In China, the “hukou” system operates as a household registration mechanism, functioning akin to a domestic passport (Chan et al., 1999). This system categorizes individuals based on their place of origin, assigning them a “hukou” status that is either “agricultural” (rural) or “non-agricultural” (urban). Chan (2013) delineates two distinct types of internal migration within China considering this system. The first type is characterized by migration with “local” residency rights, commonly referred to as hukou migration. Conversely, the second type involves migration without hukou residency rights and is termed non-hukou migration. In this study, migration is specifically defined as non-hukou migration, concentrating on middle-aged and older adults categorized as migrants. Specifically, the migrant population in this study refers to individuals whose hukou location does not consistent with their actual place of residence.

With the progression of urbanization, the middle-aged and older migrant population in China has surfaced as a pivotal component driving internal migration. Data from 2018 indicates that this group amounted to 13.04 million, constituting 5.3% of the entire migrant population. This demographic segment has witnessed a substantial increase since the turn of the century, rising from 5.03 million in 2000 to 13.04 million by 2015. Furthermore, this growth trajectory represents an average annual growth rate of 6.6%, underscoring the rapid expansion of this population segment (National Health Commission of the People’s Republic of China, 2018). Previous studies have identified family factors as significant determinants in the migration decisions of older adults in China (Lu and Qin, 2014). For example, the primary motivation for migration among urban older individuals is to care for grandchildren, while rural older adults predominantly migrate in search of employment opportunities (Cheng et al., 2019). Additionally, Gao et al. (2021) suggest that individuals in poorer health may be more inclined to migrate, seeking the support and security that coresident adult children can offer.

Studies examining the association between international migration and health showed that immigrants often exhibit better health outcomes than non-immigrants (Halliday and Kimmitt, 2008). This is attributed to two hypotheses: the “healthy migrant effect,” indicating that healthier individuals with better access to resources are more likely to migrate (Chiswick et al., 2008), and the “salmon bias hypothesis,” suggesting that aging immigrants with declining health may return to their origin (Abraido-Lanza et al., 1999). Both hypotheses have been applied to China’s internal migration (Lu and Qin, 2014), but limited research has focused on the aging migrant population in China. This group potentially faces mental health challenges due to the restrictive hukou system, raising various structural and social barriers to restrict migrants’ access to social welfare and entitlement attached to an urban hukou (Zhang et al., 2021) and the disruption of social circles. The resulting lack of emotional support can amplify depression, leading to a higher prevalence of mental health issues among migrant older individuals compared to urban aging groups (Liu et al., 2020). Therefore, protecting the mental health of migrant middle-aged and older adults should be an important public health priority.

To our best knowledge, the majority of the existing Chinese migrant studies treated depression as a whole using scales that summed individual item responses to generate a total score, implying that depressive symptoms are viewed as interchangeable presentations of the same disorder (Guo et al., 2017). However, empirical evidence revealed that individual depressive symptoms were associated with a variety of adverse outcomes, risk factors, and neurological mechanisms (Fried and Nesse, 2014). Additionally, clinical evidence suggests that the central depressive symptoms are more likely to activate other symptoms and, thus, play a significant role in causing the onset of a syndrome and maintaining it (Berlim et al., 2020). For instance, animal and human models imply that restricted sleep causes depression and anxiety symptoms, and hopelessness predicts suicide ideation (Baglioni et al., 2011). Accordingly, targeting central symptoms with biological or psychosocial interventions rather than peripheral symptoms can be highly effective (Borsboom and Cramer, 2013). This indicates that a latent factor model of depression may not fully capture the complexity of the depressive phenomenon. Hence, this study aims to employ network analysis to explore the depression structure and identify the central symptoms of depression in Chinese migrant middle-aged and older adults.

Social engagement, defined as the performance of meaningful social roles for leisure or productive activities, could protect against depression. Strong evidence suggests that activity participation is effective in decreasing depression. Namely, participation in social activities may protect individuals against depression by stimulating multiple bodily systems and reinforcing life-long patterns of attachment (James et al., 2011). However, older adults often lack opportunities for social activities due to diminished physical and motor functionality. Consequently, they are more likely to feel lonely and isolated, which in turn is associated with depression (Hao et al., 2017).

Although social activities are associated with depressive symptoms, the direction and strength of the association depend on the type of social activities. Activity theory of aging, a leading framework for understanding the link between activities and mental health, categorizes social and leisure activity participation into three broad types: informal activities involve socializing with familiar individuals such as relatives, friends, or neighbors; formal activities encompass participation in structured groups and organizations; and solitary activities, such as reading, watching television, and pursuing various hobbies (Lemon et al., 1972). These solitary activities are characterized by the ability to be performed either alone or with limited social interaction (Li and Tang, 2022). Besides, this theory asserts that informal activity is more effective than formal activity. In contrast, solitary activity is the least effective of the three activity types for helping older people improve life satisfaction (Lemon et al., 1972). However, scholars have not agreed on the order of these three types of social activities. For instance, as claimed by Ritchey et al. (2001), informal social activity was related to more aspects of well-being than formal and solitary activity. However, Longino and Kart (1982) suggested that informal activities are positively associated with life satisfaction. Other studies have suggested that solitary activities are not associated with life satisfaction and that formal activities have a negative effect on life satisfaction. Additional Asian studies also approved that informal activities might have a greater impact on geriatric depression than formal social activities in older people (Wang et al., 2020).

In some studies, however, participation in formal activities such as religious organizations and community activities is more beneficial to mental health than participation in other types of social activities in Western countries (Hao et al., 2017). The underlying reason may be that the social significance of social activities varies across cultural contexts. Specifically, previous evidence showed that Asians tend to have a family-oriented culture, and Asian older adults tend to participate in fewer formal activities than Western older adults (Chiao et al., 2011). Moreover, although Longino and Kart (1982) pointed out that there was no association between solitary activities and late-life satisfaction, solitary activities such as internet use could relieve depression among middle-aged and older migrant adults in China (Liu et al., 2020). In detail, older migrants may use the Internet to reconcile their loneliness by participating in activities, interacting with others, and strengthening connections with members of society.

Previous research has focused on exploring the different effects of informal, formal, and solitary activities on depression in older adults at a holistic level. However, the specific pathways of association between these three types of activities on depressive symptoms among migrant middle-aged and older adults are still unclear, and it is necessary to utilize a network analysis approach to deeply analyze how these three types of activities are associated with depressive symptoms and which activities are most effective in alleviating depressive symptoms. Besides, another primary purpose of this study is to understand the current patterns of social activities and depression among older adults in China’s migration context and, therefore, explore the associations between individual depressive symptoms and types of social activities at different levels from various aspects. Older individuals need to examine such patterns to select appropriate life behaviors and thus target them to improve their mental health.

Network analysis, a set of procedures based on the modeling of dynamical systems, provides a visual depiction of the complex associations among symptoms (Beard et al., 2016). Different from the traditional way of treating mental disorders as potential entities that cause symptoms, the network approach emphasizes the connection between the symptoms themselves (Borsboom and Cramer, 2013). Nodes and edges in the network are the elements of the syndrome. The higher the activation degree of the node, the thicker the edge represents the more serious mental disorder (Fried et al., 2017). Analyzing depressive symptoms from the perspective of network analysis allows us to exceed the current average level of symptoms and understand which symptoms may be particularly central to the depression experience (Borsboom, 2017). A growing body of literature utilizes network analysis to detect the depression structure among older individuals. For example, “sadness,” “empty,” “hopeless,” “pessimistic,” “suicidal thought,” and “lassitude” were identified as core symptoms of late-life depression in Korea (Kim et al., 2021). Whereas “death wishes,” “depressed mood,” “loss of interest,” and “pessimism” are identified as the central symptoms of depression among European older adults (Murri et al., 2020). Pan and Liu (2021) claimed that “I felt sad” and “I could not get going” were the two most important depressive symptoms among Chinese widowed older adults. Given the high prevalence of depression in the Chinese immigrant population, it is essential to identify the structure of depressive symptoms in this population.

Additionally, a detailed pathway indicating how factors affect the symptoms of a psychological mental disorder could be clarified when they are involved in the network (Robinaugh et al., 2014). In this study, from the network analysis perspective, depressive symptoms and social activities could be regarded as causally interacting entities, with each subset of the network of factors being different but interdependent. The most important advantage of network analysis is its feasibility of identifying core nodes and associated pathways for timely intervention (Hofmann et al., 2016). Thus, examining the association between different types of social activities and depressive symptoms helps develop tailored prevention and intervention strategies that aim to improve the mental health of migrant middle-aged and older adults in China.

This study employs data from the 2018 China Health and Retirement Longitudinal Study (CHARLS), a nationally representative sample, to examine the symptom central to the network of depression and the relationship between social activities and depressive symptoms among migrant middle-aged and older adults in China through network analysis. Specifically, building on existing research that identified depression among this demographic, this study utilizes a network analysis approach to pinpoint the symptom central to the network of depression in migrant middle-aged and older adults. Besides, we explored the specific ways in which different types of social activities are associated with individual depressive symptoms among the migrant middle-aged and older adults in China.

The China Health and Retirement Longitudinal Study (CHARLS) is a nationally representative sample of the middle- and old-age population (45+) in China. Data for this study come from the 2018 wave of the China Health and Retirement Longitudinal Study (CHARLS). Underpinned by the multistage stratified probability proportional to size (PPS) sampling technique, the CHARLS research team surveys middle- and old-aged community-dwelling residents aged 45 years old and above from 450 villages in 150 counties of 28 provinces in China (Zhao et al., 2020). The baseline CHARLS survey was carried out in 2011 and involved 17,708 individuals, with biannual follow-ups conducted in 2013 and 2015. So far, the latest version of the survey has been conducted in 2018. Data Research Topic covered a wide array of domains, including demographic characteristics, family structure, health status and functioning, health care and insurance, work, retirement and pension, income, and expenditure and assets.

We utilized the Hukou system, which regulates population movement within China, to identify a subsample of middle-aged and older migrants residing in various villages. Initially, 2,346 participants were selected. However, those who did not provide responses to questions about depressive symptoms were excluded, resulting in a final sample size of 1,926 for this research.

Age, gender, education, marital status, hukou, religious faith, living with children, household per capita consumption, chronic disease, and physical disability were used as control variables in this research. In this research, all statistical analyses were performed with R software. We estimated and visualized the structure of depressive symptoms by using the R package of qgraph. This package can automatically compute the appropriate correlations according to the types of variables (Maccallum and Bryant, 2020). We chose Spearman’s rank correlation to estimate regularized partial correlation networks (Epskamp and Fried, 2018) due to the ordinal nature of the item. To control for false positives and create a sparse and interpretable network, the GLASSO procedure used the graphical “least absolute shrinkage and selection operator” (LASSO) regularization with extended Bayesian information criterion (EBIC) model selection (Epskamp and Fried, 2018; Maccallum and Bryant, 2020). We used a regularization approach with tuning parameter γ (specifying the level of sparsity) set to 0.5 (Foygel and Drton, 2010) to compute potential spurious relationships. A psychopathology network is displayed graphically as a set of symptoms (nodes), with nodes connecting through a set of relations (edges). However, recent literature suggested that non-regularized networks might be preferable in some cases, especially for large sample sizes. Since this is not the case for our sample, we presented the non-regularized partial correlation network in Supplementary Figures S1, S2.

First, we computed one network of depressive symptoms, wherein nodes represent observed items and edges represent partial correlation coefficients between two items after conditioning for all other variables in the migrant middle-aged and older adults’ dataset (Epskamp et al., 2018). Blue edges indicate positive correlations, and red edges indicate negative correlations. The stronger the correlation between items, the thicker an edge is. Centrality is commonly used to assess the importance of nodes. We calculated three statistical indices of centrality to identify which items were the core depressive symptoms: Strength (the sum of the absolute values of the number of edges connecting to a node), Betweenness (the number of items a node acts as a bridge along the shortest path between two other nodes) and Closeness (average distance from the node to all other nodes in the network) (Beard et al., 2016).

Second, the relationships among the items of depression, informal activities, formal activities, and solitary activities were investigated using network analysis as well. We estimated one association network, including depressive symptoms, informal activities, formal activities, and solitary activities in the migrant middle-aged and older adults’ dataset, in which nodes and edges also represent observed items and partial correlation coefficients between two items, respectively. In contrast to previous studies that used network analyses in the field of psychopathology, we focused on comparing specific edges rather than centrality measures when networks including social factors were computed. These networks are composed of depressive items and several nodes about social participation factors, making the interpretation of centrality measures indicative of centrality to the network of symptoms problematic (Burger et al., 2020).

Finally, we tested the edge-weight accuracy and centrality stability of these networks by using the bootnet package in R software. The case-dropping bootstrap method was run with 5,000 samples. The correlation stability coefficient (CS-coefficient) was calculated to show the stability of node orders based on the degree index. The edge-weight accuracy was computed using a non-parametric bootstrapping method. Plots of edge-weight accuracy and centrality stability of each network were shown in Supplementary Figures S3–S6.

Table 1 outlines the characteristics of the study participants. The average age of respondents is 60.19 years, with a standard deviation of 9.58 years, and ranges from 45 to 95 years. Participants typically had 6.16 years of education, with a standard deviation of 4.76 years, and education levels varied from none to 22 years. The average household per capita consumption expenditure was 13,725.06 yuan, with a considerable variation, as indicated by a standard deviation of 22,235.95 yuan, and expenditures ranged from zero to 420,620 yuan. Just over half of the participants were female and did not have any chronic illnesses. The majority were married, held an agricultural hukou, lacked religious affiliations, lived without children, and had no functional impairments.

Table 2 shows the mean, standard deviation (SD), min value, max value, skewness, and kurtosis of depressive symptoms. The five highest means of the depressive symptoms were D05 (“I felt hopeful about the future”), D07 (“My sleep was restless”), D08 (“I was happy”), D01 (“I was bothered by things that do not usually bother me”) and D03 (“I felt depressed”), whereas D06 (“I felt fearful”) had the lowest mean.

Figure 1 shows the network frameworks of 10 depressive symptoms-related items of the respondents. In general, all of the items were positively associated with one another within the network. In particular, strong connections emerged between D05 (“I felt hopeful about the future”) and D08 (“I was happy”) (r = 0.311), D09 (“I felt lonely”), and D10 (“I could not get going”) (r = 0.286), D01 (“I was bothered by things that do not usually bother me”), and D03 (“I felt depressed”) (r = 0.277), D02 (“I had trouble keeping my mind on what I was doing”), and D03 (“I felt depressed”) (r = 0.240), D03 (“I felt depressed”), and D04 (“I felt everything I did was an effort”) (r = 0.230).

The depressive symptom network’s centrality indexes of strength, closeness, and betweenness are shown in Table 3. For strength centrality, the five nodes with the highest strength values were D03 (“I felt depressed”), D09 (“I felt lonely”), D10 (“I could not get going”), D04 (“I felt everything I did was an effort”), and D02 (“I had trouble keeping my mind on what I was doing”). For closeness centrality, the five nodes with the highest closeness were D03 (“I felt depressed”), D09 (“I felt lonely”), D10 (“I could not get going”), D04 (“I felt everything I did was an effort”), and D02 (“I had trouble keeping my mind on what I was doing”). For betweenness centrality, the five nodes with the highest betweenness were D03 (“I felt depressed”), D08 (“I was happy”), D04 (“I felt everything I did was an effort”), D09 (“I felt lonely”), and D10 (“I could not get going”). The strength index was more reliable than other indexes (closeness and betweenness) (Robinaugh et al., 2014), and the most important symptom was selected by the value of the strength index. Therefore, the results found that D03 (“I felt depressed”) was the most important depressive symptom among the migrant middle-aged and older people. In general, the CS-coefficient of this network indicated that strength (CS = 0.750), closeness (CS = 0.672), and betweenness (CS = 0.283) all reached the cut-off value of 0.25. The estimated accuracy and stability for edge weights and node centrality showed that the network was relatively stable.

Figure 2 shows the estimated networks with factors for the migrant middle-aged and older people. Overall, the associations between social activities and depressed symptoms were negative.

Interacting with friends (I01) was negatively associated with two depressive symptoms (sorted by decreasing partial correlation): D05 (r = −0.080) and D08 (r = −0.037). Playing mah-jong (I02) was negatively associated with two depressive symptoms (sorted by decreasing partial correlation): D01 (r = −0.035) and D10 (r = −0.032). Taking part in a community-related organization (F01) was negatively associated with D05 (r = −0.053). Going to a sport (S01) was negatively associated with two depressive symptoms (sorted by decreasing partial correlation): D05 (r = −0.040) and D08 (r = −0.033). Using the Internet (S02) was negatively associated with three depressive symptoms (sorted by decreasing partial correlation): D05 (r = −0.088), D10 (r = −0.062), and D04 (r = −0.046). The CS-coefficient indicated that strength (CS = 0.750) and closeness (CS = 0.439) were higher than the minimum values (0.25), while betweenness centrality (CS = 0.050) was unstable.