The worst tie matters: intergenerational relationship trajectories in multi-child families and depressive symptoms of older parents in China

Abstract

Objectives

Close intergenerational relationships are associated with better mental health in later life, yet limited longitudinal research has examined how trajectories of relationship quality, especially differentiation across multiple children within the same family, relate to older parents’ depressive symptoms. This study identifies distinct trajectories of parent–child relationship quality and examines their associations with depressive symptoms among older adults in China, while assessing whether these associations vary across socioeconomic groups.

Methods

Using nationally representative panel data from the China Longitudinal Aging Social Survey (2014–2020; N = 3,519), we applied group-based trajectory modeling to characterize six-year patterns of relationship quality, separately capturing the worst, best, and average ties within families. Linear regression models estimated associations between these trajectories and depressive symptoms at the final wave, adjusting for baseline characteristics and testing whether these associations differed by education, occupation, and hukou status.

Results

Parents in steadily or rapidly strengthening relationship trajectories reported significantly fewer depressive symptoms than those in persistently strained trajectories. When trajectories of the worst, best, and average ties were modeled simultaneously, only trajectories of the worst relationship remained significantly associated with depressive symptoms. These associations were concentrated among parents with lower socioeconomic status.

Discussion

Intergenerational relationship dynamics, particularly changes in the most strained tie, are linked to older adults’ mental health in multi-child families. Findings highlight the psychological importance of relational repair and suggest that parents with fewer socioeconomic resources may be especially sensitive to relationship change.

China is experiencing rapid population aging: by 2023, adults aged 65 and older accounted for 15.4% of the population, an increase of 119 million over the past two decades (Wu et al., 2025). With nearly one in five older adults reporting depressive symptoms (Tang et al., 2021), understanding the social determinants of late-life mental health has become increasingly urgent. Close social relationships, including ties with spouses, adult children, and friends, are central to older adults’ well-being. Research suggests that spousal relationships often show the strongest associations with well-being, followed by relationships with children and then friends (Abaei & Martin, 2023, 2026). At the same time, intergenerational ties remain among the most consistently documented predictors of older adults’ psychological health across diverse contexts (Tosi & Grundy, 2019; Zhang & Silverstein, 2022). From a life-course perspective, these relationships evolve as parents and adult children renegotiate roles and expectations, and changes in their quality may have important implications for emotional well-being over time (Umberson & Karas Montez, 2010).

Extensive research links intergenerational relationships to depressive symptoms in later life. Supportive ties with adult children, reflected in frequent contact (Tosi & Grundy, 2019), emotional closeness (Sun et al., 2022), and reciprocal support exchanges (Wang et al., 2023), are associated with fewer depressive symptoms, whereas conflict and relational strain are associated with elevated psychological distress (Li & Zhang, 2023). These associations are consistent with the stress process model (Pearlin et al., 2005), which emphasizes how enduring social roles and relational strain can accumulate as chronic stressors affecting mental health, as well as with life-course perspectives (Moen & Hernandez, 2009) that view parent–child ties as evolving relationships shaped by linked lives and changing social contexts. This literature underscores the theoretical relevance of depressive symptoms as an outcome and highlights the importance of understanding how intergenerational relationship dynamics relate to late-life mental health.

Despite this strong evidence base, three important gaps remain. First, while some studies have examined trajectories of intergenerational relationship quality in Chinese families (Chen & Chen, 2024; Lin & Chen, 2018), research linking intergenerational ties to older adults’ mental health in China typically relies on cross-sectional or short-term designs with static measures of relationship quality (e.g., Li & Zhang, 2023). As a result, the implications of dynamic changes in intergenerational relationships for parents’ mental health remain largely unexplored. Intergenerational ties can strengthen, stabilize, or decline as families renegotiate emotional and practical exchanges (Hogerbrugge & Silverstein, 2015; Zhang et al., 2025). Capturing these longer-term dynamic patterns is necessary to determine whether sustained changes in relationship quality, rather than static levels, shape older parents’ well-being. Disruptions in evolving ties may heighten vulnerability to depressive symptoms, whereas adaptive improvements may promote resilience and emotional stability (Umberson & Karas Montez, 2010).

Second, most research focuses on individual aspects of intergenerational ties, such as contact frequency or instrumental support, without addressing their multidimensional nature (Lin & Chen, 2018; J. Chen & Chen, 2024). Family relationships encompass emotional, functional, and associational dimensions that may change at different rates or even in different directions (Bengtson & Roberts, 1991; Guo et al., 2012). For example, parents can be emotionally close yet exchange little material support (Li & Zhang, 2023; Hogerbrugge & Silverstein, 2015). Relying on a single dimension risks missing meaningful shifts in overall relationship quality.

Third, studies rarely address within-family differentiation in relationship quality (e.g., Chen & Chen, 2024). In multi-child families, parents may be close to some children and strained with others (Li & Zhang, 2023), and these patterns can shift over time (Moen & Hernandez, 2009). Emerging theoretical work, including the threshold model (Suitor et al., 2018), suggests that the most strained tie may be especially consequential for well-being. Distinguishing the best, worst, and average parent–child relationships within multi-child families thus provides a fuller understanding of how intra-family inequality in relational quality affects mental health.

In addition to these core gaps, less is known about how relationship trajectories operate across socioeconomic contexts. The stress process model (Pearlin, 1989) suggests that socioeconomic resources may buffer stressors, implying that relationship change may have heightened emotional consequences among parents with fewer structural supports. Examining socioeconomic variation may therefore offer important contextual insights.

To address these gaps, we use four waves (2014–2020) of the nationally representative China Longitudinal Aging Social Survey (CLASS) to identify trajectories of multidimensional intergenerational relationships in multi-child families. We employ a novel approach that models trajectories separately for the best, worst, and average parent–child ties, capturing both dynamic change and intra-family heterogeneity. We then examine how these trajectories are associated with older parents’ depressive symptoms and how these associations vary by socioeconomic status.

This study makes three central contributions to theories of aging and family. First, it brings a developmental, life-course lens to intergenerational ties by emphasizing relationship trajectories, theorizing family relationships as unfolding processes rather than fixed states. Second, it strengthens the conceptualization of intergenerational solidarity by adopting a multidimensional understanding of relationship quality, recognizing that emotional, functional, and associational elements jointly shape parents’ relational environments. Third, it advances family systems and linked-lives perspectives by conceptualizing the family as a multi-dyadic structure, demonstrating that differentiation across a parent’s best, worst, and average relationships is theoretically consequential. Together, these contributions offer a more dynamic and nuanced framework for understanding how intergenerational relationships relate to mental health in later life.

China provides a particularly salient context in which to examine these processes. Culturally, Confucian filial norms emphasize interdependence and enduring moral obligations between generations (Zhang et al., 2024). Institutionally, however, rapid population aging has outpaced the development of formal eldercare and mental health services, creating what Riley (1987) terms a structural lag between demographic change and supportive infrastructure. As a result, family relationships remain a primary source of emotional and practical support for many older adults. In this setting, the quality and evolution of intergenerational ties may carry heightened significance for psychological well-being, particularly when institutional resources are limited.

Theoretical framework and hypotheses

A life-course perspective on intergenerational relationship change

A life-course perspective provides a foundational lens for understanding how intergenerational relationships relate to older adults’ mental health. This framework emphasizes that lives are interdependent, what Elder (1994) terms “linked lives,” and that these interdependencies unfold within changing social contexts, role transitions, and family structures. Parent–child relationships, in particular, are shaped by long histories of interaction and shared experience, yet they remain open to renegotiation across adulthood.

Prior research highlights both continuity and change in these ties. On the one hand, early patterns of affection, support, or strain often persist into later life, reflecting stable expectations and emotional bonds established long before old age (Crosnoe & Jr., 2002; Hogerbrugge & Silverstein, 2015; Schenk & Dykstra, 2012). On the other hand, consistent with the social convoy model, intergenerational relationships also show discontinuity as families adapt to major life events such as health decline, widowhood, or geographic mobility (Chen & Chen, 2024; Lin & Chen, 2018). These dual tendencies indicate that parent–child ties are neither fixed nor fully malleable. Instead, they evolve along identifiable trajectories—patterns of stability, deterioration, or improvement—that reflect both enduring relational histories and ongoing adaptation.

These trajectories are theoretically relevant for understanding late-life mental health. The stress process model (Pearlin, 1989) suggests that long-standing social roles can serve as chronic stressors or enduring sources of support. Because the parental role is both salient and identity-defining, persistent strain in relationships with adult children may accumulate over time, eroding emotional well-being (Pearlin & Skaff, 1996). Conversely, consistently positive or improving relationships may function as psychosocial resources that enhance coping and promote psychological stability (Antonucci et al., 2014). Integrating life-course and stress-process insights thus underscores the importance of the direction and persistence of relational change, not simply its level at a single point in time.

Several conceptual pathways help explain why evolving intergenerational ties may be linked to depressive symptoms, even though these pathways cannot be directly assessed in the present study. Sustained conflict may contribute to role-related strain, weakening the emotional security and sense of competence associated with the parental role (Bai, 2018; Pearlin et al., 2005). Conversely, improvements in relational quality may enhance perceived support and signal successful adaptation to life transitions, potentially buffering psychological distress (Walsh, 2003). These perspectives collectively suggest that capturing trajectories of intergenerational relationships provides a more informative view of how family dynamics relate to older adults’ mental health than static assessments alone. Based on this literature, we hypothesize that trajectories characterized by stability or improvement will be associated with fewer depressive symptoms than persistently strained trajectories (H1). 

Multidimensional and multi-dyadic family dynamics

Intergenerational relationships are inherently multidimensional. Classic formulations of intergenerational solidarity view these ties as comprising emotional, functional, associational, and normative components (Bengtson & Roberts, 1991). Because these dimensions capture different facets of intergenerational connection, a multidimensional approach provides a more comprehensive reflection of the relational environment older parents experience. Rather than relying on any single indicator, such as contact frequency or instrumental exchange, a multidimensional measure aligns with theoretical understandings of solidarity and offers a holistic summary of relationship quality as it evolves across the life course.

Family relationships are also multi-dyadic, particularly in societies such as China, where most older adults have more than one child. Research shows that parents often experience heterogeneous relationships with different children, reflecting variation in personality, proximity, socioeconomic resources, and caregiving capacities (Ward, 2008; Ward et al., 2009). These differences matter: the “best” relationship may provide emotional support and validation, while the “worst” may be a site of conflict, disappointment, or unmet expectations. Moreover, these distinctions can shift over time as family roles change, children relocate, or caregiving demands intensify (Kim et al., 2016). Conceptualizing the family as a network of linked dyads, rather than a single parent–child pair, allows a more precise understanding of how relational dynamics unfold in multi-child families.

Theoretical perspectives further suggest that within-family differentiation may have distinct implications for well-being. The threshold model (Suitor et al., 2018) proposes that even one highly strained tie may exert disproportionate psychological influence, potentially outweighing the benefits of more positive relationships. From a family systems perspective, negative interactions may be especially salient because they threaten emotional security and parental identity, whereas positive ties may be less potent in offsetting strain. These insights underscore the importance of examining a parent’s best, worst, and average relationships separately when assessing how intergenerational relationship trajectories relate to mental health. Based on this literature, we hypothesize that the parent’s worst child relationship trajectory will be particularly associated with parental depressive symptoms (H2).

Taken together, a multidimensional and multi-dyadic perspective offers a more comprehensive framework for understanding intergenerational relationships in later life. It accounts for diversity in relational experiences both across domains (emotional, functional, associational) and across children, thereby providing a more nuanced basis for studying how family dynamics may shape psychological well-being.

Socioeconomic stratification as a contextual moderator

Socioeconomic stratification shapes the conditions under which intergenerational relationship trajectories may be more or less consequential for older adults’ mental health. The stress process model emphasizes that socioeconomic resources, such as education, income, and occupational status, affect both exposure to stressors and the capacity to cope with them (Pearlin et al., 2005). Older adults with higher socioeconomic status (SES) tend to have greater access to formal supports, broader social networks, and more effective coping resources, which may buffer the psychological impact of strained or deteriorating family ties. Conversely, limited resources may heighten vulnerability to relational stress, making the emotional quality of intergenerational ties more salient for mental well-being.

These theoretical insights are particularly relevant in the Chinese context. Structural inequalities stemming from the hukou system, uneven pension coverage, and disparities in access to eldercare services create substantial variation in older adults’ reliance on family support. For parents with fewer institutional or economic resources, stable or improving relationships with adult children may represent an especially important source of emotional and practical support, whereas strained ties may carry a greater psychological cost. Although SES is not the primary focus of our theoretical contribution, considering socioeconomic variation provides important contextual insight into why the effects of intergenerational relationship change may differ across groups. Based on this literature, we hypothesize that the associations between relationship trajectories and depressive symptoms will be especially strong among older adults with lower socioeconomic status (H3).  Figure 1 illustrates our conceptual framework.

Figure 1.

Conceptual model of the study.

Method

Data and sample construction

This study utilized four waves of data (2014, 2016, 2018, and 2020) from CLASS, a nationally representative longitudinal study conducted by Renmin University of China. The survey was initiated in 2014 and adopted a multistage stratified random sampling procedure, with counties as the primary sampling units and villages in rural areas and communities in urban areas as secondary sampling units. CLASS is the only nationally representative longitudinal dataset in China that collects detailed information on various dimensions of older adults’ intergenerational relationships, which represent the focus of this study.

CLASS surveyed 11,511, 11,471, 11,418, and 11,398 respondents aged 50 and older in 2014, 2016, 2018, and 2020, respectively. Because CLASS replenished respondents in each follow-up wave after 2014, the cumulative number of unique eligible parents across all waves was 22,412. From this cumulative sample, we first restricted the analysis to parents with at least two working-age children (aged 15 and older), as older children are more likely to assume caregiving roles and are less dependent on their parents. This restriction yielded 16,994 parents (33,703 person-wave observations). Second, to enable trajectory modeling, we retained parents who were surveyed in at least three waves. This requirement resulted in 5,970 parents (19,874 observations; the parent−child sample). Third, because our analyses examine how prior relationship trajectories predict depressive symptoms at the final observed wave, we retained only the final-wave observation for each of the 5,970 eligible parents. Finally, we excluded respondents with missing information on key variables. Missingness was 14.2% for the outcome and 0.3% for the core explanatory variables. After these exclusions, the final analytic sample consisted of 3,519 older parents (the parent sample). Supplementary Figure 1 provides a detailed flowchart of the sample construction process.

For robustness, we additionally performed multiple imputation using chained equations to address missing covariates and re-estimated all models using the imputed sample of 5,108 parents across five imputed datasets, excluding those with missingness in the outcome and the core explanatory variables of intergenerational relationship (Rubin, 2018). It should be noted that the core explanatory variable, derived from trajectory modeling, is a model-generated latent construct and therefore cannot be validly imputed.

Measures

Depressive symptoms

CLASS relies on the Center for Epidemiologic Studies-Depressive symptoms (CES-D) scale to measure an older parent’s depressive symptoms (Radloff, 1977). The scale comprises three items assessing positive emotions: “Did you feel happy in the past week?,” “Did you enjoy life in the past week?,” and “Did you engage in fun or interesting activities in the past week?” Two items measure negative emotions: “Did you feel lonely in the past week?,” and “Did you feel sad in the past week?” Two items evaluate somatic symptoms: “Did you lose appetite in the past week?,” and “Did you experience restless sleep in the past week?” Two items assess feelings of worthlessness and idleness: “Did you feel worthless in the past week?,” and “Did you feel a lack of purpose in the past week?.” Each item was scored on a scale of 0 (rarely or none of the time), 1 (some of the time), or 2 (most or all of the time). After reverse-coding the first three items, we summed the frequency of nine items and created a total score ranging from 0 to 18, with a higher score indicating more depressive symptoms. In robustness, we also used a binary indicator of depressive symptoms based on a clinical cutoff of 6 on our 0–18 CES-D scale, in approximation to the standard cutoff of 16 on the 60-point scale for depressive symptoms (Zhang et al., 2025). We excluded respondents with missing one or more data in their baseline and final waves of the survey. The 9-item scale demonstrated acceptable internal consistency (Cronbach’s α ≥ .70 in Tavakol and Dennick, 2011), with Cronbach’s α values of .76 at baseline and .70 at the final wave, respectively.

Intergenerational relationship in multi-child families

Based on the observed parent–child relationship indicators across at least three waves, our analysis calculated total score of parent−child relationships and aimed to predict trajectories of the best, the worst, and the average relationship in multi-child families, respectively.

Referring to previous studies (Bai, 2018; Bengtson & Roberts, 1991; Van Gaalen & Dykstra, 2006), we first characterized the dyadic parent−child relationship using nine indicators from three dimensions in our parent−child sample: emotional affinity, functional exchange, and associational connection (see Supplementary Table 1 for more details). Specifically, the indicators of emotional affinity included self-rated closeness towards each child (scale 1–3), self-rated emotional indifference from each child (scale 1–4, reverse-coded), and self-rated excessive help required by each child (scale 1–4, reverse-coded). The indicators of functional exchange comprised the frequency of economic support (money, food, and gifts in RMB, scale 1–9) over the past 12 months (from each child to the parent and vice versa) and the frequency of housework support (scale 1–5) over the past 12 months (from each child to the parent and vice versa). The indicators of associational connection included contact frequency through in-person meetings (scale 1–5) or by phone (scale 1–5) over the past 12 months.

Next, following previous studies (Bai, 2018; Deng et al., 2025), we summed the scores of the nine measures mentioned above to identify parent−child relationship quality between parents and each of their adult children, with a higher score representing a higher parent−child relationship quality. We constructed three measures of intergenerational relationship quality by summarizing parent–child relationship scores across multiple children for each older parent: the worst relationship, the best relationship, and the average relationship. Importantly, these measures represent within-wave summaries rather than fixed child-specific dyads. Specifically, in each wave, the best relationship corresponds to the child with the highest score, the worst relationship corresponds to the child with the lowest score, and the average relationship is the mean score across all children. Because these summaries are recalculated at each wave, the child identified as the “worst” or “best” may differ across waves. This procedure was repeated across waves, and the resulting best, worst, and average scores were used to construct relationship trajectories over time. Accordingly, the trajectory models capture changes in a parent’s most strained, most supportive, and overall relationship quality across waves, rather than changes within a specific parent–child pair.

Covariates

Following previous studies (e.g., Li & Zhang, 2023), we controlled for a set of demographic and socioeconomic indicators at baseline to account for potential confounding factors that may influence both relationship quality and depressive symptoms. These variables include age (in years), gender (0 = man, 1 = woman), educational years approximated by the raw question of highest educational attainment (0 years  =  illiterate/literacy class, 6 years = primary school, 9 years = junior high school, 12 years  =  senior high school/vocational high school, and 15 years = junior college or above), parents’ total income (RMB, in logarithm), hukou status (0 = non-agricultural or urban hukou, 1 = agricultural or rural hukou), homeownership (0 = no, 1 = yes), marital status (0 = unmarried, 1 = married), number of adult children (2–5), parents’ instrumental activities of daily living score (0–16, higher scores indicating greater difficulty), and parents’ CES-D score (0–18, higher scores indicating more symptoms). Among these variables, hukou is China’s household registration system, which assigns individuals to either urban or rural status and determines access to welfare benefits, public services, and economic opportunities. Including hukou status allows us to capture structural inequalities that may shape family relations and mental well-being.

Stratified variables

To evaluate whether the associations between trajectories of intergenerational relationships and depressive symptoms differ across SES among older Chinese parents, we constructed three stratified variables at baseline: occupational class (non–white-collar positions vs. white-collar positions); educational attainment (under junior high school vs. junior high school and above; different from educational years as a continuous covariate in the main models); hukou status (0 = non-agricultural or urban hukou vs. agricultural or rural hukou). Referring to Enroth et al. (2013) and Martikainen et al. (2008), occupational class was measured based on current occupations for employed persons and previous occupations for unemployed and retired people, respectively: leaders of government or enterprise/public institutions, professional and technical personnel, and general office staff are classified into white collar positions; others are categorized into non-white collar positions. In addition, the categorization of educational attainment at the junior high school threshold is motivated by both empirical and institutional reasons. First, the distribution of schooling in our sample is concentrated at a lower level, with 63.09% of respondents having primary education or less. Second, the cohort under study (mean birth year 1952) completed their schooling well before the implementation of China’s 1986 Compulsory Education Law. Progression beyond primary school was not institutionally guaranteed and therefore represents meaningful variation in educational attainment for this cohort (Fang et al., 2012).

Table 1 presents the descriptive statistics of the sample at baseline. The majority of respondents were men (52.54%), married (75.62%), property owners (89.94%), and held a rural hukou (52.83%). The mean age of older parents was 68.43 years, while their adult children had an average age of 44.42 years (not included as a covariate). On average, respondents had 6.30 years of education, an annual income of 18,609 RMB (approximately 2,593 US dollars), and around three children. Instrumental activities of daily living (IADL) scores were low (mean = 0.77). Moderate depressive symptoms were observed at both baseline (mean = 5.66) and in 2020 (mean = 6.71).

Table 1.

Descriptive statistics of older parents in the CLASS.

Variables	Mean (SD)	Range	Frequency	(%)
Depressive symptoms (final wave)
CES-D scores	6.71 (3.20)	0–18
Parent−child relationship trajectories
The worst parent−child relationship
Persistently strained relationship			366	10.40
Gradual decline with sustained positivity			343	9.75
Steadily strengthening relationship			2,245	63.80
Rapidly strengthening relationship			565	16.06
The best parent−child relationship
Persistently strained relationship			263	7.47
Gradual decline with sustained positivity			216	6.14
Steadily strengthening relationship			2,069	58.80
Rapidly strengthening relationship			971	27.59
The average parent−child relationship
Persistently strained relationship			257	7.30
Gradual decline with sustained positivity			266	7.56
Steadily strengthening relationship			2,155	61.24
Rapidly strengthening relationship			841	23.90
Covariate (baseline wave)
Age	68.43 (6.40)	59–93
Woman			1,670	47.46
Educational years	6.30 (4.15)	0–16
Income (RMB, in logarithm)	8.88 (2.02)	0–13.20
Rural hukou holder			1,859	52.83
Home property owner			3,165	89.94
Married			2,661	75.62
Number of children	2.82 (0.92)	2–5
IADL score	0.77 (1.85)	0–16
Depressive symptoms	5.66 (3.30)	0–18

Note. CES-D = Center for Epidemiologic Studies Depression Scale; CLASS = China Longitudinal Aging Social Survey; IADL = Instrumental Activities of Daily Living. N = 3,519. Depressive symptoms are measured in the final wave of the survey, while parent−child relationship trajectories are based on data from at least three waves of the survey in 2014, 2016, 2018, and 2020. All covariates are derived from the baseline observation.

Analytical strategy

To identify the association between trajectory of intergenerational relationship quality and parents’ depressive symptoms in multi-child families, we first used Group-Based Trajectory Modeling (GBTM) to identify and characterize the trajectories of the lowest, the highest, and the mean parent−child relationship quality based on at least three waves of indicators in the parent−child sample. Unlike conventional growth curve approaches such as latent growth models or growth mixture models, GBTM does not assume that individual trajectories within a class vary randomly around a common functional form. Instead, it represents each latent group with its own distinct trajectory shape (e.g., linear, quadratic), allowing researchers to compare and select among different functional forms for each group (Nagin, 2005).

Censored normal models were used to analyze intergenerational relationship, as it was treated as a continuous variable. To identify the optimal number and shape of trajectory groups, we fitted models with two to five groups and tested the significance of intercept, linear, and quadratic terms of time (Huang et al., 2024). Model fit was evaluated using the Bayesian information criterion (BIC), where a lower value indicates a better fit. We also assessed the model’s fit by examining the average posterior probability (AvePP), ensuring that the AvePP for each trajectory group was at least 0.7, which suggests a good fit (Nagin, 2005). Specifically, we found that the four-trajectory model had a relatively lower BIC and most AvePP values above 0.7 compared to models with other numbers of trajectories, across all relationship trajectory types (best, worst, and average). See Supplementary Tables 2–4 for details. In addition, models with more groups did not substantively change the overall patterns. Therefore, we selected the four-trajectory model (comprising 2 linear and 2 quadratic trajectories) as the best-fit model.

Next, we matched the parent sample with the types of trajectories of parent−child relationship quality derived from the parent–child sample. Older parents represent our analytical unit for the association between parent–child relationship trajectory and parent’s depressive symptoms in multi-child families. Based on the parent sample, we used the ordinary least square (OLS) models to test how trajectories of the best, the worst, and the average parent−child relationship quality for the period 2014–2020 are associated with older parents’ depressive symptoms in their final observed or final wave (2018 or 2020) in multi-child families. Specifically, Model 1–3 assess H1 that no matter in best, worst, or average tie, improvements in intergenerational relationships are associated with fewer depressive symptoms. Model 4 puts best, worst, and average together and examines H2 that the parent’s worst child relationship trajectory is particularly associated with parental depressive symptoms. To test H3 that the association between relationship trajectories and depressive symptoms is especially strong among low-SES parents, we performed moderation analysis using six OLS models on stratified samples based on parental occupation status (non–white-collar vs white-collar), educational status (below middle school vs middle school and above), and hukou status (rural vs urban).

It should be noted that our OLS models are rational because the distribution of depressive symptoms in the final wave shown in Supplementary Figure 2 is nearly normal. In robustness, we employed Probit models to predict the probability of exhibiting depressive symptoms based on the clinical cutoff discussed in the variable section (see Supplementary Table 5). All statistical analyses were performed by STATA 17.

Results

Intergenerational relationship trajectories

Across all three relationship contexts, most parent–child ties followed steady or rapid strengthening pattern over time, indicating that positive intergenerational dynamics were the dominant long-term trajectory. For the worst relationships (Figure 2A), the persistently strained trajectory (10.40%) exhibited declining relationships, with scores declining from approximately 23 to 20. The gradual decline with sustained positivity trajectory (9.75%) showed a declined more gradually from 33 to 30. The steadily strengthening trajectory (63.80%) improved from 25 to 27, while the rapidly strengthening trajectory (16.06%) demonstrated the largest improvement, rising sharply from 24 to 32. For the best relationships (Figure 2B), the persistently strained trajectory (7.47%) declined from 25 to 22. The gradual decline with sustained positivity trajectory (6.14%) began with the highest scores (36), declining slightly to 34. The steadily strengthening trajectory (58.80%) first declined slightly from 29 to 27 and then improved from 27 to 29, while the rapidly strengthening trajectory (27.59%) exhibited significant improvement, rising from 27 to 34. For the average relationships (Figure 2C), the persistently strained trajectory (7.30%) declined from 24 to 21, while the gradual decline with sustained positivity trajectory (7.56%) decreased from 34 to 32. The steadily strengthening trajectory (61.24%) improved from 26 to 28, and the rapidly strengthening trajectory (23.90%) showed rapid improvement from 26 to 31.

Figure 2.

Four trajectories of the worst, the best, and average intergenerational relationship estimated by group-based trajectory modeling. Solid lines are the estimated trajectories. Dot symbols are the observed group means at each time point. Dashed lines are 95% pointwise confidence intervals on the estimated trajectories (7488 cases in total, referred to as the parent−child sample).

Intergenerational relationship trajectories and parents’ depressive symptoms

Table 2 presents the results for testing H1 (Model 1–3) and H2 (Model 4). Overall, strengthening relationship trajectories were associated with fewer depressive symptoms than persistently strained relationships. When accounting for multiple relationship dimensions simultaneously, only trajectories of the worst relationship remained significantly associated with depressive symptoms.

Table 2.

Ordinary least squares models predicting parental depressive symptoms.

Variables	Model 1	Model 2	Model 3	Model 4
Worst relationship (ref: persistently strained relationship)
Gradual decline with sustained positivity	−0.954** (0.284)			−1.012* (0.397)
Steadily strengthening relationship	−1.261*** (0.190)			−0.393 (0.261)
Rapidly strengthening relationship	−1.411*** (0.208)			−0.958** (0.344)
Best relationship (ref: persistently strained relationship)
Gradual decline with sustained positivity		−1.320*** (0.244)		0.147 (0.437)
Steadily strengthening relationship		−0.999*** (0.174)		−0.539† (0.312)
Rapidly strengthening relationship		−1.411*** (0.217)		−0.476 (0.347)
Average relationship (ref: persistently strained relationship)
Gradual decline with sustained positivity			−1.428*** (0.277)	−0.516 (0.569)
Steadily strengthening relationship			−1.324*** (0.196)	−0.560 (0.398)
Rapidly strengthening relationship			−1.542*** (0.221)	−0.434 (0.464)
Comparison of relationship trajectories
Gradual decline with sustained positivity
Worst relationship—best relationship				0.043
Worst relationship—average relationship				0.572
Best relationship—average relationship				0.465
Steadily strengthening relationship
Worst relationship—best relationship				0.715
Worst relationship—average relationship				0.775
Best relationship—average relationship				0.973
Rapidly strengthening relationship
Worst relationship—Best relationship				0.320
Worst relationship—average relationship				0.469
Best relationship—average relationship				0.955
Observation	3,519	3,519	3,519	3,519

Note. Coefficients (β), with robust standard errors in parentheses, are reported. All models control for baseline parental background information, including income, education, home ownership, gender, age, hukou status, marital status, number of children, instrumental activities of daily living (IADL), and depressive symptoms. In Model 4, the p-values from the Wald test comparing the relationship trajectories by relationship dimensions are reported.

^† p < .10.

^* p < .05.

^** p < .01.

^*** p < .001.

Across Models 1–3, strengthening trajectories for the worst, best, and average ties were each associated with fewer depressive symptoms. Specifically, Model 1–3 show that gradual decline with sustained positivity (β_worst = −0.954, p < .01; β_best  =  −1.320, p < .001; β_average = −1.428, p < .001), steadily strengthening (β_worst  = −1.261, p < .001; β_best = −0.999, p < .001; β_average =−1.324, p < .001), and rapidly strengthening trajectories (β_worst  =  −1.411, p < .001; β_best  =  −1.411, p < .001; β_average = −1.542, p < .001) were significantly associated with fewer depressive symptoms compared to persistently strained relationships, after adjusting for demographic factors, health covariates, and depressive symptoms at baseline. These results support H1 that improvements in the relationships are associated with fewer depressive symptoms for older parents.

In Model 4, after accounting for the worst and best relationship trajectories, the effects of average relationship trajectories were no longer significant. For the worst relationship, gradual decline with sustained positivity (β = −1.012, p < .05), and rapidly strengthening trajectories (β = −0.958, p < 0.01) were significantly associated with fewer depressive symptoms compared to persistently strained relationships after adjusting for demographic factors, health covariates, and depressive symptoms at baseline. For the best relationships, steadily strengthening (β = −0.539, p < 0.10) exhibited a marginally significant association, though to a lesser extent. These findings suggest that the worst intergenerational relationship trajectory is significantly associated with parental depressive symptoms, supporting H2.

To further examine whether the effects differ across relationship dimensions, we conducted Wald tests comparing coefficients for worst, best, and average relationship trajectories within Model 4. The results suggest that these differences are in general not statistically significant (see Table 2 notes).

Differences in intergenerational relationship trajectories and depressive symptoms by parents’ socioeconomic resources

Table 3 presents subgroup analyses based on parents’ SES, which include occupation, education, and hukou status. Significant associations between worst or best relationship trajectories and parents’ depressive symptoms were observed only among parents with low socioeconomic status (low occupational class, below middle school education, or rural hukou). As for the worst relationship, compared to the persistently strained relationship, gradual decline with sustained positivity (β_occupation = −1.056, p < .05; β_education = −1.096, p < .05; β_hukou = −1.150, p < .10), and rapidly strengthening trajectories (β_occupation = −1.043, p < .01; β_education = −1.187, SE = 0.403, p < .01; β_hukou = −0.845, p < .10) were associated with fewer depressive symptoms. For the best relationship, in contrast to the persistently strained relationship, steadily strengthening relationship (β_occupation = −0.715, p < .05; β_education = −0.795, p < .05; β_hukou = −0.557, p < .10), and rapidly strengthening trajectories (β_occupation = −0.776, p < .05; β_education = −0.877, p < .05; β_hukou = −0.673, p < .10) were associated with fewer depressive symptoms. No significant associations were found for high SES groups. These results indicate that the association between relationship trajectories and depressive symptoms is especially strong among older adults with lower socioeconomic status, supporting H3.

Table 3.

Ordinary least squares models predicting parental depressive symptoms in 2020: Sub-group analysis.

Variables	Non-White collar	White-collar	Below middle school	Middle school and above	Rural hukou	Urban hukou
Worst relationship (reference group: persistently strained relationship)
Gradual decline with sustained positivity	−1.056* (0.447)	−1.070 (0.892)	−1.096* (0.544)	−0.864 (0.699)	−1.150† (0.658)	−0.769 (0.590)
Steadily strengthening relationship	−0.393 (0.280)	−0.540 (0.679)	−0.273 (0.280)	−0.636 (0.579)	−0.572† (0.313)	−0.110 (0.475)
Rapidly strengthening relationship	−1.043** (0.382)	−0.732 (0.808)	−1.187** (0.403)	−0.626 (0.673)	−0.845† (0.476)	−0.827 (0.554)
Best relationship (reference group: persistently strained relationship)
Gradual decline with sustained positivity	−0.091 (0.476)	1.808 (1.164)	−0.318 (0.539)	1.067 (0.764)	−0.597 (0.642)	0.646 (0.954)
Steadily strengthening relationship	−0.715* (0.316)	0.679 (1.012)	−0.795* (0.353)	0.051 (0.611)	−0.557† (0.313)	−0.516 (0.880)
Rapidly strengthening relationship	−0.776* (0.359)	1.266 (1.069)	−0.877* (0.401)	0.352 (0.662)	−0.673† (0.380)	−0.193 (0.906)
Average relationship (reference group: persistently strained relationship)
Gradual decline with sustained positivity	−0.441 (0.623)	−1.172 (1.491)	−0.363 (0.750)	−1.365 (0.991)	−0.216 (0.873)	−1.236 (1.129)
Steadily strengthening relationship	−0.424 (0.405)	−1.141 (1.269)	−0.345 (0.443)	−1.128 (0.811)	−0.352 (0.419)	−0.953 (1.012)
Rapidly strengthening relationship	−0.172 (0.487)	−1.639 (1.363)	0.182 (0.540)	−1.696† (0.897)	−0.011 (0.546)	−1.101 (1.068)
Observation	2,852	667	2,220	1,299	1,859	1,660

Note. Coefficients (β), with robust standard errors in parentheses, are reported. All models control for baseline parental background information, including income, education, home ownership, gender, age, hukou status, marital status, number of children, instrumental activities of daily living (IADL), and depressive symptoms. The worst, best, and average relationship trajectories are included in a single model for all subgroup analyses (see Model 4).

^† p < .10.

^* p < .05.

^** p < .01.

Discussion

This study contributes new evidence on how intergenerational relationships shape older adults’ mental health by showing that meaningful variation lies not only in the level of relationship quality but in its developmental course over time. Using 6 years of nationally representative data from China, we identify distinct trajectories of the worst/best relationship summary and demonstrate how these trajectories relate to depressive symptoms. By centering depressive symptoms as a theoretically grounded indicator of late-life mental health, our findings directly extend prior research linking intergenerational ties to psychological well-being. Our findings highlight three contributions that extend existing understandings of aging, family dynamics, and mental health.

First, our results underscore the importance of treating intergenerational relationships as dynamic processes. While prior studies emphasize the benefits of supportive ties among adults (e.g., Li & Zhang, 2023; Sun et al., 2022), few have examined how changes in these ties accumulate and matter for well-being. Consistent with the stress process model (Pearlin et al., 2005), which emphasizes the cumulative impact of chronic relational strain, we find that parents in persistently strained trajectories report the highest levels of depressive symptoms. By contrast, parents whose relationships steadily or rapidly improve report substantially fewer symptoms, even though their baseline relationship quality is lower than that of the gradual-decline-with-sustained-positivity trajectory, which begins at a relatively high level but gradually declines. These findings suggest that the direction and persistence of relational change, rather than baseline quality alone, carry meaningful implications for depressive symptom risk. These patterns show that relationship repair is not merely a return to baseline but a potentially powerful resilience-building process, consistent with recent empirical studies showing that children’s functional exchanges with parents often increase and geographic distance narrows in response to parental health declines (Reyes & Shang, 2024; Y. Zhang et al., 2025). By empirically capturing sustained deterioration versus improvement, our findings advance life-course perspectives that view family relationships as unfolding trajectories and demonstrate that shifts in these trajectories, not just their static levels, carry important mental-health consequences.

Second, by distinguishing the best, worst, and average child–parent ties, this study provides dynamic support for the threshold model and expands how scholars conceptualize multi-dyadic family systems. Prior research suggests that one strained relationship may disproportionately undermine well-being (Suitor et al., 2018), but most evidence has been cross-sectional. When trajectories of the worst, best, and average relationships are modeled simultaneously, only trajectories of the worst relationship remain significantly associated with depressive symptoms. These findings refine the threshold model by showing that the “worst tie” is not simply a fixed source of burden; its direction of change matters. Improvements in the most strained tie were associated with lower depressive symptoms, indicating that relational resilience in this specific dyad can offset years of accumulated strain. Rather than implying comparative magnitude, our results demonstrate the independent salience of the most strained tie within the broader relational context of multi-child families. This reveals an overlooked pathway through which families adapt and through which parents emotionally recalibrate their relationships in later life.

Third, our findings illuminate how socioeconomic stratification shapes the emotional significance of family relationships. We find that trajectory–mental health associations are concentrated among parents with lower education, lower occupational status, or rural hukou. For these parents, both persistent strain and improvement in their worst ties are strongly linked to depressive symptoms, whereas no such associations appear among high-SES parents. These patterns suggest that low-SES parents are simultaneously more vulnerable to relational deterioration (Jiang et al., 2024) and more responsive to relational recovery, reflecting their greater reliance on family-based support in the absence of robust institutional and economic resources. In line with the stress process framework (Pearlin, 1989), these results indicate that socioeconomic resources moderate the psychological consequences of relational change, shaping both exposure to and recovery from relational stressors.

These findings should be interpreted in light of several limitations. First, due to constraints in the CLASS dataset, we rely solely on older parents’ reports of emotional cohesion, functional exchange, and associational connection with each child to measure intergenerational relationships. Prior research indicates that parents’ reports may be positively biased relative to children’s assessments (Lendon et al., 2014), and parent–child reports often diverge in systematic ways (Chen & Chen, 2024). Future research should incorporate dyadic or multi-informant data to capture a more complete picture of intergenerational relationship trajectories. Second, the CLASS panel experiences substantial attrition, and detailed reasons for loss to follow-up are not always documented. This limits the generalizability of our findings, particularly if parents with deteriorating relationships or declining health are more likely to exit the panel. Nonetheless, the trajectory patterns we identify align with those in studies drawing on larger national samples (Chen & Chen, 2024), suggesting some degree of external consistency. Moreover, we conducted multiple imputation for missing covariates to address potential bias from complete-case analysis. As shown in Supplementary Tables 6 and 7, individuals with lower socioeconomic status and poorer health status were more likely to be excluded. Specifically, attrition respondents were more likely to be women, had fewer years of education and lower income, and were more likely to have rural hukou. They were also less likely to own a home and less likely to be married, had slightly more children, and reported worse physical functional status. Because respondents lost to follow-up were disproportionately socioeconomically disadvantaged and in poorer health, attrition likely results in conservative estimates of the relationship between strained ties and depressive symptoms. However, the imputed regression results (Supplementary Table 8) closely align with the complete-case findings, suggesting that the main results are not driven by missing-data patterns and remain stable across analytic specifications. Third, although we used a multidimensional composite index to capture emotional, functional, and associational domains of relationship quality, we were unable to assess whether the underlying measurement properties of this index remained stable across waves. Future research should test for longitudinal measurement invariance or examine each domain separately to better understand potential differences in how emotional and instrumental ties evolve over time. Fourth, although we adjusted for a large set of covariates, including baseline depressive symptoms, and leveraged the temporal ordering of trajectories prior to psychological outcomes, we cannot rule out reverse causality or unobserved confounding. Parents experiencing worsening depressive symptoms may themselves contribute to relationship strain, and our observational design cannot fully separate these processes. These findings should therefore be interpreted as associations rather than evidence of causal effects. Finally, while our theoretical framework identifies several pathways through which intergenerational relationship change may shape psychological well-being (e.g., role-related strain, perceived support, relational resilience), the CLASS data do not include the measures necessary to directly test these mechanisms. Future research should incorporate more detailed assessments of stress processes, coping resources, and relational expectations to clarify how and why certain relationship trajectories affect mental health.

Despite these limitations, this study offers meaningful theoretical and practical contributions to the study of aging and mental health. By tracing intergenerational relationships over 6 years, it shows that parents’ well-being is shaped not only by the quality of their ties but by how those ties evolve. In particular, the finding that the most strained relationship carries disproportionate weight, and that improvements in this tie yield meaningful psychological gains, deepens understanding of relational inequality within families and clarifies how emotional recovery can occur late in life. The results also demonstrate that socioeconomic disadvantage amplifies sensitivity to relational change, highlighting the unequal conditions under which families navigate support, conflict, and repair. Collectively, these insights point to the importance of attending to processes of family change and the structural environments in which they unfold. Practically, the findings suggest that strengthening family communication, investing in programs that help repair strained parent–child relationships, and expanding formal supports for low-SES older adults may bolster mental health in aging populations. More broadly, recognizing relationship trajectories as a meaningful dimension of inequality opens new avenues for research and policy aimed at promoting resilience and well-being in later life.

Supplementary materials

Supplementary material is available at The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences online.

Funding

None declared.

Conflicts of interest

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability

The data in this article are available in the website of the China Longitudinal Aging Social Survey (CLASS), at http://class.ruc.edu.cn/English/Home.htm. The codes for replication can be accessed at https://github.com/jgongsoc/IR_Dep_Project.git.