Caregiving can predict the changes in depression symptoms both positively and negatively. Role Accumulation Theory and Role Tension Theory are competing perspectives that have been widely used to study the impacts of grandparenting on mental health. Role Accumulation Theory postulates that multiple social roles are associated with higher life satisfaction and self-efficacy (9) when achieving social integration in different areas (25). As individuals age, retire, and become less socially integrated, the need to strengthen social roles may be greater than ever (26). As a kind of social activity, caring for grandchildren meets intergenerational emotional needs (10) and promotes social integration (27) for grandparents through meaningful engagement. Role Tension Theory, in contrast, suggests that individuals may face role conflicts and role stress when requested to perform specific obligations of different roles (28). Providing childcare may limit grandparents’ time and opportunities to care for themselves (29) and assume the role of marriage (8), making them more vulnerable to emotional stress. Furthermore, older people frequently feel unvalued and disrespected as a result of generational conflicts over child-rearing understanding, which is harmful to psychological health (30).

Rural–urban residence and gender roles shape intergenerational care and depressive symptoms for Chinese grandparents (13). The traditional family division of labour has shaped varied responsibilities between grandmothers and grandfathers. Women are expected to take on more intensive duties such as feeding and cooking, while men usually play a companion role as playmates or helpers (31). In this way, engaging in the care of grandchildren is regarded as a deviation from the norm for men (31). As shown in a previous study, compared with grandmothers, grandfathers suffered from a greater risk of depression when continuously providing high-intensity care (22). However, because women are the main caregivers, some findings showed that the positive mental health outcomes of caregiving were reported significantly only for grandmothers, whereas this was previously thought to be the effect of the whole sample (21, 32). In addition, when considering the intersection of gender roles and the rural–urban context, it may turn out that only urban grandfathers who were financially independent and did not seek an intergenerational time-for-money exchange were associated with fewer depressive symptoms from grandchild care (31). The health effects of continuous caring on gender and urban/rural subgroups have yet to be fully studied.

The consequences of grandchild care are associated with the residence circumstances of the care. First, in recent decades, numerous rural workers have migrated to urban areas, leaving their young children to be cared for by grandparents. It results in rural grandparents having to do more extensive care duties and becoming acclimated to it (1). Second, rural state pension coverage is less extensive than in urban areas, implying that intergenerational care is more of a burden than a reward for rural grandparents (31). This means that rural grandparents have to shoulder the financial and physical burden of caring for grandchildren (33). Third, the one-child policy, which was enforced more rigorously in urban areas, leads to greater involvement and less burden for urban grandparents (31). Affluent urban grandparents tend to compete for opportunities to nurture and care for shrinking numbers of grandchildren to gain emotional rewards (34). Compared to their urban counterparts, rural grandparents are more dependent on their children’s financial support (35–37). In this way, rural grandparents may be more likely to consider grandchild care as a reciprocal form of intergenerational reward rather than an emotionally rewarding activity (33). Forth, rural grandparents could be more likely to become the ‘sandwich’ generation at a young age and face conflicts between providing care for their parents and grandchildren (38), which may make intergenerational care particularly stressful.

Our research contributes to the literature in three ways. First, we used national panel data from China to investigate the relationship between stability and changes in grandparental childcare and depression, expanding on previous studies that were either cross-sectional or had specific region observation windows. Considering that little attention has been paid to the duration and changes of grandparenting on mental health, we classified it based on the intensity of caregiving rather than simple engagement. Second, we adopted generalized estimating equations to estimate the effects of caregiving transitions on depressive symptoms to handle time-dependent autocorrelated data (39). Finally, we performed an exploratory investigation to explore how the relationships between the changes in grandparenting intensity and depression vary between different subgroups of location and gender, while prior studies mostly focused on how the relations differ across gender (40). Based on current evidence, we expect that changes in grandparenting intensity is associated with depression symptoms in later life (Hypothesis 1). Furthermore, the effects of grandparenting change on depression operate differently by gender and urban/rural areas (Hypothesis 2).

The data came from China Health and Retirement Longitudinal Surveys (CHARLS) hosted by the National Development Research Institute of Peking University. CHARLS adopted probability proportional to size sampling to collect a nationally representative sample of Chinese residents aged 45 and above, which included 450 villages in 150 counties and districts throughout China. We obtained national tracking survey data in 2015 and 2018 from the official website.¹ Our analysis was limited to the respondents with follow-up information (n = 36,274) above 45 (n = 34,530) who reported having at least one grandchild under 16 at baseline (n = 25,186) and provided complete answers to the depression level (n = 21,434), the independent variable (n = 18,835), demographics and family characteristics (n = 17,701) in both waves (see Supplementary Figure S1). We compared the differences in socioeconomic characteristics between excluded and retained samples (see Supplementary Table S1), and the retained participants were younger, more males, more rural residences, more at married or cohabiting status, more at fair or good health status, at higher level of individual expenditure, and had higher levels of education, which also showed up in other similar studies (39, 41). Our samples were not significantly different in depressive symptoms when compared with the excluded samples, but it may be necessary to be careful to generalize our results.

The dependent variable was the level of depression. According to the epidemiology research center Depression Scale (CES-D), respondents were asked to assess their psychological and emotional states within a week. The scale included ten subitems. The 4-point responses were rescaled from little or no (0) to most of the time (3). The total score was between 0 and 30. The higher the score, the more serious the depression. Depression levels were log-transformed to fit a normal distribution of the variables and ranged from 0 to 3.434 (Cronbach α = 0.77).

We constructed annual hours of grandparenting to measure the intensity of care and built the classification of changes on that basis. In two waves, respondents with at least one grandchild were asked if they had cared for them in the previous 12 months. If they did, they were asked how many weeks and, on average, how many hours they would have cared for their grandchildren per week in the last year. Using this information, we distinguished between three types of grandparent care: high-intensity care (i.e., those who cared for grandchildren at least 40 h per week on average or for over 2080 h in the last year), low-intensity care (i.e., those who cared for grandchildren for less than 40 h per week on average or under 2080 h last year), and noncaregivers (i.e., those who did not care for grandchildren). We chose 40 h per week for intensive grandparent care because according to the Labor Law of China and previous studies, 40 h per week (five-day working week and no more than 8 h a day) is equivalent to having a full-time job (40, 42). Then, we created a 7-category stability showing the changes in the provision of grandchild care in two waves. We distinguished those who (1) continued to provide high-intensity or (2) low-intensity care at both waves; (3) never provided care; (4) started caregiving (did not care at Wave 1 but provided high-intensity or low-intensity care at Wave 2); (5) ended caregiving (only providing high-intensity or low-intensity care at Wave 1); (6) provided less intensive care (from high-intensity care at Wave 1 to low-intensity care at Wave 2); and (7) provided more intensive care (from low-intensity care at Wave 1 to high-intensity care at Wave 2).

The control variables at baseline included gender (female = 0); location (rural = 0); education level (illiterate = 0; primary school and below = 1; middle school graduation = 2; high school graduation = 3; college and above = 4); marital status (separation, divorce, widowhood, never married = 0; married, cohabiting = 1); self-rated health status (poor = 0; fair =1; good =2); age; whether to live with children (yes = 1); number of grandchildren under the age of 16; number of surviving children; and per capita household expenditure.

We compared the differences of depression levels and other baseline characteristics across four subgroups of rural grandmothers, rural grandfathers, urban grandmothers, and urban grandfathers (see Table 1) and seven category changes over time in grandparental childcare (see Supplementary Table S2), including the chi-square test and variance (ANOVA) for categorical variables and normally distributed continuous variables. Then, we set the models of generalized estimating equations (GEE) to evaluate the effects of changes in caregiving intensity on the level of depressive symptoms, which is flexible to analyse correlated data from the same subjects over time (43) and control the confounding variables that change over time, such as self-rated health status (44). Model 1 was specified to evaluate the effect of the caregiving transition on the dependent variable after taking into account a range of baseline covariates. To test Hypothesis 2 regarding whether the effect of transition of caregiving on depressive symptoms differed in gender and location groups, we divided subgroups in model 2, model 3, model 4 and model 5. Instead of cross-terms for care intensity with gender and urban/rural residence, we operated models by groups to better understand the internal characteristics of different samples.

Table 1 suggests descriptive information of the sample at baseline. According to the stability and change in grandchild care, only 33.44% of grandparents provided no care at either wave. Close to one-fifth of grandparents in Wave 2 continued to provide the same level of intergenerational care reported in Wave 1: 10.78% continued to provide high-intensity care, and 8.39% continued to provide low-intensity care in Wave 2. More than 20% of grandparents increased their level of grandparenting: 15.01% provided no care at Wave 1 and any care for grandchildren in Wave 2; 6.94% provided low-intensity care at Wave 1 and high-intensity care at Wave 2. More than 25% of grandparents decreased the intensity of intergenerational care: 16.84% cared for grandchildren at Wave 1 and ended up at Wave 2; 8.59% provided high-intensity care at Wave 1 and low-intensity care at Wave 2.

There were significantly differences among four groups. Rural grandmothers had the highest scores for depressive symptoms, and urban grandfathers scored the lowest. Rural grandparents had a higher proportion of noncaregivers than urban samples, while urban grandparents had a higher proportion of continuous low-intensity and decreased-intensity care. In total, urban grandfathers were the most educated, the least single, had the best self-assessed health status, had the least number of grandchildren, and had the highest level of individual household expenditure. However, rural grandmothers seemed to be the most disadvantaged in these socio-economic indicators.

The results of the longitudinal impact of grandparenting on depression are presented in Table 2. After controlling for sociodemographic characteristics and family information variables, stability and changes in grandparental childcare were associated with caregivers’ depressive symptoms. Hypothesis 1 was supported. Grandparents who provided high-intensity care in Wave 1 and low-intensity care in Wave 2, who stopped childcare at Wave 2, and who offered low-intensity care in 2 waves were associated with a lower level of depression compared with those providing no childcare. Associations with other covariates were broadly similar to previous studies. Males, living in urban areas, higher-level education, being married or cohabiting, better self-assessed health status, older age, less children, and higher-level of household individual expenditure were associated with decreased depression.

The mental impact of change in grandparenting intensity revealed urban–rural and gender heterogeneity when respondents were divided into four groups. Hypothesis 2 was verified. Model 2 suggested that, compared with rural grandmothers who never participated in grandparenting, those who continued to provide high-intensity childcare were associated with an increased level of depression, and those who decreased the intensity of childcare were associated with a decreased level of depression. Model 3 showed that rural grandfathers who stopped childcare, and those who increased the intensity of childcare were associated with decreased depression when compared with noncaregivers. Model 4 suggested that urban grandmothers who provided continuous high-intensity or low-intensity childcare and those who increased the intensity of childcare were associated with decreased depression when compared with noncaregivers. Model 5 pointed out that the changes in childcare intensity were not significantly associated with urban grandfathers’ depressive symptoms. Figure 1 more clearly displayed differences across groups.

We examined the effects of 16 classification changes in grandparenting intensity based on distinguishment between four types of grandparent care: high-intensity care (i.e., over 40 h per week on average), moderate-intensity care (i.e., more than 10 h but less than 40 h per week on average), low-intensity care (i.e., under 10 h per week on average) and noncaregivers (13). The results (see Supplementary Table S3) indicated that, compared with noncaregivers, grandparents who provided continuous moderate-intensity caregiving, decreased the levels of grandparenting from high intensity to lower intensity, stopped caregiving in Wave 2 from moderate-intensity or low-intensity care in Wave 1, and started caregiving at a low intensity in Wave 2, were associated with decreased levels of depressive symptoms. The results were roughly the same but more detailed than the results above. In addition, providing continuous low-intensity childcare and starting caregiving at a high intensity in Wave 2 were associated with increased depressive symptoms among urban grandfathers. More details were given in Supplementary Table S3.

There are some limitations in this study. First, specific contents of grandparenting activities, grandchild characteristics such as age or health, whether individuals care grandchildren collectively with their spouses, and the reasons for changes in caregiving, which may be associated with caregivers’ actual or perceived intensity of care activities but cannot be measured solely by caregiving time, were not available in the data. Thus, the impacts of intergenerational care could not be more accurate, and there may be some potential confounders that have not been controlled. Second, our study excluded participants who did not provide complete independent variables, dependent variables, and covariates, which may cause selection bias. Although there were no significant differences in depressive symptoms between the excluded samples and the included samples, the participants in our study had a higher likelihood of providing continuous caregiving and more favourable socioeconomic characteristics, which might be associated with a lower level of depression. This finding suggested that grandparents who were more capable of providing care might be oversampled, and favourable mental outcomes should be treated with caution. Third, although we used longitudinal data, these findings may not be a strict casual but only a causal inference under the social science paradigm, because intergenerational care and depressive symptoms were observed simultaneously.