Gender differences in depressive symptoms of rural Chinese grandparents caring for grandchildren

Depression, a non-communicable disease, is globally prevalent regardless of gaps in economic development [13], and social and cultural factors in various regions [14]. It is evidenced that depression can be a contributory cause of numerous physical health problems, such as inflammation [15], Parkinson disease [16], type-2 diabetes [17] and cardiovascular disease [18], resulting in an increasing burden being placed on individual families and communities, as well as affecting statistics at national level. Moreover, globally, depression is most common in middle- and old-age [13, 19]. As the Chinese population ages rapidly, there is an increasing interest in literature on depressive symptoms of grandparents caring for grandchildren. The literature indicates that caring for grandchildren contributed to reduced depressive symptoms in grandparents in some countries [8, 20, 21], the same scenario is seen in China [22‐24]. Based on the theory of role enhancement, multiple social roles lead to improved wellbeing, since individuals gain social integration and gratification from these different areas of social participation [25, 26]. As a type of social role, caring for grandchildren provides grandparents with stronger and more frequent emotional connections with the younger generation and more opportunities for receiving informal and formal support and forging social bonds [27]. According to the existing literature and the theory of role enhancement, we therefore expect that rural Chinese grandparents providing grandchild care have fewer depressive symptoms, compared with those who do not.

Despite the evidence that caring for grandchildren relieves grandparents’ depressive symptoms, a few studies still show it to have an overall negative effect [28‐30], mainly owing to the added stress involved in caregiving. Role strain theory argues that, when individuals play multiple social roles and undertake a series of social obligations, negative health problems can result when they exceed their physical and psychological capabilities [31, 32]. Following the theory of role strain, the heterogeneous findings in the literature regarding the association between grandchild care and grandparents’ depressive symptoms can be explained by the intensity of grandchild-care involvement. Caring for grandchildren may become a stressful and potentially overwhelming task that can have dire consequences for the physical and psychological health of the individual, particularly for grandparents providing extensive care [33], as they find they have little energy and limited time to maintain other personal social activities or social ties [34]. Studies have shown that high intensity of caregiving is more likely to damage the cognitive health of grandparents [35] and indicate a decline in self-reported health [1]. However, a reduction in care intensity was associated with an increase in life satisfaction in the middle-aged and elderly population in China [36]. Moreover, less intensive grandchild care was demonstrated to have an association with a reduction in depressive symptoms for grandparents in France [37]. Meanwhile increasing grandchild care to an intensive level increased depressive symptoms among grandmothers in Poland, Spain and Sweden [37]. The association between intensity of grandchild care and depressive symptoms of caregivers is seen across Western countries. We hypothesize that moderate-intensity grandchild care is likely to result in fewer depressive symptoms, whereas high-intensity grandchild care is associated with more depressive symptoms for rural grandparents in China.

Despite the well-established association between depressive symptoms and grandparents caring for grandchildren, and multiple influencing factors on caregivers’ depressive symptoms (including demographic characteristics, health status and social participation of the caregiver, grandchildren’s characteristics and living arrangements), most of them did not distinguish between grandmothers and grandfathers. Although important gender differences were found in the grandchild-care experience [38‐40] and depressive symptoms were different for males and females [28, 41], most studies analysed grandparents as a demographic group, preventing us from determining a gender differential effect in grandchild-care involvement and depressive symptoms. Given that there are gendered responsibilities and expectations [42], grandmothers usually provide a greater portion of grandchild care than do grandfathers [38, 39, 43]. Empirical evidence on the association between grandchild care and grandparents’ mental health suggests that grandmothers who provide grandchild care are not only at lower risk of depression than those who do not [20], but also experience a higher level of life satisfaction [9, 44] than do grandfathers. Another study using the longitudinal structure of the Survey of Health, Ageing and Retirement in Europe (SHARE) showed a reduction in depressive symptoms in grandmothers who became caregivers [37]. Based on previous evidence around the grandparental role in grandchild care, and depressive symptoms of grandparents caring for grandchildren (considering gender difference), we therefore hypothesize that grandmothers had more depressive symptoms than grandfathers, and grandmothers rather than grandfathers caring for grandchildren are significantly associated with depressive symptoms in rural China.

We attempt to contribute to the literature on grandchild care and depressive symptoms among grandparents by examining the gender differences in depressive symptoms of rural Chinese grandparents caring for grandchildren, using nationally representative data. We propose the following hypotheses: (1) compared with non-caregivers, caregivers are expected to have fewer depressive symptoms; (2) caregivers providing non-intensive of grandchild care are expected to have fewer depressive symptoms, whereas those with intensive grandchild-care involvement are likely to have more depressive symptoms; (3) the above associations are applicable to grandmothers rather than grandfathers; (4) grandmothers suffer more from depressive symptoms than grandfathers.

Data for this study were drawn from the China Health and Retirement Longitudinal Study (CHARLS) conducted in 2015, a nationally representative survey targeting middle- and old-aged adults in China. CHARLS 2015 surveyed 21,789 individuals selected from 450 villages/resident committees in 150 counties/districts in 28 provinces across the country, using a four-stage, stratified, cluster sampling method to select reviewers. The detailed sampling design had been introduced previously [45, 46]. The present study selected grandparents registered as the ‘Agricultural Hukou’ under the Hukou household-registration system [47], with at least one grandchild under 16 years old [27], so as to avoid the possible selection bias suggesting that people without grandchildren are inherently different from those who have grandchildren in terms of health status [48]. Hukou divides Chinese citizens into two categories, with Agricultural Hukou for households in rural areas and Non-agricultural Hukou for those in urban areas [49]. Excluding missing values, the study sample contained 4833 individuals.

Depressive symptoms served as a dependent variable in this study, measured by the 10-item Center for Epidemiologic Studies Depression Scale (CES-D). The respondents were required to assess their mental and emotional states accurately during the week prior to the interview. Each item used a four-point Likert scale, with an answering category ranging from “Rarely or none of the time (< 1 day)” to “Most or all of the time (5–7 days)”, coding from 0 to 3. The sum score ranged from 0 to 30, with higher values presenting more depressive symptoms. Previous studies suggest that CES-D has good reliability and validity among the Chinese population [50]. According to previous studies that a cut-off point of 12 provides the optimal threshold by which to identify clinically significant depression [51, 52], a score of 12 was also used in this study to generate a binary depression variable (No = 0, Yes = 1). It was used to describe further the prevalence of grandparents’ depression. In order to make the continuous variable CES-D score normally distributed to fit the regression, we transferred the CES-D score to its square root. As Fig. 1 shows, the distribution of its square root was broadly normally distributed, although it skewed slightly left (skewness = − 0.28 and kurtosis = 2.78) [53].

Grandchild care was an independent variable in this study, measured by three indicators. One was grandparents’ self-reported answer to the question, “Did you spend any time taking care of your grandchildren last year”, which was divided into ‘Non-caregivers’ and ‘Caregivers’, a binary variable. The other was the answer to the question, “Approximately how many hours per week did you spend last year on taking care of grandchildren”, which was regarded as a continuous variable (duration) and a categorical variable (no care, low intensity, moderate intensity, high intensity), respectively. Based on previous studies [1, 39, 53, 54], three categories of grandchild-care intensity were identified: low intensity (spending 1–14 h per week), moderate intensity (spending 15–39 h per week) and high intensity (spending ≥40 h per week). The cut-off point of 15 h was referred to in the previous literatures [1, 53], and 40 h was chosen based on standard working time per week (five-day working week and no more than 8 h a day) according to the Labour Law of China and previous studies [55, 56]. The gender of grandparents was used to stratify the full sample into the grandfathers and grandmothers subsamples.

Control variables included age (45–59 years old; ≥60 years old); education level (illiterate; primary school or below; middle school; high school or above); marital status (separated/divorced/widowed/never married; married); work status (unemployed/working); annual household income (poorest/2nd quintile/middle/4th quintile/richest); receiving intergenerational support from children (yes/no); co-habiting with children (yes/no); engaging in social activities (yes/no); suffering from chronic disease (yes/no); ADLs score; IADLs score; and number of grandchildren. Annual household income included individual and other household members’ income (e.g., wages, assets, subsidies, government transfers and other income sources), which was divided into quintiles. Intergenerational support from children included cash and in-kind transfers. Engagement in social activities was defined as any activity participated in between individuals, such as interacting with friends; playing mah-jongg/chess/cards or attending community clubs; providing help for family/friends/neighbours outside the household; attending a sport/social/other club; taking part in community-related organizations; undertaking voluntary or charity work; caring for a sick or disabled adult outside the household; attending an educational or training course; stocking investment and using the internet. We considered using the internet as a type of social activity not only because it is one of the options to this question in the CHALRS questionnaire, but also because prior studies have categorised using the internet as a type of social activity, given that individuals can communicate and form social ties without being affected by limitations on mobility and activity for older adults, and it consequently reduces social isolation, and enhances social integration and support [52, 57, 58]. ADLs score was the sum score of items asking interviewers whether they had difficulties with dressing, bathing or showering, eating, getting into or out of bed, using the toilet, and controlling urination and defecation, with each coding from 0 (no difficulties) to 3 (cannot perform it). IADLs score was measured by totalling the items asking respondents whether they encountered difficulties in performing household tasks, preparing hot meals, shopping for groceries, managing personal finances, making phone calls and taking medications, with the same coding as the former. Detailed definitions and codes of variables are presented in Table 1.

Evidence shows that both matched sampling and regression adjustment can be expected to reduce bias [59]. Matching method application is more robust than regression analysis alone [60]. Initially, we used the Coarsened exact matching (CEM) method put forward by Iacus et al. [61, 62] to balance the multidimensional distribution of covariates between the two compared groups (non-caregivers and caregivers), and thereby reduce the explanatory variable’s degree of dependence on the estimation model and further decrease the biases. CEM is a matching method of the class Monotonic Imbalance Bounding (MIB), which shows the basic advantage over other matching methods that the bound on balance for one covariate can be studied and improved in isolation, as this won’t affect any other covariates chosen for balancing [62, 63]. It is preferable to other matching procedures (e.g., propensity score matching, PSM) in terms of processing more efficiently and reducing model dependence, estimation error, variance and bias [64]. It does not require further conduct balance checking or restrict data to common support, as is required by PSM.

The CEM algorithm consists of three principal procedures [61]. Firstly, each variable is coarsened by recoding, and thereby indistinguishable values are grouped and allotted the same numerical value (groups may have the same or different sizes). Secondly, the coarsened data are matched using an “exact matching” algorithm, and unmatched units are pruned. Thirdly, the coarsened data are removed and the uncoarsened (original) values of the matched data are retained. Additionally, a CEM-weights variable is generated to equalize the number of observations within comparison groups, where unmatched units are 0 and matched units are larger than 0 but less than 1 [61]. For balance checking of two compared groups, multivariate imbalance measure L₁ is employed, of which size depends on the dataset and the variables selected. L₁ ranges from 0 to 1, where 0 and 1 represent perfect global balance and complete separation, respectively, and a larger value indicates greater imbalance between two groups. A good match usually reduces the value of L₁ [65]. The L₁ statistic is calculated as follows [66]. Firstly, the covariates are coarsened into bins. Then, the discretized variables are cross-tabulated as X₁× … … × X_k for the treated and control groups, respectively, and k-dimensional relative frequencies are recorded for the treated \( {\mathcal{f}}_{\ell 1\dots \mathit{\ell k}} \) and the control ℊ_{ℓ1. . ℓk} units. Finally, the measure of imbalance is the absolute difference over all the cell values:

In the current study, we matched the socioeconomic characteristics and variables related to grandchild care according to the literature on CEM progress [53, 67], including age, education level, marital status, work status, annual household income, whether receiving intergenerational support from children, whether co-habiting with children, and number of grandchildren.

We first examined the moderation effect of gender on the association between grandchild care and depressive symptoms by creating interaction terms in multilevel linear regression. The predictive margins and the average marginal effects were presented to interpret the gender differences visually. Furthermore, taking gender differences in life expectancy (e.g., females generally live longer than males), socioeconomic status (e.g., males have higher educational attainment and better financial resources than females), and labour market (e.g., males have longer employment history than females) into account, we maintain that they are also likely to contribute to the gender gaps in the association between grandchild care and depressive symptoms. We therefore stratified the analyses by gender and further explored the possible gender differences in such an association.

The Chi-square test for categorical variables and Univariate ANOVA for continuous variables were used to compare caregivers and non-caregivers in the unmatched and matched cohorts. Matched weights were considered in all analyses in matched groups. Since the data of this study were drawn from CHARLS, with a four-stage stratified cluster sampling, the dependence among observations could exist on several levels of the hierarchy. To remove the cluster effect of observations at different levels of the data hierarchy [68], we fitted four-level multiple linear regression models (individual at level 1; nested within the community at level 2; nested within the city at level 3; nested within the province at level 4). An Intra-class Correlation Coefficient (ICC) is used to check the applicability and validity of the multilevel model. The ICC measures correlation among observations within a cluster, ranging from 0 to 1. A multilevel regression model is appropriate for the analysis when ICC is greater than 0 [68, 69]. In this study, the ICC values were 0.038, 0.071 and 0.113 in the grandfather subsample; and 0.023, 0.060 and 0.112 in the grandmother subsample, at province, city and community levels, respectively, which made multilevel linear regression suitable.

Subsequently, we examined the association between grandchild-care duration and depressive symptoms among all participants (including caregivers and non-caregivers) in Model 1; the association between grandchild-care provision and depressive symptoms among all participants in Model 2; and the association between grandchild-care intensity and depressive symptoms, respectively, among all participants in Model 3 and caregivers in Model 4, by using multilevel linear regression, controlling for grandparents’ socioeconomic characteristics (including age, education level, marital status, work status, annual household income) and health status (including chronic disease, ADLs, IADLs), receiving intergenerational support from children, co-habiting with children, number of grandchildren and level of social engagement. Stata statistical software (version 15.0; StataCorp LP, College Station, Texas) was used for all analyses.

We acknowledge several limitations of the current study and, therefore, the findings should be interpreted with caution. Firstly, owing to the limited data on characteristics of grandchildren in CHARLS, we took only the number of grandchildren into consideration. Detailed characteristics of grandchildren (age, gender, health status and so on) cannot be extracted from the questionnaire, which may make estimation less precise. Secondly, we have no information regarding the experiences and assignments of grandparent caregivers. No data are available on their reasons for assuming the responsibility for grandchild care (e.g., out of a sense of obligation); the caregiving pattern (e.g., temporary or primary caregiver, custodial or non-custodial caregiver, sole caregiver or caregiver with other helpers); their feelings during care involvement; the daily matters and activities they perform; the quality of caregiving; or their relationships with their adult children. Future studies covering such information may contribute to identifying the causal pathways underlying the association between grandparental childcare and depressive symptoms, as well as gender differences. Thirdly, grandparents’ self-reported caregiving hours can be subject to recall bias and social-desirability bias [9]. Fourthly, in view of past studies indicating that grandparents in rural China are more likely to be involved in intensive caregiving [11, 36], and their potential medical problems owing to a lack of healthcare resources and supporting alternative childcare facilities compared with urban areas [1], we focused on a rural sample in this study. The possible rural-urban gender differences in depressive symptoms of grandparents caring for grandchildren are needed to be identified. Such a study would contribute to detecting potential gender disparities in grandchild-care engagement and health conditions between rural and urban grandparents, and may be used to further improve policy formulation. Fifthly, although we found there to be an association between grandchild care and depressive symptoms among grandmothers rather than grandfathers, further investigation, including qualitative and quantitative studies, is needed to identify the causal mechanisms, and to assess the extent to which the observed gender differences depend on grandmothers’ and grandfathers’ differential roles, expectations and experiences.

In addition, we declined to use the longitudinal dataset for CHARLS to examine the gender differences in the causal relationship between grandchild care and depressive symptoms among grandparents, owing to limited grandchild information. In order to reduce biased estimation as far as possible, the present study examined only the association between grandchild care and depressive symptoms, although CEM, a quasi-experimental matching method for causal inference, was employed [88]. Based on the findings of this study, it is necessary to examine the causal relationship between grandparental caregiving patterns and depressive symptoms, as well as the gender differences, by using other, more detailed, precise and comprehensive longitudinal data related to grandchild care.