2007; Zhang, et al., 2008). Some studies suggest family-based support is central for improving SWB among the Chinese elderly (Cheng, et al., 2009; Phillips et al., 2008; Shen and Yeatts, 2013; Siu, & Phillips, 2002; Weng, 1998), and the Chinese family traditions that involve high levels of social support which are protective for depression (Chen, Copeland, & Wei, 1999; Chen, et al., 2005). Like other age groups, elderly Chinese living alone is more likely to report higher levels of depressive symptoms (Chen & Short, 2008; Chou & Chi, 2000; Chou, Ho & Chi, 2006). Silverstein, Cong, and Li (2006) have reported that traditional family arrangements (living with children) in China are beneficial, as elderly people’s stronger emotional cohesion and tighter intergenerational exchange with children will improve their psychological well-being (Chen & Silverstein, 2000). Chen and Short (2008) also found that co-residence with immediate family is positively associated with subjective well-being among oldest-old in rural China. According to the previous research, a positive correlation is predicted between living with descendants and better SWB. The crosstab result (Graph 1 & Graph 2 in Introduction) that SWB is lower among elderly Chinese living with descendants after 2005 is contradictory with what previous studies have found. As the literature shows, some factors are both associated with living arrangements and SWB, such as self- rated health, functional limitation and SES. The negative association may be caused by selection
SLIDE 4
4
bias due to these confounders. For example, disabled elders tend to live with descendants for care and support, and the disability may also lower SWB. This study aims to control for the confounders and examine whether living with descendants causes lower SWB (as the crosstab shows), high SWB (as the literature predicts), or no significant effects. Data, Measures and Methods Data Data for this study come from two nationally representative surveys - China Health and Retirement Longitudinal Study (CHARLS) in 2011(baseline) and 2013 (wave 2), and Chinese Longitudinal Healthy Longevity Survey (CLHLS) in 2002, 2005 and 2008. The CHARLS is a household survey
- f Chinese adults above 45 years old. The sample size for the CHARLS in 2011 is 17,708 individuals
and 10,287 households, and 18605 individuals in 2013 including 15570 individuals re-interviewed, and available spouses in the same household were both interviewed. The sample in this study is restricted to older adults above 60 years old1, and the robust cluster estimator is used to address the violation of stable unit treatment value assumption (SUTVA) among couples, as they live together and will influence each other’s SWB. The CLHLS was originally designed to survey the oldest-old population from 1998 (baseline), and the 2002 wave firstly expanded to ages 65+, with a total sample size of 16,057 elders. I have combined the 2002 wave with two following waves in 2005 (15,613 elders) and 2008 (16,540 elders) for panel data analyses. Since many elders deceased and new respondents were added every wave, the elderly deaths could be another confounder and make panel data unbalanced. I restrict the sample to respondents who have been interviewed at least twice in all three waves, and show another analysis of surviving respondents over all three waves in Appendix
- D. Both datasets provide relevant information for this study, including demographics, family
information, living arrangements, health status, functional limitation, work and retirement, and SWB questions etc. To control for the influence of childlessness on both living arrangement and SWB, I restrict both samples to elderly Chinese having living children. Measures The outcome variables subjective well-being (SWB) are predicted by the factor analyses of responses to different SWB questions in two surveys. In CHARLS, eight depressive symptoms questions as a different SWB outcome measures to examine the similar effects, such as the frequencies of feeling depressed, feeling fearful and restless sleep, etc. The Cronbach's alpha is 0.6863. There are three questions more similar to CLHLS’s questions (feeling hopeful, being happy
1 60 is used as a cutoff to refer to Chinese old population, because currently civil servants’ retirement age in China is 60 for men and
55 for women. (From State Council Circular on Issuing "State Council Temporary Measures on Providing for Old, Weak, Sick, and Handicapped Cadres" and "State Council Temporary Measures on Workers' Retirement, Resignation" (Chinese Text) http://www.cecc.gov/resources/legal-provisions/state-council-circular-on-issuing-state-council-temporary-measures-on)
SLIDE 5
5
and life satisfaction), but not as closely related as the depressive symptoms with Cronbach’s alpha 0.5644. Additional analyses compare the two outcome measures. In CLHLS, six SWB questions (e.g. life ratings and happiness) are used, and the Cronbach’s alpha is 0.8090. Both outcome variables have only one leading factor with eigenvalue over two, and they are standardized continuous variable with mean equals to zero, and standard deviation equals to one. The outcome variable in CHARLS measures depressive symptoms, and a higher value means that respondents have more depressive symptoms, whereas the outcome variable in CLHLS measures SWB, and a higher value means better
- SWB. The details of these questions are listed in the Appendix A.
The treatment variable is this study is a binary variable measuring whether respondents having living children2 co-reside with their descendants or not. The descendants include children (and/or spouses) and grandchildren (and/or spouses). Since there is no direct question about co-residence with descendants in CHARLS, I use family relationship information in the household roster to generate the variable. The control variables are selected based on prior literature regarding the most influential factors determining both living arrangements and SWB. The common variables in both two surveys include gender, age, education (categorical in the CHARLS; continuous - years of schooling in the CLHLS), pension ownership (dummy - yes or no), marital status (categorical variable - married, separated/divorced, widowed or never married 3 ), residence (rural/urban in the CHARLS; village/town/city in the CLHLS), self-rated health (continuous), experience of child/children loss (dead) (dummy - yes or no), functional limitation including activities of daily life (ADL) and instrumental activities of daily life (IADL). The functional limitation variables ADL and IADL are generated through factor analyses of all questions regarding the difficulties of ADL (e.g. dressing, bathing and eating, etc.) and IADL (e.g. shopping and cooking, etc.). All functional limitation variables are continuous variables on a standardized scale with Cronbach's alpha over 0.8. A larger value of functional limitation variables means more limitation and difficulties of doing activities of daily life. Questions about ADL and IADL are different in the CHARLS and the CLHLS, and detailed questions are shown in Appendix. B. Ethnicity variable (dummy - Han or Minority) is included in the CLHLS, but not available in the CHARLS. Since the CHARLS is a more comprehensive survey with more questions that are relevant to this study, I also include hukou (categorical - agricultural, non-agricultural, or other), disability (dummy
- yes or no), other health problems (dummy - yes or no) including any of cancer or malignant tumor,
disease, stroke, memory-related disease or emotional, nervous or psychiatric problems. Both surveys have questions about living arrangement preference. In the CHARLS, two questions about living arrangement preferences for either individuals having a spouse or not with same five answers - living
2 Both surveys do not have information about the live status of respondents’ grandchildren.
3The sample of being never married is too small in the CHARLS dataset to give any estimation of its coefficients. Thus
this category is not shown in the results.
SLIDE 6
6
with adult children in the same household (1), not living with adult children in the same household but in the same village or community (2), not in the same community (3), living in a nursing institute (4), or other (5). These two questions are both coded as categorical variables in the analysis. In the CLHLS, there is a similar question in wave 2005 and 2008, not available in wave 2002, thus the preference variable is not included in the analysis. I also include another variable measuring expectation of source of support (dummy - from children, or from other sources including savings, pension or insurance) among elderly. Details of these questions are available in Appendix B. Methods I firstly use propensity score matching techniques to address the selectivity issue using CHARLS 2011 cross-sectional data. For example, the functional limitation may increase the likelihood of living with descendants and having more depressive symptoms. Without controlling for the functional limitation, potential positive effects (if any) of living with descendants on mental health will be deflated. Thus confounders (e.g. self-reported, health, pension, etc.) may either deflate or inflate the outcome effects. Logistic regression models using different weights are firstly built to predict the selectivity process that determines living with descendants by using all the covariates mentioned above except ethnicity. To address the violation of stable unit treatment value assumption (SUTVA) among couples, I use the robust cluster estimator in the logistic regression. The model is used to estimate propensity score (the conditional probability of treatment given a set of observed covariates) to create a “randomly assigned” treatment (living with descendants) and control (not living descendants), and then use kernel matching and nearest neighbor matching techniques to estimate the average treatment effect between outcomes of these two groups. The important feature
- f matching is that comparable treatment and control groups are balanced on a set of covariates. That
means individuals with same propensity score should have the same distribution of observed covariates (Rosenbaum, 2010; Rosenbaum and Rubin, 1983). An examination of Common region of support and balance tests will be shown in Appendix C. Secondly, I use a weighted regression approach where the ATE weights are derived from the propensity score (p). For subjects in the treatment group, this weight is equal to 1/p and for subjects in the control group the weight is equal to 1/(1-p) (Morgan, & Winship, 2014). CHALRS survey weights (individual weights with both household and individual non-responsement adjustments) are used in a second model. Since the survey weights have some extreme values that could cause inaccurate results, I also show a third model with 95%-trimmed4 survey weights. Appendix C. shows a comparison of the untrimmed survey weights and the trimmed survey weights. Doubly robust regression models are used to estimate the causal effects of living with descendants on depressive symptoms.
4 I top the weights to 95% and trim the weights above to 95%.
SLIDE 7
7
It is possible that there is no causal effect of living with descendants on SWB, but rather that some
- f the same individual characteristics that both influence the treatment and outcome variables are
- unobserved. Datasets lack measures of many relevant characteristics, such as descendants’
personality and their living arrangement preferences, etc. I combine the inclusion of measured control variables with person-specific fixed-effects modeling to deal with heterogeneity on time- invariant unmeasured characteristics. Fixed-effects regression models require panel data that measure variables at least two points in time, thus I include those interviewed twice (2011 and 2013) in the CHARLS data and only include respondents that have records in at least two waves among three waves (2002, 2005 and 2008) in the CLHLS. Since the fixed effects model cannot estimate the coefficients for time-invariant variables using conditional likelihood methods, I use Allison’s hybrid model that combines the fixed effects and random effects approaches by decomposing each time- varying predictor into a within-person component (deviation from that person-specific mean) and a between-person component, and then fitting a random effects with both components (Allison, 2009). Because the CLHLS datasets are very unbalanced due to the deceased people and newly added respondents, I employ both models to examine the robustness of the estimation. Results What are the factors associated with Chinese elders’ living arrangements? Table 1 shows the results of three logistic regression models that are not weighted, weighted by untrimmed survey weights, and weighted by 95%-trimmed weights respectively. The models are used to predict the propensity score that establishes the likelihood of living with descendants. The models also illustrate the factors associated with Chinese elders’ living arrangements among those who have living children. The odds that an elderly Chinese living with descendants is about 1.6 times higher in urban areas than in rural areas. This might be resulted from large numbers of rural-to-urban migrants at working age and housing price versus income in urban areas. For an elderly Chinese with non-agricultural hukou (household registration system), the odds of living with descendants is about two thirds of those with agricultural hukou. Different from the place of residence, hukou system specifies certain benefits for certain hukou holders - an agricultural hukou provides access to farmland and a nonagricultural hukou provides access to jobs, housing, food, and state-sponsored benefits including urban health insurance (Fan, 2008). Thus, the negative association might suggest that the elderly having a non-agricultural hukou need less support from their descendants. Having pension is also significantly correlated with less likelihood of co-residing with descendants in the first model, but not statistically significant once the survey weights are employed. As the source of support variable indicates, Chinese elders who reported the expectation of main support from children rather than other sources (savings, pension or insurance, etc.) are more likely to live with their descendants. Furthermore, the living arrangement preferences are significantly correlated with actual living
- arrangements. The odds for the elderly Chinese who prefer not living with children in the same
SLIDE 8
8
household (same community or not) to live with their descendants is almost half as large as those who prefer the co-residence with children. The odds are generally smaller in the first question assuming the choice of individuals with spouses than the second question assuming the choice of individuals without spouses. The first question is more appropriate for most respondents living with their spouses in CHALRS survey, and the result suggests that the first question has largest effects. In addition, widowhood and strong functional limitation of IADL predict higher likelihood of living with descendants, as they might need more family support. Females and older age are associated with less likelihood of co-residence with descendants. Although the odds of functional limitation of ADL are close to one, ADL and IADL show opposite directions of correlation. As I group all grandchildren and children under descendant’s category, the correlations of living with children or grandchildren are entangled. The elderly with functional limitation to ADL (e.g. dressing, eating, toileting, etc.) need more support and cannot provide support for their children, such as taking care
- f their grandchildren, housework, etc. These coefficients with statistical significance are hard to
interpret due to the complexity of intergenerational transfer and communication. In general, the logistic regression models indicate that Chinese elders who have non-agricultural hukou, pension and other main social support sources rather than depending on children are less likely to co-reside with their descendants, whereas those who are widowed, have functional limitation to IADL and strong preferences of co-residence with children tend to live with their descendants to get more family support. What is the causal relation between living with descendants and Chinese elders’ subjective well-being (SWB)? The main goal of this paper is to examine the causal relation between living arrangements and SWB. I firstly use the propensity score generated from the logistic regression models above to run kernel matching and nearest neighbor as Table 2 shows. For the unmatched group, the average treatment effect is 0.0556, which means living with descendants will increase depressive symptoms by around 6% of standard deviation. Using kernel matching, the treatment effect is reduced to almost zero. Using propensity score predicted with survey weights, the treatment effects show a negative sign, which indicates co-residence with descendants may reduce depressive symptoms. Nearest neighbor matching also reduce a large amount of treatment effects by controlling the confounders, but still show positive effects. I also include matching within the region of common support, only a few cases are dropped. The treatment and control groups created are generally balanced with biases mostly less than 2%. Details and balance tests are shown in Appendix C. Generally, propensity score matching suggests that selection biases cause a large proportion of living arrangements’ negative effects on depressive symptoms.
SLIDE 9
9
TABLE 1. Logistic regression models predicting living with descendants (N =6,955 from CHARLS 2011)
Odds Ratio Robust S.E. Odds Ratio (svy weights) Robust S.E. Odds Ratio (Trimmed svy weights) Robust S.E. Age (60+) 0.950*** 0.005 0.962*** 0.006 0.960*** 0.005 Gender (Male-0, Female-1) 0.825*** 0.033 0.844** 0.042 0.852*** 0.038 Residence (Rural -0, Urban-1) 1.652*** 0.191 1.660** 0.271 1.596*** 0.204 Hukou (Agricultural -1) Non-agricultural (2) 0.679** 0.078 0.667* 0.107 0.678** 0.088 Other (3) 0.635 0.236 1.116 0.508 0.903** 0.366 Marital (Married -1) Separated/Divorced (2) 1.003 0.245 0.961 0.256 0.970 0.256 Widowed(3) 1.787*** 0.143 2.065*** 0.197 1.997*** 0.176 Education (Illiterate -1) Somewhat elementary (2) 0.908 0.068 0.898 0.082 0.920 0.076 Elementary school (3) 0.875 0.068 0.864 0.082 0.868 0.073 Middle school (4) 0.832 0.082 0.896 0.112 0.885 0.096 High school and above (5) 0.930 0.125 0.969 0.164 0.930 0.131 Self-reported health (Poor-1, Fair-2, Good-3) 0.976 0.030 0.953 0.036 0.969 0.032 Disability (No-0, Yes-1) 0.975 0.066 0.946 0.077 0.943 0.071 Other health problems (No-0, Yes-1) 0.944 0.062 0.905 0.073 0.898 0.065 Deceased child (No-0, Yes-1) 1.055 0.100 1.082 0.122 1.090 0.110 Functional limitation ADL 0.919* 0.037 0.908* 0.042 0.906* 0.040 IADL 1.230*** 0.052 1.276*** 0.063 1.283*** 0.060 Pension (No-0, Yes-1) 0.837* 0.068 0.903 0.084 0.873 0.076 Source of support (Other-0, From children-1) 1.237** 0.097 1.271* 0.118 1.247* 0.107 Living arrangement preference with spouse (Co-residence with children -1) Not same household, but same community (2) 0.452*** 0.040 0.424*** 0.047 0.427*** 0.041 Not same household or community (3) 0.410*** 0.080 0.397*** 0.087 0.414*** 0.089 Nursing institution (4) 0.624 0.174 0.662 0.212 0.659 0.207 Other (5) 0.492* 0.174 0.486 0.190 0.506 0.197 Living arrangement preference without spouse (Co-residence with children -1) Not same household, but same community (2) 0.684*** 0.065 0.671** 0.078 0.676*** 0.070 Not same household or community (3) 0.509** 0.132 0.472* 0.138 0.476** 0.136 Nursing institution (4) 0.677 0.148 0.617 0.160 0.635 0.157 Other (5) 0.571 0.195 0.569 0.204 0.575 0.205 p***<0.001 p**<0.01 p*<0.05
SLIDE 10
10
TABLE 2. Propensity score matching results of depressive symptoms outcomes (CHARLS, 2011)
Treated Controls Difference (ATT effects) S.E. Not weighted Unmatched (N=6,355) 0.0162
- 0.0394
0.0556 0.0250 Matched (Kernel ) (N=6,355) 0.0162 0.0140 0.0067 0.0012 Matched (Kernel within the region of common support) (N=6,349) 0.0157 0.0130 0.0028 0.0282 Matched (5-Nearest Neighbor) (N=6,355) 0.0162 0.0028 0.0133 0.0307 Matched (5-Nearest Neighbor within the region of common support) (N=6,349) 0.0157 0.0009 0.0149 0.0307 Weighted by the untrimmed survey weights Unmatched (N=6,355) 0.0162
- 0.0394
0.0556 0.0250 Matched (Kernel-based) (N=6,355) 0.0162 0.0236
- 0.0074
0.0282 Matched (Kernel-based within the region of common support) (N=6,353) 0.0159 0.0238
- 0.0079
0.0282 Matched (5-Nearest Neighbor) (N=6,355) 0.0162 0.0075 0.0087 0.0305 Matched (5-Nearest Neighbor within the region of common support) (N=6,353) 0.0159 0.0078 0.0081 0.0305 Weighted by the 95%-trimmed survey weights Unmatched (N=6,355) 0.0162
- 0.0394
0.0556 0.0250 Matched (Kernel -based) (N=6,355) 0.0162 0.0251
- 0.0089
0.0282 Matched (Kernel-based within the region of common support) (N=6,353) 0.0159 0.0253
- 0.0094
0.0282 Matched (5-Nearest Neighbor) (N=6,355) 0.0162 0.0091 0.00714 0.0303 Matched (5-Nearest Neighbor within the region of common support) (N=6,353) 0.0159 0.0093 0.0065 0.0303
Then I use doubly-robust regression models to address effect heterogeneity issues by using average treatment effects (ATE) weights from the same propensity scores used in the matching. Survey weights are incorporated into the second and the third model shown in Table 3. All three models using different weights indicate that the hypothesis that living with descendants has no causal effects
- n depressive symptoms cannot be rejected.
Since propensity score matching and weighted regression can only account for observed covariates, unobserved variables may cause hidden bias remained after matching and regression. Finally I use the fixed effects model and Allison’s hybrid model to investigate causal effects using panel data (two
SLIDE 11
11
waves of CHARLS and three waves of CLHLS) to control for time-invariant unobserved variables. Table 4 using CLHLS shows co-residence with descendants can significantly increase Chinese elders’ SWB by around 10% of standard deviation, as the prior literature predicts. I examine both the potential endogenous issue and the common time trends assumption. The living arrangements in later waves are not associated with SWB in prior waves, and treatment group (co-residence with children) and control group have common time trends. Therefore, both assumptions are not violated. However, Table 5 using hybrid model from CHARLS data doesn’t suggest the same significant effects of living with descendants, which is similar to the results of weighted regression. In sum, my causal inference generally supports the prior literature that living with descendants tend to have positive influence on Chinese elders’ SWB or at least no statically significant negative effects are found to be associated with living with descendants. The propensity matching removes part of selection bias, and weighted regression and hybrid models using CHARLS data cannot reject the null hypothesis of nonexistent causal effects. By controlling for time-invariant unobserved variables, both the fixed effects model and the hybrid model using CLHLS show that residence with descendants has significantly positive effects on SWB. What are the factors associated with Chinese elders’ subjective well-being? The doubly-robust regression models, the fixed effects model and the hybrid models all indicate that depressive symptoms are strongly associated with physical health. The Chinese elderly with better self-rated health conditions are more likely to report less depressive symptoms and better SWB. More functional limitation of both ADL and IADL is strongly associated more depressive symptoms and worse SWB. Disability and other health problems (e.g. stroke, cancer, etc.) also predict more depressive symptoms. In the fixed effects model and the hybrid model, I add an interaction term between self-reported health and wave to see the time trend of health conditions. The significance
- f the interaction of wave 2 indicates that in 2005 Chinese elders’ self-reported health was
significantly improved from 2002. On the contrary, the hybrid model shows that SWB among Chinese elderly was significantly decreasing over the three waves (-0.248 from 2002 to 2005, and - 0.305 from 2005 to 2008). This result suggests that the physical health of the Chinese elderly was improving, whereas their mental health was deteriorating with lower SWB. More studies are needed to investigate more about the time trends of Chinese elders’ mental health. Both weighted regression models and hybrid model present that the female Chines elderly are more likely to have poor SWB and more depressive symptoms than males. This could be one indicator of gender inequality in terms of mental health. The Chinese elderly women may be trapped by their traditional gender roles and experience dramatic changes of social attitudes, or they might have more adverse experiences than men because of gender disparities. The analysis above has found females are also less likely to live with their descendants. The interaction between gender and living arrangements should be considered in the future research. Marital status, particularly widowhood, is strongly associated with SWB. The Chinese elders who are separated, divorced or widowed are more
SLIDE 12
12
likely to report depressive symptoms and lower SWB. More social resources and higher SES, including majority ethnicity (Han), pension, urban (town/city) residence, non-agricultural hukou and better education, are significantly associated with less depressive symptoms and/or higher SWB. The experience of child/children loss is a statistically significant factor predicting more depressive symptoms in weighted regression models, but not significant in the panel data analysis. Living arrangement preferences are not significantly associated with depressive symptoms, but the expectation that main source of support is from children can significantly decrease the likelihood of having depressive symptoms. Based on the analyses of both cross-sectional data and longitudinal data, the most important factors associated with SWB are physical health (disability, functional limitation, disease, etc), gender, marital status and SES (residence, hukou, education, ethnicity, pension, etc.). The female Chinese elders with poor health, more functional limitation and less SES are the most vulnerable individuals who tend to have more depressive symptoms and worse SWB. More social support should be provided to help this group, especially those who do not live with their descendants in the lack of family support.
Discussion
This study employs propensity score matching, weighted regression and fixed effects models to examine the causal relations between living arrangements (co-residence with descendants among elderly Chinese having living children) and subjective well-being (depressive symptoms in the CHARLS data analysis and SWB in the CLHLS data analysis). The findings suggest that the negative correlation between living with descendants and lower SWB is spurious due to the selectivity. The first logistic regression models report that Chinese elders are more likely to live with descendants if they have fewer social support resources (e.g. agricultural hukou, no pension, widowed), have difficulties doing IADL, and have strong preference of co-residence with
- children. As the doubly-robust regression models and the hybrid model show, these factors associated with
living arrangements are also associated with SWB, such as functional limitation, gender, marital status and
- SES. The positive causal effects on mental health tend to be deflated by most of these confounders. My causal
analyses generally support the positive association found by the previous research, and examine the causal
- effects. By controlling for all the observed confounders, the null hypothesis of no causal effects cannot be
rejected in the weighted regression model and hybrid model using CHARLS data, and both the fixed effects model and Allison’s hybrid model using CLHLS data show the positive causal effects of co-residence with descendants on SWB after controlling for time-invariant unobserved variables that could not be controlled in matching and regression.
SLIDE 13
13
TABLE 3. Doubly-robust regression models predicting depressive symptoms (CHARLS, 2011)
Depressive Symptoms Only ATE weights Robust S.E. Incorporating Survey Weights Robust S.E. Incorporating Trimmed survey weights Robust S.E. Living with descendants (No-0, Yes-1) 0.006 0.026
- 0.006
0.028
- 0.009
0.025 Age (60+)
- 0.014***
0.002
- 0.016***
0.002
- 0.014***
0.002 Gender (Male-0, Female-1) 0.225*** 0.024 0.235*** 0.027 0.225*** 0.028 Residence (Rural -0, Urban-1)
- 0.152***
0.044
- 0.195***
0.047
- 0.164***
0.044 Hukou (Agricultural -1) Non-agricultural (2)
- 0.211***
0.047
- 0.123*
0.053
- 0.153**
0.048 Other (3)
- 0.280*
0.125*
- 0.228
0.167
- 0.185
0.132 Marital (Married -1) Separated/Divorced (2) 0.585*** 0.124 0.502*** 0.129 0.499*** 0.129 Widowed(3) 0.191*** 0.038 0.178*** 0.043 0.184*** 0.041 Education (Illiterate -1) Somewhat elementary (2)
- 0.001
0.036
- 0.030
0.039
- 0.026
0.038 Elementary school (3)
- 0.094*
0.035
- 0.117**
0.041
- 0.099**
0.038 Middle school (4)
- 0.178***
0.045
- 0.222***
0.049
- 0.207***
0.046 High school and above (5)
- 0.256***
0.050
- 0.248***
0.056
- 0.243***
0.052 Self-reported health (Poor-1, Fair-2, Good-3)
- 0.169***
0.014
- 0.176***
0.015
- 0.178***
0.014 Disability (No-0, Yes-1) 0.160*** 0.033 0.182*** 0.034 0.175*** 0.034 Other health problems (No-0, Yes-1) 0.195*** 0.032 0.175*** 0.035 0.172*** 0.034 Deceased child (No-0, Yes-1) 0.126** 0.040 0.123** 0.043 0.131*** 0.039 Functional limitation ADL 0.170*** 0.026 0.168*** 0.027 0.167*** 0.027 IADL 0.196*** 0.021 0.187*** 0.022 0.191*** 0.022 Pension (No-0, Yes-1)
- 0.158***
0.031
- 0.175***
0.032
- 0.170***
0.029 Source of support (Other-0, From children-1)
- 0.129***
0.036
- 0.100*
0.040
- 0.097**
0.036 Living arrangement preference with spouse (Co-residence with children -1) Not same household, but same community (2) 0.029 0.034 0.031 0.037 0.035 0.032 Not same household or community (3) 0.162* 0.073 0.092 0.077 0.096 0.069 Nursing institution (4)
- 0.016
0.105
- 0.023
0.106
- 0.032
0.093 Other (5)
- 0.046
0.143 0.010 0.120 0.006 0.118 Living arrangement preference without spouse (Co-residence with children -1) Not same household, but same community (2) 0.024 0.036
- 0.011
0.038
- 0.003
0.033
SLIDE 14
14
Not same household or community (3)
- 0.145
0.107
- 0.093
0.104
- 0.094
0.104 Nursing institution (4) 0.064 0.085 0.034 0.086 0.045 0.075 Other (5)
- 0.108
0.125
- 0.141
0.111
- 0.141
0.109 N 6,355 6,355 6,355 p***<0.001 p**<0.01 p*<0.05
TABLE 4 The fixed effects model and Allison’s hybrid model predicting subjective well-being (CLHLS)
Subjective well-being Fixed effects model S.E. Hybrid model S.E. Living with descendants (No-0, Yes-1) 0.102*** 0.021 0.111*** 0.021 Age (65+) 0.017 0.025 0.006 0.006 Gender (Male-0, Female-1) 0.000 0.124
- 0.054***
0.014 Ethnicity (Minority-0, Han-1)
- 0.009
0.072 0.058* 0.025 Residence (Country - 1) Town (2) 0.002 0.025 0.043** 0.015 City (3) 0.033 0.047 0.190*** 0.018 Education (years of schooling)
- 0.004
0.303
- 0.011
0.302 Marital (Married -1) Divorced (2)
- 0.131
0.147
- 0.086
0.142 Widowed (3)
- 0.250***
0.030
- 0.244***
0.029 Never married (4)
- 0.120
0.548
- 0.021
0.319 Deceased child (No-0, Yes-1) 0.000 0.020 0.006 0.019 Pension (No-0, Yes-1) 0.114** 0.041 0.113** 0.039 Functional limitation ADL
- 0.037**
0.013
- 0.029*
0.011 IADL
- 0.121***
0.014
- 0.117***
0.013 Self-reported health (Very bad-1, Bad -2, Fair-3, Good-4, Very good-5) 0.342*** 0.015 0.366*** 0.009 Wave (2002 -1) 2005 (2)
- 0.219*
0.098
- 0.248***
0.065 2008 (3)
- 0.276
0.169
- 0.305***
0.079 Interaction -wave##health (2002-1) 2005 (2) 0.048** 0.018 0.067*** 0.018 2008 (3) 0.017 0.020 0.044* 0.021 N 22,920 22,920 Individuals 10,796 10,796 p***<0.001 p**<0.01 p*<0.05
SLIDE 15
15
TABLE 5 Allison’s hybrid model predicting depressive symptoms and subjective well-being (CHARLS)
Depressive Symptoms (more) S.E. Subjective Wellbeing(better) S.E. Living with descendants (No-0, Yes-1)
- 0.043
0.023
- 0.003
0.024 Age (60+)
- 0.016***
0.002 0.008*** 0.002 Gender (Male-0, Female-1) 0.246*** 0.025
- 0.025
0.023 Graduated Elementary School (No-0, Yes-1)
- 0.079**
0.029 0.117*** 0.030 Residence (Rural – 0, Urban- 1)
- 0.104**
0.037 0.103** 0.039 Hukou (Agricultural -0, Non-agricultural-1)
- 0.151***
0.035 0.101** 0.037 Marital (Married -1) Separated/Divorced (2) 0.507*** 0.116
- 0.473***
0.125 Widowed (3) 0.176*** 0.029
- 0.054
0.031 Disabled (No-0, Yes-1 0.298*** 0.039
- 0.178***
0.041 Deceased child (No-0, Yes-1) 0.210*** 0.034
- 0.146***
0.036 Pension (No-0, Yes-1) 0.076 0.043
- 0.079
0.044 Functional limitation ADL 0.183*** 0.020
- 0.100***
0.021 IADL 0.152*** 0.018
- 0.064**
0.020 Self-reported health (Poor-1, Fair-2, Good-3) 0.240*** 0.016
- 0.208***
0.016 Wave (2011 -1) 2012 (2) 0.042 0.031
- 0.102**
0.036 N 10,868 10381 Individuals 6,121 5958 p***<0.001 p**<0.01 p*<0.05
Since I use different datasets (the CHARLS 2011 & 2013, the CLHLS 2002, 2005 & 2008), different
- utcome variables (depressive symptoms and SWB) and different methods (matching and weighted
regression and fixed effects modeling) to study the same question, the comparisons between cross- sectional data analyses and longitudinal data analyses are limited and should be carefully interpreted. For instance, it is possible that co-residence with children has no causal effects on depressive symptoms, but have positive effects on SBW that measures more about positive attitudes. I examined this by using the subjective wellbeing measure created through three questions in CHARLS data, and the results are similar (presented in Table 5). Many variables that are relevant to this causal questions are not available in the datasets, such as communication and closeness between the elderly and their children, children’s personality and the elderly’s social network, etc. Some of these unobserved variables are time-varying, and cannot be controlled by fixed effects models. To find out how robust the results are, I incorporate survey weights and 95%-trimmed survey weights in the
SLIDE 16
16
cross-sectional data analysis, and both the fixed effects model and the hybrid model in the panel data
- analysis. The results are similar, except the fixed effects model cannot measure the coefficients of
time-invariant variables. Since deaths could be a confounder in my panel data analysis and the deceased respondents make the datasets very unbalanced, I include another panel data analysis on respondents who survived all three waves to see whether the effects of deaths may change my results in Appendix D. Both the fixed effects model and the hybrid model show similar results with smaller coefficients of treatment, which may be caused by the small sample size and the selectivity of those survivors who have lower risks (or frailty). Another limitation mentioned before is my treatment variable, grouping all descendants together confounds the intricate relations between the Chinese elderly and their children. For instance, they might help their adult children to take care of their grandchildren. Future research should decompose the treatment variable to study the complexity of intergenerational relations. Also, the intergenerational transfer (the direction and scale of wealth and care flow) is not included in this study, but is also important to understand how Chinese elders make decisions about their living arrangements and how this will influence their mental health. As the number of empty-nest household is still growing (Yi, et al, 2008), and the general subjective well- being of Chinese elders is getting worse, what policy or intervention should be made to improve this increasing population’s subjective well-being? What living arrangement for elderly people in an aging society is ideal? On the other hand, what can we do to help the Chinese elders cope with the rapid change of living arrangements? The factors, such as SES, that both affect living arrangements and mental health should be taken into consideration. Improving universal healthcare coverage, pension coverage may empower the elderly to be less dependent on their children, but the public wealth transfer to the elderly Chinese could be more difficult in an aging society with low fertility. Other options are providing education and employment opportunities for the elderly, building better community support and balancing resources distribution in rural and urban area (more investments in rural area). None of the interventions mentioned above is easy to implement, more research is needed to find out the effective interventions.
SLIDE 17
17
Bibliography
Allison, P. D. (2009). Fixed effects regression models (Vol. 160). SAGE publications. 28-48. Chen, B. (2009). Change of intergenerational relations and the elderly suicide. Sociological Research. (4), 157-176. (Chinese Text) 陈柏峰. (2009). 代际关系变动与老年人自杀. 社会学研究, (4), 157-176.. Chen, F., & Short, S. E. (2008). Household context and subjective well-being among the oldest old in
- China. Journal of family issues.
Chen, R., Wei, L., Hu, Z., Qin, X., Copeland, J. R., & Hemingway, H. (2005). Depression in older people in rural China. Archives of Internal Medicine,165(17), 2019-2025. Chen, R., Copeland, J. R. M., & Wei, L. (1999). A meta‐ analysis of epidemiological studies in depression
- f older people in the People's Republic of China. International journal of geriatric psychiatry, 14(10), 821-
830. Chen, X., & Silverstein, M. (2000). Intergenerational social support and the psychological well-being of
- lder parents in China. Research on aging, 22(1), 43-65.
Cheng, S. T., Lee, C. K., Chan, A. C., Leung, E. M., & Lee, J. J. (2009). Social network types and subjective well-being in Chinese older adults. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, gbp075. Chou, K. L., & Chi, I. (2000). Comparison between elderly Chinese living alone and those living with
- thers. Journal of Gerontological Social Work,33(4), 51-66.
Chou, K. L., Ho, A. H. Y., & Chi, I. (2006). Living alone and depression in Chinese older adults. Aging and Mental Health, 10(6), 583-591. Deng, J., Hu, J., Wu, W., Dong, B., & Wu, H. (2010). Subjective well‐ being, social support, and age‐ related functioning among the very old in China. International journal of geriatric psychiatry, 25(7), 697-703. Fan, C. C. (2008). Migration, hukou, and the city. China urbanizes: Consequences, strategies, and policies, 65-89. Gao, S., Jin, Y., Unverzagt, F. W., et al. (2009). Correlates of depressive symptoms in rural elderly Chinese. International journal of geriatric psychiatry, 24(12), 1358-1366. Jing, J., Zhang, J., & Wu, X. (2011). Suicide among the elderly people in urban China. Demographic Research, (3), 84-95. (Chinese Text) 景军, 张杰, & 吴学雅. (2011). 中国城市老人自杀问题分析. 人口研究, 3, 84-95. Li, X., Xiao, Z., & Xiao, S. (2009). Suicide among the elderly in mainland China. Psychogeriatrics, 9(2), 62-66. Liu, C. Y., Wang, S. J., Teng, E. L., et al. (1997). Depressive disorders among older residents in a Chinese rural community. Psychological medicine, 27(04), 943-949. Liu, Y. (2011). Suicide in rural China (1980 - 2009) - Discussion with Dr. Jing. Youth Studies, (6), 72-82. (Chinese Text) 刘燕舞. (2011). 中国农村的自杀问题 (1980—2009)——兼与景军先生等商榷. 青年研究, (6), 72-82. Morgan, S. L., & Winship, C. (2014). Counterfactuals and causal inference. Cambridge University Press. 188-224. Phillips, D. R., Siu, O. L., Yeh, A. G., & Cheng, K. H. (2008). Informal social support and older persons’ psychological well-being in Hong Kong. Journal of cross-cultural gerontology, 23(1), 39-55. Rosenbaum, P.R., (2010). Design of Observational Studies. Springer, New York. Rosenbaum, P.R., & Rubin, D.B., (1983). The central role of the propensity score in
- bservational studies for causal effects. Biometrika 70, 41e55.
SLIDE 18
18
Sereny, M. (2011). Living arrangements of older adults in China: The interplay among preferences, realities, and health. Research on Aging, 33(2), 172-204. Shen, Y., & Yeatts, D. E. (2013). Social support and life satisfaction among older adults in China: family- based support versus community-based support. The International Journal of Aging and Human Development, 77(3), 189-209. Silverstein, M., Cong, Z., & Li, S. (2006). Intergenerational transfers and living arrangements of older people in rural China: Consequences for psychological well-being. The Journals of Gerontology Series B: Psychological Sciences and Social Sciences, 61(5), S256-S266. Siu, O. L., & Phillips, D. R. (2002). A study of family support, friendship, and psychological well-being among older women in Hong Kong. The International Journal of Aging and Human Development, 55(4), 299-319. Weng, B. K. (1998). Social network and subjective well-being of the elderly in Hong Kong. Asia Pacific Journal of Social Work and Development, 8(2), 5-15. Yi, Z., Wang, Z., Jiang, L., & Gu, D. (2008). Future trend of family households and elderly living arrangement in China. Genus, 64(1-2), 9-36. Yip, W., Subramanian, S. V., Mitchell, A. D., Lee, D. T., Wang, J., & Kawachi, I. (2007). Does social capital enhance health and well-being? Evidence from rural China. Social science & medicine, 64(1), 35-49. Zhang, J. P., Huang, H. S., Ye, M., & Zeng, H. (2008). Factors influencing the subjective well being (SWB) in a sample of older adults in an economically depressed area of China. Archives of gerontology and geriatrics, 46(3), 335-347. Zhang, W., & Liu, G. (2007). Childlessness, psychological well-being, and life satisfaction among the elderly in China. Journal of cross-cultural gerontology, 22(2), 185-203. Zimmer, Z. (2005). Health and living arrangement transitions among China’s oldest-old. Research on Aging, 27(5), 526-555.
SLIDE 19
19
- Appendix. A Subjective Well-being (SWB) Questions
The SWB outcome variables are standardized continuous variables created by the factor analyses in two surveys. The self-reported subjective wellbeing questions and answers in the CHARLS: DC013 I felt hopeful about the future. DC016 I was happy. DC028 Life’s satisfaction. The self-reported depressive symptoms questions and answers in the CHARLS: DC009 I was bothered by things that don’t usually bother me DC010 I had trouble keeping my mind on what I was doing. DC011 I felt depressed. DC012 I felt everything I did was an effort. DC014 I felt fearful. DC015 My sleep was restless. DC017 I felt lonely. DC018 I could not get ”going.” Same multiple choices for frequency of certain feelings within a week:
- 1. Rarely or none of the time ( < 1 day)
- 2. Some or a little of the time (1- 2 days)
- 3. Occasionally or a moderate amount of the time (3 - 4 days)
- 4. Most or all of the time (5-7 days)
The SWB questions and answers in the CLHLS B1.1 How do you rate your life at present? 1 very good 2 good 3 so-so 4 bad 5 very bad 8 not able to answer B2.1 Do you always look on the bright side of things? 1 always 2 often 3sometimes 4 seldom 5 never 8 not able to answer B2.7 Are you as happy as when you were younger? 1 happier and same 2 often 3 sometimes 4 seldom 5 never 8 not able to answer B2.3 Do you often feel fearful or anxious? 1 always 2 often 3sometimes 4 seldom 5 never 8 not able to answer B2.4 Do you often feel lonely and isolated? 1 always 2 often 3sometimes 4 seldom 5 never 8 not able to answer B2.6 Do you feel the older you get, the more useless you are? 1 always 2 often 3sometimes 4 seldom 5 never 8 not able to answer
SLIDE 20
20
- Appendix. B Control Variables Questions
Functional Limitation ADL questions in CHARLS DB010 Because of health and memory problems, do you have any difficulty with dressing? Dressing includes taking clothes out from a closet, putting them on, buttoning up, and fastening a belt. DB011 Because of health and memory problems, do you have any difficulty with bathing
- r showering?
DB012 Because of health and memory problems, do you have any difficulty with eating, such as cutting up your food? (Definition: By eating, we mean eating food by oneself when it is ready.) DB013 Do you have any difficulty with getting into or out of bed? DB014 Because of health and memory problems, do you have any difficulties with using the toilet, including getting up and down? DB015 Because of health and memory problems, do you have any difficulties with controlling urination and defecation? If you use a catheter (conduit) or a pouch by yourself, then you are not considered to have difficulties. Same multiple choices for above questions: (1) No, I don’t have any difficulty. (2) I have difficulty but can still do it. (3) Yes, I have difficulty and need help. (4) I cannot do it. ADL questions in CLHLS E1 Bathing – either sponge bath, tub bath, shower or washing the body. 1 receives no assistance (gets in and out of tub alone if tub is usual means of bathing) 2 receives assistance in bathing only for part of the body (such as back or a leg) 3 receives assistance in bathing more than one part of the body (or doesn’t bathe) E2 Dressing – gets clothes from closets and drawers – including underwear, outer garments and fasteners (including suspenders, if worn) 1 gets clothes and gets completely dressed without assistance 2 gets clothes and gets dressed without assistance except for tying shoes 3 receives assistance in getting clothes or in getting dressed, or stays partly or completely undressed E3 Toilet – going to the toilet; cleaning oneself afterwards 1 goes to the toilet, cleans self, and arranges clothes without assistance (may use object for support such as cane, walker, or wheelchair) 2 can partly manage on his/her own, and receives assistance in going to the toilet or in cleaning self or in arranging clothes afterwards or in use of night bedpan or commode 3 bedridden and needs complete assistance in use of night bedpan or commode in bed. E4 Indoor Transfer 1 gets in and out of bed as well as in and out of a chair without assistance (may use object for support such as cane or walker)
SLIDE 21
21
2 gets in and out of bed or chair with assistance 3 bedridden E5 Continence 1 has complete control of urination and bowel movement without assistance 2 has occasional ‘accidents’ 3 supervision helps keep urine or bowel control; catheter is used or elder is incontinent E6 Eating 1.feeds self without assistance 2 feeds self, with some help 3 receives assistance in feeding or is fed partly or completely intravenously IADL questions in CHARLS
- DB016. Because of health and memory problems, do you have any difficulties with doing household
chores?(Definition: By doing household chores, we mean house cleaning, doing dishes, making the bed, and arranging the house.) [IWER: If R cannot mop the floor, but can scrub, or R cannot fold heavy bedding, but is able to do light
- nes, then mark (3).]
- DB017. Because of health and memory problems, do you have any difficulties with preparing hot meals?
(Definition: By preparing hot meals, we mean preparing ingredients, cooking, and serving food.) [IWER: If another person prepares ingredients or if R can cook rice, but is not able to prepare side dishes, then mark (3).]
- DB018. Because of health and memory problems, do you have any difficulties with shopping for
groceries? By shopping, we mean deciding what to buy and paying for it.
- DB019. Because of health and memory problems, do you have any difficulties with managing your money,
such as paying your bills, keeping track of expenses, or managing assets?
- DB020. Because of health and memory problems, do you have any difficulties with taking medications? By
taking medications, we mean taking the right portion of medication right on time. Same multiple choices for above questions: (1) No, I don’t have any difficulty (2) I have difficulty but can still do it. (3) Yes, I have difficulty and need help. (4) I can not do it. IADL questions in CLHLS E7 Can you visit your neighbors by yourself? E8 Can you go shopping by yourself? E9 Can you cook a meal by yourself whenever necessary? E10 Can you wash clothing by yourself whenever necessary? E11 Can you walk continuously for 1 kilometer at a time by yourself?
SLIDE 22
22
E12 Can you lift a weight of 5kg, such as a heavy bag of groceries? E13 Can you continuously crouch and stand up three times? E14 Can you take public transportation by yourself? Same multiple choices for above questions: 1 yes, independently 2 yes, but need some help 3 no, can’t Living Arrangement Preferences Question in CHARLS:
- CG001. Suppose an elderly person has a spouse and adult children, and has good relationship with them
What do you think is the best living arrangement for the elderly person? (1)Live with adult children (2) Don’t live with them in the same house, but live in the same community or village. (3) Don’t live with them in the same house and the same community or village. (4) Live in a nursing house (5) Other
- CG002. Suppose an elderly person has no spouse but has adult children, and has good relationship with
- them. What do you think is the best living arrangement for him/her?
(1) Live with adult children (2) Don’t live with them in the same house, but live in the same community or village. (3) Don’t live with them in the same house and the same community or village. (4) Live in a nursing house (5) Other Living Arrangement Preferences Question in CLHLS:
- F16. Which living arrangement do you prefer?
1 living alone (or with spouse), no matter how far children live 2 living alone (or with spouse), but it is better that children live nearby 3 coresidence with children 4 institutions (elderly center, elderly home, etc.) 5 do not know Source of Support Question in CLHLS:
- FN080. Whom do you think you can rely on for old-age support？
(1) Children (2) Savings (3) Pension or retirement salary (4) Commercial pension insurance (5) Other
SLIDE 23
23
Appendix C The region of common support, balance tests, and survey weights. C.1 The region of common support is good with less than ten extreme cases on both ends in the treatment group. C.2 Balance tests
Table 5. Balance tests (Standardized biases (%) between treatment and control groups using different weights) Covariates ATE weights With untrimmed survey weights With trimmed survey weights Age (60+) 0.5
- 9.3
- 7.2
Gender (Male-0, Female-1) 1.9 0.1
- 0.0
Residence (Rural -0, Urban-1)
- 0.8
- 0.8
0.8 Hukou (Agricultural -1) Non-agricultural (2)
- 2.3
- 1.9
- 0.9
Other (3) 1.7
- 2.9
- 0.6
Marital (Married -1) Separated/Divorced (2)
- 1.8
- 0.9
- 1.1
Widowed (3) 3.3
- 2.9
- 1.7
Education (Illiterate -1) Somewhat elementary (2) 1.2 1.8 0.7 Elementary school (3)
- 1.9
0.4 0.2 Middle school (4)
- 0.5
- 0.9
- 0.6
High school and above (5)
- 0.3
0.5 1.6 Self-reported health (Poor-1, Fair-2, Good-3) 0.0 2.1 1.1
SLIDE 24
24
Disability (No-0, Yes-1)
- 0.5
- .0.8
- 0.7
Other health problems (No-0, Yes-1) 0.5 1.0 1.7 Deceased child (No-0, Yes-1)
- 0.6
- 3.0
- 3.1
Functional limitation ADL
- 1.2
- 3.2
- 3.0
IADL
- 1.3
- 4.2
- 4.2
Pension (No-0, Yes-1) 2.2
- 0.8
0.6 Source of support (Other-0, From children-1) 3.3 2.7 2.3 Living arrangement preference with spouse (Corresidence with children -1) Not same houeshold, but same community (2)
- 2.5
- 1.5
- 1.2
Not same household or community (3) 2.0 1.9 1.6 Nurisng institution (4)
- 0.5
- 0.1
- 0.4
Other (5)
- 0.2
- 0.6
- 0.8
Living arrangement preference with spouse (Corresidence with children -1) Not same houeshold, but same community (2)
- 2.3
- 2.9
- 2.4
Not same household or community (3) 1.3 1.1 1.0 Nurisng institution (4) 0.3 1.2 1.0 Other (5)
- 0.3
- 0.5
- 0.7
C.3 Survey weights with extreme values and 95% trimmed survey weights
SLIDE 25
25
Appendix D.
TABLE 6 The fixed effects model and Allison’s hybrid model predicting subjective well-being among surviving respondents over all the three waves SWB Fixed effects model S.E. Hybrid model S.E. Living with descendants (No-0, Yes-1) 0.093** 0.027 0.098*** 0.027 Age (65+) 0.020 0.035 0.018 0.034 Gender (Male-0, Female-1) 0.187 0.201
- 0.055
0.021 Ethnicity (Minority-0, Han-1)
- 0.029
0.097 0.070 0.035 Residence (Country - 1) Town (2) 0.027 0.032 0.047* 0.021 City (3)
- 0.021
0.063 0.151*** 0.029 Education (years of schooling)
- 0.174
0.317
- 0.162
0.316 Marital (Married -1) Divorced (2)
- 0.078
0.196
- 0.010
0.189 Widowed (3)
- 0.207***
0.037
- 0.202***
0.037 Never married (4)
- 0.152
1.110 0.182 0.590 Deceased child (No-0, Yes-1) 0.029 0.026 0.027 0.026 Pension (No-0, Yes-1) 0.109* 0.054 0.105* 0.052 Functional limitation ADL
- 0.022
0.022
- 0.018
0.019 IADL
- 0.166***
0.019
- 0.161***
0.018 Self-reported health (Very bad-1, Bad - 2, Fair-3, Good-4, Very good-5) 0.357*** 0.018 0.353*** 0.012 Wave (2002 -1) 2005 (2)
- 0.105
0.134
- 0.226
0.141 2008 (3)
- 0.148
0.230
- 0.275
0.232 wave#health (2002-1) 2005 (2) 0.019 0.024 0.055 0.028 2008 (3)
- 0.018