The Developmental Trajectory of Social Adjustment Among Left‐Behind Children in China: The Predictive Role of Gender, Left‐Behind Status and Parent‐Child Cohesion

ABSTRACT

Few studies have systematically examined social adjustment of left‐behind children. A longitudinal study with three waves was conducted on 512 Chinese left‐behind children to address the gaps. The unconditional latent growth curve model revealed a linear decline in problem behaviors over time, whereas prosocial behaviors and school engagement showed an increasing trend. The conditional latent growth curve model indicated significant differences by gender and left‐behind status in the initial levels of problem behavior, prosocial behavior, and school engagement. Children with different left‐behind statuses exhibited distinct developmental patterns of prosocial behaviors. Parent‐child cohesion played a protective role in problem behaviors, prosocial behaviors, and school engagement. These findings underline the protective role of parent‐child cohesion and highlight the importance of providing targeted social services for different types of left behind children.

1. Introduction

Left‐behind children refer to minors who are left in their household registration place and cannot live with parents because both or one of their parents has been working away from home for more than six months (Duan and Yang 2008). Consistent with this definition, the 2020 Seventh National Population Census estimated that there were approximately 41.77 million such children in China (National Bureau of Statistics of China, 2023). Compared to non‐left‐behind children, left‐behind children face risk factors such as lack of parental care, poor learning management, inadequate living care, insufficient supervision, poor family coping abilities, and lack of communication between children and caregivers, placing them in a state of family adversity (Fan et al. 2023). Long‐term separation resulting from parental labor migration disrupts children's normal family structure and undermines the emotional function of the family. Studies have shown that compared with non‐left‐behind children, left‐behind children are more prone to various social adjustment issues (Chen et al. 2020; Fellmeth et al. 2018), with emotional adjustment (such as depression and loneliness), behavioral adjustment (such as problem behaviors, prosocial behaviors, school engagement) being particularly prominent (Zhao et al. 2019).

However, some research suggests that children possess a certain degree of adaptability to social life, and that their developmental potential may play a positive regulatory role even in a living environment away from their parents (Ma et al. 2018). The adverse effects of early parental departure on the social adjustment of left‐behind children may gradually diminish over the long term of being left behind (Li et al. 2024). Middle childhood represents a critical stage for the formation of an individual's psychological traits and behavioral habits. Previous studies on the social adjustment of left‐behind children have predominantly focused on left‐behind adolescents and often employed short‐term longitudinal design, making it difficult to determine how prolonged experiences of being left behind affect the social adjustment of children in middle childhood. Therefore, it is necessary to examine whether the social adjustment of left‐behind children changes over a longer period, what types of changes occur, and whether protective factors have a positive impact on their social adjustment. Hence, based on previous studies (Wen and Lin 2012; Zhao et al. 2019), this study selected loneliness, depressive symptoms, problem behaviors, prosocial behaviors, and school engagement as indicators of social adjustment for left‐behind children. Through a two‐year, three‐wave follow‐up research design, this study systematically investigated their adjustment development trajectories and analyzed the influencing factors, providing empirical evidence for addressing the social adjustment issues of left‐behind children and promoting healthy development.

Summary

• This three‐wave longitudinal study maps the multidimensional social adjustment trajectories of Chinese left‐behind children, addressing gaps in long‐term developmental dynamics.

• Social adjustment showed heterogeneous trajectories: behavioral domains exhibited positive developmental propensities, whereas emotional domains, like depressive symptoms, demonstrated persistent stagnation.

• Parent‐child cohesion consistently protects multiple social adjustment domains, highlighting its critical role as an actionable intervention target for migrant families.

• Gender and left‐behind status significantly moderate the initial level, highlighting the necessity for stratified, precision‐based interventions targeting high‐risk both‐parent migration subgroups.

1.1. Social Adjustment of Left‐Behind Children

The first objective of this study is to determine the measurement indicators for the social adjustment of left‐behind children. Social adjustment refers to the extent to which an individual adjusts their physical and mental states during interactions with the environment to meet the developmental standards expected by society, which are appropriate for their age and cultural group (Wang et al. 2010). It encompasses two dimensions: emotional adjustment and behavioral adjustment (Niu et al. 2019). Currently, there are two primary approaches for measuring social adjustment. The first approach involves developing distinct measures of social adjustment tailored to different age groups based on their social structures. While this approach can, to a certain extent, reflect the characteristics of social adjustment among children at various developmental stages, it has limitations when it comes to measuring developmental changes across different age groups. The second approach constructs a framework for children's social adjustment by selecting multiple indicators, both positive and negative, within the realm of mental health. This approach employs standardized questionnaires that are not specific to children at a particular stage, making them more suitable for examining the developmental trajectory of children's social adjustment. Given that the focus of this study is to investigate the long‐term development of social adjustment among left‐behind children and its influencing factors, following the second research approach, this study selects corresponding indicators from the two dimensions of emotional adjustment and behavioral adjustment to construct the social adjustment framework for left‐behind children.

Currently, some researchers have investigated the social adjustment of left‐behind children based on the second research approach. For instance, Fan et al. (2009) employed self‐esteem, life satisfaction, loneliness, depression, social anxiety, and problem behaviors as indicators of children's social adjustment. In a recent study, Fan et al. (2023) utilized four indicators—depression, well‐being, problem behaviors, and positive behaviors—to examine the social adjustment of left‐behind children from grades 4 to 9. Su et al. (2015) also measured the social adjustment of left‐behind children using depression, loneliness, and social anxiety. Additionally, Niu et al. (2019), drawing on Fan et al. (2009) ’s earlier research, selected problem behaviors, self‐esteem, depression, and life satisfaction as indicators for assessing the social adjustment of left‐behind children. The above review indicates that, in examining the social adjustment of left‐behind children, researchers have predominantly focused on loneliness and depression as indicators of emotional adjustment and on problem behaviors as an indicator of behavioral adjustment. Accordingly, this study also adopts these three indicators to assess their social adjustment.

Loneliness and depression are representative emotional experiences for left‐behind children (Zhao and Liu 2010). When parents leave, loneliness is often the first emotion that children perceive and is the most representative in their reported emotional responses (Liu et al. 2008). According to social support theory, the more support an individual receives from others, the better equipped they are to cope positively with life's challenges (Qin 2019). Children's social support comes from multiple systems, including parents and other guardians, teachers, and classmates. However, for primary school left‐behind children, due to their parents’ labor migration, family support diminishes, making them more vulnerable to peer victimization or discrimination at school and less likely to feel recognized by teachers. These disruptions in support contribute to significantly higher levels of loneliness among primary school left‐behind children compared to their non‐left‐behind counterparts (Q. Wang et al. 2020). Furthermore, the results of a longitudinal study indicate that the loneliness of left‐behind children shows an upward trend over the course of a year (Huang 2021). This is in contrast to the significant downward trend in loneliness observed among general children (Liu et al. 2013; Zhao 2004).

Additionally, depression symptoms serve as a crucial indicator for measuring children's psychological adjustment. According to the diathesis‐stress theory of depression (Monroe and Simons 1991), the onset of depression is jointly influenced by stress and diathesis. Here, stress broadly refers to major life events and adverse changes in life. For left‐behind children, the departure of their parents represents a significant life event. The lack of direct interaction between parents and children leads to a deficiency in parental care, thereby increasing the risk of depression (Fan et al. 2018). Simultaneously, the absence of parents can also trigger adverse changes in children's lives, such as taking on a substantial amount of household chores and farm work, experiencing more daily setbacks, becoming sensitive and self‐abasement, and losing confidence in themselves, which undoubtedly elevates the risk of depression (Zhang et al. 2023). A one‐and‐a‐half‐year longitudinal study involving four waves of data collection among Chinese children in Grades 2 through 5 revealed an upward trend in their levels of depressive symptoms (Xiao 2023). In contrast, among Western samples, the depression levels of children in grades 4 to 6 exhibited a linear decline (Garber et al. 2002; MaCarty et al. 2013). However, a study on left‐behind adolescents found a decreasing trend in their depression levels (Liu et al. 2024).

Problem behaviors among left‐behind children, due to their potential to disrupt social safety, not only constitute a focal point of concern for researchers (F. Wang et al. 2020) but also represent a critical issue that government departments aim to address. Research has shown that, owing to disruptions in family structure and functioning, left‐behind children in disadvantaged circumstances exhibit more problem behaviors than their non‐left‐behind counterparts (Fan et al. 2010; Hu et al. 2014). The special report of the 19th National Congress of the Communist Party of China pointed out that some left‐behind children exhibit behaviors such as non‐compliance with school regulations, fighting, and conflicts with caregivers, and even engage in illegal activities like extortion and drug abuse. Previous studies have found that the problem behaviors of general children change over time. Specifically, from childhood to adolescence, children's problem behaviors tend to show a significant decline over time (Ji 2007). However, the longitudinal developmental changes in the problem behaviors of left‐behind children remain unknown.

Considering that loneliness, depression, and problem behaviors are all negative indicators, this study selects prosocial behavior and school engagement as positive indicators to assess the social adjustment of left‐behind children. The specific reasons are as follows.

During the process of socialization, individuals are bound to acquire prosocial behaviors that are crucial for the socialization process. Prosocial behaviors encompass a wide range of actions beneficial to others or having an impact on social development, such as sharing, cooperation, altruism, and helping (Wei and Chen 2010). Therefore, prosocial behavior not only serves as an indicator of the moral development of left‐behind children but also represents a key pathway for them to actively integrate into society and establish positive interpersonal relationships. However, previous researchers have predominantly focused on the negative social behaviors of left‐behind children, paying relatively little attention to their positive social behaviors, which undoubtedly increases the likelihood of left‐behind children being labeled as “problem children”. Existing studies have found that there is no significant difference in prosocial tendencies between left‐behind children and general children (Ye 2014). Meanwhile, research on general children has revealed that as an individual's empathy level and moral cognitive ability continuously develop during childhood, their prosocial tendencies also increase accordingly (Lin et al. 2006). Nevertheless, there has been no exploration into the developmental changes in the prosocial tendencies of left‐behind children.

One of the main reasons for parental labor migration is to provide more educational resources for their children and assist them in achieving higher academic levels. Thus, the school engagement and academic performance of left‐behind children are the focal points of their parents’ attention. Additionally, school is the primary setting for children's socialization, and school engagement serves as a key indicator for measuring children's level of social adjustment. For left‐behind children, the emotional deprivation, insufficient guardianship, and lack of academic guidance caused by the absence of parents have a negative impact on their learning engagement, classroom participation, and sense of belonging to the school (Shi et al. 2020; Wen and Lin 2012). Moreover, school engagement is an important resource for them to obtain support and make up for the deficiencies in family functions.

There are two main deficiencies in research on the social adjustment of left‐behind children. On the one hand, most existing studies on the problem behaviors, prosocial behaviors, and school participation of left‐behind children are cross‐sectional, lacking follow‐up studies on their dynamic developmental trajectories. Especially during middle childhood, a critical stage for the formation of psychological and behavioral habits, current research has failed to explain how social adjustment in middle childhood evolves over time. On the other hand, longitudinal studies on the loneliness and depression of left‐behind children have mostly been multiple longitudinal surveys conducted over relatively short periods, with the research subjects primarily being left‐behind adolescents. This makes it impossible to clarify the impact of a longer period of being left behind on the social adjustment of children in middle childhood. Overall, this study selects depression and loneliness as indicators of emotional adjustment, and problem behaviors, prosocial behaviors, and school participation as indicators of behavioral adjustment. Together, these five indicators serve as measures of social adjustment, used to examine the developmental trajectories of left‐behind children.

1.2. The Influencing Factors of Social Adjustment of Left‐Behind Children

The second objective is to explore the impact of individual characteristics and environmental factors on the developmental trajectories of the five social adjustment indicators among left‐behind children. The participants in this study were in middle childhood, which is a critical turning point in forming views about their talent and future development. Exploring the impact of the left‐behind environment on the social adjustment trajectories of left‐behind children has an important guiding significance for their future development.

The social adjustment of left‐behind children is influenced by various factors such as individuals and families. In terms of individual factors, previous studies have not reached a unified consensus on the gender and left‐behind status differences in the social adjustment of left‐behind children. In terms of gender, some research has suggested that girls are more independent and resilient than boys, and have a greater understanding of the hardships their parents face while working away from home, resulting in a lower incidence of psychological problems among girls compared to boys (Zhang et al. 2012). However, other studies have pointed out that girls have weaker self‐protection skills and are more likely to respond in silence when facing harm, which prevents them from effectively seeking help from others (Wu et al. 2019). The observed inconsistencies in prior research may stem from their predominant reliance on cross‐sectional designs, which primarily examine static gender differences in social adjustment among left‐behind children across discrete age groups. This approach overlooks how gender factors dynamically influence developmental trajectories of social adjustment over time. Furthermore, stage‐specific shifts in gender roles, which may significantly impact adjustment patterns, cannot be adequately captured through cross‐sectional data. Therefore, to clarify the impact of gender on the social adjustment of left‐behind children, it is necessary to adopt a longitudinal perspective to examine developmental differences in social adjustment between left‐behind children of different genders.

Left‐behind status is categorized based on the parental migrated situation into father‐only migration, mother‐only migration, and both‐parent migration (Wen and Lin 2012). The seventh national population census found that, compared to both‐parent migration (45.6%), mother‐only migration (29.8%) and father‐only migration (24.6%) were less common (National Bureau of Statistics of China, 2023). The left‐behind status constitutes a macro system in the development process of left‐behind children. It can have direct or indirect influences on the development of these children. However, evidence regarding the impact of different left‐behind statuses on children's development is contradictory. Some studies suggest that both parents working away can increase household income, which may increase investments in children's health and future development (Guo 2012). Conversely, other researchers argue that the higher income of families in which both parents have migrated does not offset the negative effects of parental absence (Zhou et al. 2015). Rather, children whose parents have both migrated are reported to have the poorest social adjustment, as they are often cared for by grandparents, relatives, or even themselves, undertake more household chores and labor, and may lack assistance when facing difficulties (Wang et al. 2019). Similarly, research has shown that compared to children with father‐only migration, those with both‐parent migration have weaker family nurturing capabilities and are more likely to live in poverty (Shi et al. 2020). Additionally, mothers are considered more important for children's growth than fathers (Deng and Li 2014; Wen and Lin 2012). However, these prior studies predominantly employ cross‐sectional designs and fail to examine the impact of different left‐behind statuses on children's social adjustment from a developmental perspective. This methodological limitation may account for the inconsistencies observed in existing literature. Specifically, while parental migration inflicts immediate emotional harm on left‐behind children, the improvements in family material conditions resulting from migration typically exhibit delayed effects. Thus, the present study addresses this gap by employing a longitudinal design to investigate how left‐behind status shapes the developmental trajectories of children's social adjustment, thereby overcoming the constraints of previous research.

According to ecological systems theory, family is a critical factor for the normal development of children (Bronfenbrenner and Morris 2006), and Steinberg (2001) believed that among family factors, the quality of parent‐child relationships has the greatest impact on children's development. Parent‐child cohesion is an important indicator of the quality of parent‐child relationships, a protective factor for children's healthy development, and one of the basic features of normal family functioning (Bean et al. 2006). Research shows that parent‐child cohesion has a protective effect on children's psychological, behavioral, and school adjustment (Li 2020), effectively promoting positive individual development. For left‐behind children, although parental absence leads to a decrease in parent‐child cohesion, advances of communication technology have alleviated the emotional deficits caused by parent‐child separation. Studies indicate that parent‐child cohesion has both direct and indirect protective effects on the development of left‐behind children. On one hand, parent‐child cohesion can significantly reduce negative emotions and problem behaviors among left‐behind children (Zhang et al. 2019). On the other hand, parent‐child cohesion can indirectly influence the emotional experiences of left‐behind children through mediating variables (e.g., adversity beliefs: Zhao et al. 2017). Furthermore, research shows that parent‐child cohesion can negatively predict depressive symptoms and loneliness in left‐behind children one year later (Zhang et al. 2019). Therefore, to further deepen research in this field, this study also examined the impact of parent‐child cohesion on the developmental trajectories of social adjustment in left‐behind children.

1.3. The Current Study

In summary, through a two‐year longitudinal investigation with three waves of data collection, this study aimed to describe the longitudinal developmental trajectories of five social adjustment indicators (depressive symptoms, loneliness, problem behaviors, prosocial behaviors, and school engagement) in left‐behind children to gain a deeper understanding of the long‐term impact of being left behind on children's psychological and behavioral adjustment. Additionally, the effects of individual characteristics (gender, left‐behind status) and environmental factors (parent‐child cohesion) on the social adjustment trajectories of left‐behind children were investigated to identify the factors that can affect the developmental trajectories and identify the direction for the development of intervention measures.

2. Methods

2.1. Participants

Fourth‐grade left‐behind children from 11 rural primary schools in three central provinces (Shanxi, Sichuan, and Guizhou) were selected as participants. The first survey (T1) was conducted in April 2016, and subsequent surveys were administered annually, resulting in three surveys. The final sample comprised 512 participants: 233 males (45.5%) and 279 females (54.5%). The average age was 9.67 years (SD = 0.73). To control for the potential confounding effects of major negative family life events (e.g., parental divorce or bereavement) on children's psychological development, all participants in the current study were strictly limited to those from two‐parent intact families (i.e., both parents are living and not divorced) and had no chronic or acute illnesses. Most absent parents were engaged in construction‐, factory production‐, or service‐related work. In this study, some left‐behind children might have experienced attrition because they dropped out of school or moved with their parents. At Wave 1 (T1), 512 fourth‐grade elementary school students participated in the questionnaire survey. At Wave 2 (T2), 50 students did not participate in the follow‐up survey, while at Wave 3 (T3), 63 students were absent. First, we defined missing data using a single‐wave absence criterion (i.e., missing at either T2 or T3). Logistic regression analyses were conducted separately for each outcome variable, including depressive symptoms, loneliness, problem behaviors, prosocial behaviors, and school engagement, to examine whether missingness at T2 or T3 was associated with these variables. Results indicated that all models were non‐significant (p > 0.05), suggesting that single‐wave missing data may be missing completely at random (MCAR). Subsequently, we redefined missing data as absence at both T2 and T3. Logistic regression analyses were repeated for the same outcome variables based on this new criterion. Results again showed non‐significant associations (p > 0.05), supporting the assumption that two‐wave missing data may also be MCAR. Therefore, all participants were included in the primary analyses. We employed full information maximum likelihood (FIML) estimation to handle missing data in the latent variable growth models. Additionally, since all outcome variables (depressive symptoms, loneliness, problem behaviors, prosocial behaviors, and school engagement) exhibited non‐normal distributions, robust maximum likelihood estimation was utilized to address non‐normality. Finally, longitudinal measurement invariance testing revealed that the depression (ΔCFI = 0.009, ΔRMSEA = 0.001), loneliness (ΔCFI = 0.009, ΔRMSEA = 0.001), and prosocial behavior (ΔCFI = 0.007, ΔRMSEA = 0.001) scales all met the scalar invariance criteria of Chen (2007) (ΔCFI ≤ 0.010, ΔRMSEA ≤ 0.015). For the problem behavior (ΔCFI = 0.014, ΔRMSEA = 0.001) and school engagement (ΔCFI = 0.012, ΔRMSEA = 0.006) scales, although the CFI changes slightly exceeded the recommended threshold, the minimal RMSEA changes and good absolute fit indices indicated that acceptable scalar invariance was achieved.

2.2. Instruments

2.2.1. Depressive Symptoms

The Depression Self‐Rating Scale for Children, developed by Birleson (1981) and revised by Su et al. (2003), was used. It consists of 18 items (e.g., “Sleep soundly”), with 10 items scored in reverse, and each item was rated using a three‐point Likert scale from 0 (never) to 2 (often). Higher scores indicated higher levels of depressive symptoms. This questionnaire has been widely used in Chinese child populations (Su et al. 2003) and has good reliability and validity. The internal consistency coefficients of the three measurements were 0.69, 0.72, and 0.71. Confirmatory factor analysis (CFA) demonstrated that the scale possesses sound psychometric properties among the sample of rural left‐behind children.

2.2.2. Loneliness

The Chinese version of the Loneliness Scale (LS) developed by Asher et al. (1984) was used, consisting of 16 items (e.g., “No one talks to me”), with 10 items scored in reverse, and each item was rated using a five‐point Likert scale from 1 (always) to 5 (never). Higher scores indicated higher levels of loneliness. In the present study, the internal consistency alpha coefficients of the scale across three administrations were 0.81, 0.85, and 0.88, respectively.

2.2.3. Problem Behaviors

Problem behavior was assessed using a 12‐item scale adapted by Shen (2009), which integrates selected items from the antisocial behavior subscale of the Child Behavior Checklist (Achenbach and Edelbrock 1987) and the revised Problem Behavior Scale (Fang and Li 1996). A representative item is, “I often hang out with troublemakers.” Participants rated the items on a 4‐point Likert scale ranging from 1 (never) to 4 (often), with higher aggregate scores reflecting greater levels of problem behavior. This specific adjustment has demonstrated robust reliability and validity in prior domestic research (Zhao and Liu 2010). In the current study, the internal consistency coefficients (Cronbach's α) across the three measurement waves were 0.77, 0.80, and 0.80, respectively. CFA demonstrated that the scale possesses sound psychometric properties among the sample of rural left‐behind children.

2.2.4. Prosocial Behaviors

Selected items from the prosocial behavior section of the Child Behavior Checklist developed by Achenbach and Edelbrock (1987) (e.g., “I comfort classmates who are upset”) were used, totaling 4 items. To more accurately measure the prosocial behaviors of left‐behind children, the original three‐point scoring system was modified to a four‐point system, where 1 represents “never” and 4 represents “often”. Higher scores indicated higher levels of prosocial behavior. In this study, the internal consistency alpha coefficients for the three administrations were 0.71, 0.70, and 0.75, respectively.

2.2.5. School Engagement

The School Engagement Questionnaire developed by Wen and Lin (2012) was used, consisting of five items (e.g., “I can go to school on time”), rated on a scale from “1” (rarely) to “3” (always). Higher scores indicate higher levels of school engagement. This questionnaire has been widely used with rural adolescent populations in China and has good reliability and validity (Wen and Lin 2012). In this study, the internal consistency alpha coefficients were 0.72, 0.75, and 0.76.

2.2.6. Parent‐Child Cohesion

The Parent‐Child Cohesion Subscale of the Family Adaptation and Cohesion Evaluation Scale (FACES) II, developed by Olson et al. (1979), was used. It consists of 12 items (e.g., “I feel very close to my mother/father”), with 4 items scored in reverse. The scale was divided into two identical questionnaires, one for fathers and one for mothers, and each item was rated using a 5‐point Likert scale ranging from 1 (almost never) to 5 (almost always). Higher scores indicated higher levels of parent‐child cohesion. In this study, the internal consistency alpha coefficients were 0.83.

2.3. Research Procedure

The present study was approved by the Institutional Review Board of [Blinded for Review] (Approval No.: [Blinded for Review]). A cluster sampling approach was employed, utilizing classes as the primary sampling units; all students within the selected classes were included in the survey. Prior to the formal assessment, informed consent was obtained from all participating students and their legal guardians. To ensure response accuracy and data quality, particularly for younger children, stringent administration controls were implemented. The assessments were administered by uniformly trained researchers with expertise in psychology or education. Before the assessment began, administrators read the instructions aloud, utilizing concrete life examples to elucidate abstract emotional vocabulary and frequency scales, followed by a comprehension check. During the assessment, administrators controlled the pacing by reading each item aloud sequentially. They simultaneously monitored the classroom to address any reading or comprehension inquiries, ensuring that each participant completed the questionnaire independently with a full understanding of the items. The administration procedures remained strictly consistent across all three time points (Time 1, Time 2, and Time 3).

2.4. Data Analysis

SPSS (version 25.0) was used for data entry, organization, and correlation analyses. To address the research objectives, Mplus 8.3 software was employed to conduct latent growth curve modeling (LGCM) on three‐wave longitudinal data assessing social adjustment among left‐behind children, encompassing depressive symptoms, loneliness, problem behaviors, prosocial behaviors, and school engagement. This modeling approach yielded estimates of the initial level (intercept) at Grade 4 and the developmental rate (slope) from Grade 4 to Grade 6 for each outcome variable. Subsequently, these intercept and slope factors served as dependent variables in a conditional latent growth model. The model incorporated father‐child cohesion, mother‐child cohesion, left‐behind status (with father‐only migration as the reference category), and gender (with male as the reference category) as independent variables. Model fit was evaluated using a comprehensive set of indices: Parsimony‐adjusted indices: χ²/df ratio; Relative fit indices: Comparative Fit Index (CFI) and Tucker‐Lewis Index (TLI); Absolute fit indices: Root Mean Square Error of Approximation (RMSEA) and Standardized Root Mean Square Residual (SRMR). Following established criteria, good model fit was indicated by: χ²/df < 3, CFI > 0.95, TLI > 0.90, RMSEA < 0.08, SRMR < 0.08 (Byrne 2012).

3. Results

3.1. Descriptive Statistics and Correlation Analysis

Table 1 presents the means and standard deviations of depressive symptoms, loneliness, problem behaviors, prosocial behaviors, and school engagement across the three waves of data. Table 2 shows the bivariate correlations between individual characteristics, parent‐child cohesion, and dependent variables at time point 1 (T1). The correlation analysis results indicated that gender was significantly correlated with mother‐only migration, both‐parent migration, problem behaviors, prosocial behaviors, and school engagement. Apart from father‐child cohesion, mother‐only migration was significantly correlated with all other variables in this study. Additionally, father‐child cohesion, mother‐child cohesion, depressive symptoms, loneliness, problem behaviors, prosocial behaviors, and school engagement were significantly correlated.

TABLE 1.

The mean and standard deviation of the main variables.

	Time 1		Time 2		Time 3
Variables	M	SD	M	SD	M	SD
Depressive symptoms	12.81	4.94	12.54	4.88	12.46	4.85
Loneliness	35.74	10.47	32.96	10.65	34.08	11.31
Problem behaviors	16.29	4.18	16.16	4.33	15.69	4.12
Prosocial behaviors	10.74	2.90	11.20	2.74	11.19	2.91
School engagement	10.72	2.31	11.14	2.41	11.44	2.26

TABLE 2.

The correlation between personal characteristics, parent‐child cohesion and social adjustment (Time 1).

Variables	1	2	3	4	5	6	7	8	9	10
1. Gendera	1
2. Mother‐only migration b	−0.09 *	1
3. Both‐parent migration c	0.09 *	−.29 **	1
4. Father‐child cohesion	0.07	−.06	−0.05	1
5. Mother‐child cohesion	0.03	−.12 **	−0.10 *	0.51 **	1
6. Depressive symptoms	−0.02	.10 *	−0.01	−0.47 **	−0.41 **	1
7. Loneliness	0.02	.11 *	−0.03	−0.43 **	−0.40 **	0.49 **	1
8. Problem behaviors	−0.23 **	.10 *	−0.01	−0.32 **	−0.30 **	0.28 **	0.31 **	1
9. Prosocial behaviors	0.12 **	−0.08	−0.04	0.41 **	.34 **	−0.29 **	−0.33 **	−0.32 **	1
10. School engagement	0.10 *	−0.10 *	−0.10	.40 **	.34 **	−.27 **	−.33 **	−.37 **	0.50 **	1

^aGender: 0 = male, 1 = female.

^bMother‐only migration: 0 = Father‐only migration, 1 = Mother‐only migration.

^cboth‐parent migration: 0 = Father‐only migration, 1 = Both‐parent migration.

^* p<.05, ^** p<.01.

3.2. Trajectories of Social Adjustment

To examine the developmental trends of depressive symptoms, loneliness, problem behaviors, prosocial behaviors, and school engagement among left‐behind children, unconditional latent growth curve models (LGCMs) were first constructed to investigate the developmental trajectories of social adjustment. The fit indices for the unconditional LGCMs of left‐behind children's social adjustment are presented in Table 3. The unconditional LGCMs for depressive symptoms, problem behaviors, and school engagement demonstrated good fit. The fit indices for the unconditional LGCM of prosocial behavior were acceptable. However, neither the unconditional nor the conditional latent growth models for loneliness achieved adequate fit indices. This indicates that a simple linear growth modeling approach is insufficient to accurately delineate the developmental dynamics of loneliness within the current sample. It is plausible that loneliness exhibits higher‐order non‐linear fluctuations or is disproportionately confounded by measurement error over time. Specifically, when the target variable exhibits significant individual differences or is influenced by unmeasured external factors, the unconditional model fails to account for this variance, resulting in poor model fit. Furthermore, as the study utilized three time points of data, it was not feasible to specify a nonlinear growth model incorporating a quadratic term.

TABLE 3.

Fit indices of unconditional latent growth model of social adjustment of left‐behind children.

Variables	χ2	df	χ2/df	CFI	TLI	RMSEA	SRMR
Depressive symptoms	0.33	1	0.33	1.00	1.00	0.00	0.01
Loneliness	14.74	1	14.74	0.91	0.73	0.16	0.04
Problem behaviors	0.71	1	0.71	1.01	1.00	0.00	0.01
Prosocial behaviors	3.45	1	3.45	0.98	0.94	0.07	0.03
School engagement	0.68	1	0.68	1.00	1.01	0.00	0.01

Consequently, to maintain strict analytical rigor, no substantive conclusions predicated on this inadequately fitted model will be drawn in the ensuing discussion.

For the unconditional LGCM of depressive symptoms, the mean of the intercept factor was significant (M = 12.77, p < 0.001), and the variance of the intercept factor was also significant (σ² = 8.72, p < 0.001), indicating significant individual differences in the initial level of depressive symptoms among left‐behind children. However, the variance of the slope factor was negative (σ² = −1.25, p = 0.334), violating the statistical assumption of non‐negative variance. This suggests that the current linear growth model may not adequately capture the true pattern of change in depressive symptoms, necessitating further model modification for verification. To address this model misspecification, a boundary constraint method was employed, fixing the slope variance at zero to ensure model identifiability and the validity of parameter estimates. The constrained model retained individual differences only in the intercept factor, while the slope factor reflected the group‐level trend. The constrained model demonstrated good fit (χ²(3) = 0.52, p = 0.668, RMSEA = 0.00, SRMR = 0.02, CFI = 1.00, TLI = 1.01). Overall, depressive symptoms among left‐behind children remained stable (M = −0.12, p = 0.36).

Regarding problem behaviors, both the mean of the intercept factor (M = 16.34, p < 0.001) and the mean of the slope factor (M = −0.26, p = 0.023) were significant, indicating that the level of problem behaviors among left‐behind children showed a downward trend over the two‐year period. The variance of the intercept factor was significant (σ² = 6.80, p < 0.001), while the variance of the slope factor was not significant (σ² = 0.90, p = 0.394). This indicates significant individual differences in the initial level of problem behaviors, but no significant individual differences in the rate of change. Additionally, the covariance between the intercept and slope factors was not significant (r = −0.42, p = 0.717).

In contrast, for prosocial behavior, both the mean of the intercept factor (M = 10.81, p < 0.001) and the mean of the slope factor (M = 0.20, p = 0.011) were significant, indicating that the level of prosocial behavior among left‐behind children showed an upward trend over the two‐year period. The variance of the intercept factor was significant (σ² = 3.37, p < 0.001), while the variance of the slope factor was not significant (σ² = 0.66, p = 0.103). This indicates significant individual differences in the initial level of prosocial behavior, but no significant individual differences in the rate of change. Furthermore, the covariance between the intercept and slope factors was not significant (r = −0.49, p = 0.302).

Similarly, for school engagement, the mean of the intercept factor was significant (M = 10.74, p < 0.001), and the variance of the intercept factor was also significant (σ² = 1.76, p < 0.001), indicating significant individual differences in the initial level of school engagement among left‐behind children. However, the variance of the slope factor was negative (σ² = −0.33, p = 0.194), violating the statistical assumption of non‐negative variance. This suggests that the current linear growth model may not adequately capture the true pattern of change in school engagement, necessitating further model modification for verification. To address this model misspecification, a boundary constraint method was employed, fixing the slope variance at zero to ensure model identifiability and the validity of parameter estimates. The constrained model retained individual differences only in the intercept factor, while the slope factor reflected the group‐level trend. The constrained model demonstrated good fit (χ²(3) = 0.90, p = 0.439, RMSEA = 0.00, SRMR = 0.05, CFI = 1.00, TLI = 1.00). Overall, school engagement among left‐behind children showed a significant upward trend (M = 0.34, p < 0.001).

Focusing on the core “slope mean” parameter across the four successfully fitted outcome variables (depressive symptoms, problem behavior, prosocial behavior, and school engagement), we applied the Benjamini–Hochberg (B–H) False Discovery Rate (FDR) procedure to adjust the p‐values, as widely recommended in current literature (Benjamini and Hochberg 1995). Following the B‐H FDR correction, all variables that demonstrated significant trajectories in the initial analyses retained their statistical significance. Specifically, the slopes for problem behavior (original p = 0.023, adjusted p = 0.031), prosocial behavior (original p = 0.011, adjusted p = 0.022), and school engagement (adjusted p < 0.001) consistently remained below the 0.05 significance threshold. Conversely, the slope for depressive symptoms remained non‐significant.

3.3. The Predictions of Gender, Left‐Behind Status and Parent‐Child Cohesion on Social Adjustment Trajectories

Building upon the unconditional latent growth models, conditional latent growth models were specified by incorporating baseline (Time 1) predictors: father‐child cohesion, mother‐child cohesion, left‐behind status (dummy‐coded, with father‐only migration as the reference group, contrasting mother‐only and both‐parent migration), and gender (with male as the reference group).

Regarding depressive symptoms and school engagement, the unconditional models initially yielded non‐significant negative variances. To investigate whether these negative variances were artifacts of omitted variable bias, the initial construction of the conditional models attempted to freely estimate the slope variances, allowing the aforementioned covariates to predict them. However, the free estimation resulted in a non‐positive definite latent variable covariance matrix, and the residual variances of the slopes remained negative (σ² = −0.39; σ² = −0.17). Consequently, the standardized estimates pertaining to the slopes failed to converge. This statistical anomaly (non‐positive definite matrix) indicates that the current linear model specification could not reliably estimate the inter‐individual variability in the developmental rates of these two variables within the constraints of the three‐wave data. Therefore, adhering to established methodological guidelines for latent growth modeling (Bollen and Curran 2006), the current study ultimately adopted a boundary‐constrained approach by strictly fixing the slope variances to zero. To ensure the validity and interpretability of the parameter estimates, the final models only examined the predictive effects of the covariates on the initial intercepts of depressive symptoms and school engagement. As detailed in Table 4, the conditional models demonstrated robust fit for depressive symptoms, problem behavior, prosocial behavior, and school engagement.

TABLE 4.

Fit indices of conditional latent growth model of social adjustment of left‐behind children.

Variables	χ2	df	χ2/df	CFI	TLI	RMSEA	SRMR
Depressive symptoms	14.55	13	1.12	1.00	0.99	0.02	0.04
Loneliness	18.07 **	6	3.01	0.96	0.88	0.06	0.02
Problem behaviors	7.08	6	1.18	1.00	0.99	0.02	0.02
Prosocial behaviors	12.71 *	6	2.12	0.97	0.91	0.05	0.05
School engagement	22.79 *	13	1.75	0.97	0.96	0.04	0.06

^* p < 0.05, ^** p < 01

As shown in Table 5, regarding the prediction of initial levels, both father‐child cohesion and mother‐child cohesion exhibited significant negative effects on the intercepts of depressive symptoms (β = −0.35, p < 0.001; β = −0.34, p < 0.001) and problem behavior (β = −0.14, p = 0.034; β = −0.39, p < 0.001) among left‐behind children. Concurrently, they robustly and positively predicted the initial levels of prosocial behavior (β = 0.38, p < 0.001; β = 0.25, p = 0.003) and school engagement (β = 0.30, p < 0.001; β = 0.22, p < 0.001). Furthermore, concerning gender differences, girls exhibited significantly lower baseline scores for problem behavior compared to boys (β = −0.34, p < 0.001), while demonstrating significantly higher initial scores for both prosocial behavior (β = 0.16, p = 0.008) and school engagement (β = 0.19, p < 0.001). Regarding parental migration patterns, relative to the reference group of father‐only migration, children experiencing both‐parent migration displayed significantly lower initial levels of school engagement (β = −0.11, p = 0.031). However, the migration type configurations did not yield significant predictive effects on the initial levels of the remaining three indicators. In terms of predicting developmental trajectories (slopes), the covariate cluster demonstrated limited explanatory power for the slopes of most indicators. Specifically, none of the covariates significantly predicted the developmental slopes of depressive symptoms, problem behavior, or school engagement. Nevertheless, within the conditional model for prosocial behavior, father‐child cohesion significantly and negatively predicted the developmental slope (β = −0.25, p = 0.021). Additionally, compared to father‐only migration, both‐parent migration exerted a significant positive predictive effect on the developmental trajectory of prosocial behavior (β = 0.20, p = 0.034).

TABLE 5.

Path coefficients for parent‐child cohesion, left‐behind status and gender on growth factors.

Variables	Depressive symptoms				Problem behaviors				Prosocial behaviors				School engagement
	intercept		slope		intercept		slope		intercept		slope		intercept		slope
	B(SE)	β	B(SE)	β	B(SE)	β	B(SE)	β	B(SE)	β	B(SE)	β	B(SE)	β	B(SE)	β
Gender	0.37(0.30)	0.06			−1.80(0.33) ***	−0.34	0.19(0.22)	0.10	0.63(0.23) **	0.16	−0.03(0.16)	−0.02	0.60(0.15) ***	0.19
Mother‐only migration	0.66(0.45)	0.07			0.62(0.55)	0.08	−0.23(0.33)	−0.08	−0.31(0.35)	−0.06	0.34(0.22)	0.12	0.33(0.24)	−0.07
Both‐parent migration	−0.30(0.33)	−0.05			0.16(0.35)	0.03	0.21(0.24)	0.11	−0.31(0.24)	−0.08	0.40(0.17) *	0.20	−0.35(0.16) *	−0.11
Mother‐child cohesion	−0.21(0.04) ***	−0.35			−0.20(0.04) ***	−0.39	0.02(0.03)	0.09	0.09(0.03) **	0.25	0.00(0.02)	−0.00	0.06(0.02) ***	0.22
Father‐child cohesion	−0.20(0.03) ***	−0.34			−0.07(0.03) *	−0.14	0.05(0.02) *	0.27	0.14(0.03) ***	0.38	−0.04(0.02) *	−0.25	0.09(0.02) ***	0.30
R2(SE)	0.37(0.05) ***				0.38 (0.10) ***		0.15(0.18)		0.35(0.08) ***		0.11(0.07)		0.28(0.05) ***

^* p < 0.05; ^**p < 0.01; ^*** p < 0.001.

Regarding explanatory capacity, the covariates described above (mother‐child cohesion, father‐child cohesion, gender, and migration type) accounted for a significant proportion of individual differences in the inter‐individual variances in the initial levels of the social adjustment indicators. The proportions of variance explained were 38% for problem behavior, 36.9% for depressive symptoms, 35% for prosocial behavior, and 28% for school engagement. By contrast, the total variance explained by these covariates for the developmental trajectories did not reach statistical significance for any of the outcomes.

4. Discussion

This Longitudinal Study Employed a Three‐wave Design Over Two Years to Investigate the Dynamic Development of Social Adjustment Among Left‐Behind Children, With Particular Emphasis on the Effects of Parent‐child Cohesion, Gender, and Left‐Behind Status. Our Findings Provide Novel Insights Into the Protective Role of Parent‐child Cohesion in Social Adjustment Processes, Highlighting Key Factors That May Inform Interventions to Promote Positive Development Among Left‐Behind Children.

4.1. The Developmental Trajectories of Social Adjustment Among Left‐Behind Children

This study found that among the four indicators of social adjustment for left‐behind children, three showed significant changes. Specifically, problem behaviors decreased, while prosocial behaviors and school engagement increased. These findings indicate that, over the two‐year longitudinal window, left‐behind children exhibited positive developmental trajectories across multiple facets of social adjustment. At the individual level, children in Grades 4 through 6 continue to growth in self‐regulatory and cognitive capacities (Blakemore and Choudhury 2006), which may equip them to better manage problem behaviors and engage actively in school and social activities (Ma and Su 2020). At the contextual level, when parental migration compromises the availability of day‐to‐day family supervision and emotional support, schools may assume a compensatory socialization function. The positive adjustment trajectories may reflect a synergistic interplay between the normative maturation of self‐regulatory abilities and compensatory school‐based support. However, the present study did not directly measure school environment variables; the compensatory mechanism discussed above therefore remains inferential. Furthermore, the absence of a non‐left‐behind comparison group precludes a clear determination of the extent to which the observed positive trajectories are specific to left‐behind children, as opposed to reflecting broader normative developmental patterns characteristic of this age period.

In contrast to the three indicators that showed significant change, depressive symptoms remained relatively stable across the two‐year study period. This null finding is consistent with some prior work (Lei et al. 2021), although other studies have documented rising depressive symptoms among left‐behind children over a one‐year interval (Li et al. 2024). Differences in follow‐up duration and measurement spacing may partly account for these inconsistencies: depressive symptoms may exhibit short‐term fluctuations in response to stressful life events while reverting toward individual baseline levels over longer observation windows. Furthermore, the positive developmental changes observed in problem behavior, prosocial behavior, and school engagement likely benefited from the normative maturation of self‐regulatory capacities and compensatory school‐based support (as discussed above), but these mechanisms may offer limited buffering against internalizing distress. Depressive symptoms in middle childhood are characterized by relatively high rank‐order stability and are strongly constrained by temperamental affective reactivity styles and chronic psychosocial stressors, rendering them less susceptible to change through cognitive maturation or isolated environmental factors alone (Klein et al. 2011).

In summary, the social adjustment of left‐behind children showed a trend toward continuous improvement or stability, consistent with longitudinal evidence from children exposed to other adverse environments (Lima et al. 2020). However, longer‐term monitoring is required to confirm the sustainability of this improvement.

4.2. The Predictive Effects of Father‐child Cohesion, Mother‐child Cohesion, Left‐Behind Status and Gender on Social Adjustment Trajectories

The present study found significant gender differences in left‐behind children's social adjustment at T1. Specifically, relative to boys, girls exhibited lower levels of problem behavior and higher levels of prosocial behavior and school engagement. This pattern aligns with prior findings (Cui et al. 2021; Wen and Lin 2012) and is consistent with the robust cross‐cultural finding in the developmental literature that girls tend to outperform boys in behavioral regulation and prosociality (Else‐Quest et al. 2006). These results suggest that gender differences observed among left‐behind children at least partly reflect normative developmental features characteristic of this age period, and that the experience of being left behind does not erase these differences.

The present study further revealed that, compared with children from father‐only migration families, children from both‐parent migration families scored significantly lower on prosocial behavior and school engagement at T1. This finding converges with prior evidence (Fan et al. 2009; Wen and Lin 2012). In rural Chinese families, mothers typically serve as the primary providers of daily care, emotional support, and behavioral socialization. When only the father migrates, the mother remains present and continues these functions. By contrast, both‐parent migration entails the physical absence of the mother, and children in these families are predominantly placed under grandparental guardianship. Owing to grandparents’ limitations in physical stamina, educational awareness, and disciplinary capacity, they often struggle to provide adequate academic guidance and behavioral regulation, with their caregiving largely centered on meeting children's basic physical needs (Wen and Lin 2012). These children thus face the compounding of two layers of disadvantage: the loss of direct maternal caregiving and the limited quality of surrogate care. This may help explain their lower baseline levels of prosocial behavior and school engagement.

Furthermore, the present study found that left‐behind children from both‐parent migration families showed a steeper upward trajectory in prosocial behavior across the two‐year period than did children from father‐only migration families. Two accounts warrant consideration. First, given the lower baseline levels observed in the both‐parent‐migration group, part of the apparent increase may reflect regression to the mean, that is, a natural rebound from a transiently low score, rather than substantive developmental change. Second, children from both‐parent migration families may accrue compensatory socialization resources through peer relationships and teacher‐student interactions within the school context. It is important to emphasize, however, that both accounts remain at the level of inference. Future research should systematically incorporate longitudinal measures of peer support, teacher‐student relationship quality, the frequency and quality of remote family communication, and children's cognitive maturation, so as to directly evaluate the relative explanatory power of these competing or complementary hypotheses.

Consistent with prior findings (Song et al. 2017), the present study provides longitudinal evidence for the protective role of parent‐child cohesion in the social adjustment of left‐behind children. At baseline (T1), left‐behind children who reported higher levels of parent‐child cohesion scored lower on depressive symptoms and problem behavior, and higher on prosocial behavior and school engagement. These results converge with cross‐sectional evidence (Zhao et al. 2013), indicating that parent‐child cohesion plays a pivotal role in fostering positive social adjustment among left‐behind children. High‐quality parent‐child cohesion provides a secure base that sustains children's self‐worth, emotional security, and internalization of behavioral norms despite parental physical absence (Bowlby 1982; Eisenberg et al. 2006). High parent‐child cohesion may also facilitate more frequent, higher‐quality parent‐child remote communication, enabling parents to remain responsive to children's developmental needs across distance.

Interestingly, the conditional LGCM revealed that initial father‐child cohesion negatively predicted the developmental slope of prosocial behavior. Children with high father‐child cohesion may have already approached a developmental ceiling for their age group at baseline, yielding a high‐stable pattern. Conversely, children with low initial father‐child cohesion exhibited comparatively lower baseline prosocial behavior, which objectively afforded greater room for upward growth. As time spent in school accumulated and peer relationships deepened, these children may have gained compensatory momentum for prosocial development through alternative socialization channels, thereby exhibiting a faster rate of growth. Caution is warranted in interpreting these accounts, however, as the present study did not directly measure compensatory variables. Future longitudinal research should incorporate such variables to test these possible compensatory mechanisms.

Finally, regarding the explanatory power of the covariates on the social adjustment of left‐behind children, the current study found that the covariate cluster (parent‐child cohesion, gender, and migration type) accounted for 28% to 38% of the variance in the initial levels (intercepts) across all indicators, yet exhibited limited overall explanatory power for the developmental trajectories (slopes). First, individual and family characteristics established at T1, namely parent‐child cohesion, gender, and left‐behind status, showed moderate to strong associations with the baseline levels of left‐behind children's social adjustment. Among these, the quality of the long‐distance emotional bond captured by parent‐child cohesion, and the family structure differences captured by left‐behind status, may represent key family‐level factors shaping initial adjustment disparities. Second, the low proportion of explained variance in the slope warrants careful interpretation. On the one hand, the relatively static family and individual factors measured at T1 may indeed possess only limited predictive power for the rate of developmental change in children's adjustment, with their influence primarily operating through the setting of initial levels. On the other hand, the reliability of the slope variance itself depends on the number of follow‐up waves and measurement precision; if the slope variance estimate is compressed due to insufficient reliability, it becomes inherently difficult for any covariate to explain it meaningfully. Furthermore, time‐invariant covariates can capture only linear predictions of the slope from between‐person differences, and are unable to characterize the dynamic influence of time‐varying environmental factors, such as changes in peer relationships or fluctuations in school support, on the rate of change.

5. Limitations

First, regarding the longitudinal design and model fit, the measurement frequency and tracking duration of the current study remained relatively limited. Constrained by theoretical frameworks and practical feasibility, the current longitudinal design restricted our ability to thoroughly investigate the non‐linear developmental trajectories (e.g., quadratic or higher‐order fluctuating patterns) of left‐behind children's social adjustment. Notably, the latent growth curve model (LGCM) for loneliness failed to achieve an adequate model fit. This suggests that loneliness, functioning as a highly sensitive and volatile internalizing emotion, may exhibit more complex, non‐linear developmental characteristics. Future research is urgently needed to further elucidate the true dynamic evolution of complex emotions by increasing the number of measurement waves (e.g., implementing high‐frequency Ecological Momentary Assessment, EMA) and extending the tracking duration.

Furthermore, an important methodological limitation concerns the estimation of developmental trajectories for depressive symptoms and school engagement. During the LGCM estimation, these variables exhibited negative slope residual variances (resulting in non‐positive definite matrices), necessitating the boundary‐constraint approach where slope variances were fixed to zero. Rather than implying true group homogeneity, this statistical issue suggests that a simple linear growth trajectory may not adequately capture the true, potentially complex (e.g., non‐linear or fluctuating) developmental patterns of these indicators over the two‐year period. Alternatively, it may stem from observation errors or the limited temporal resolution inherent in a three‐wave longitudinal design. Future research employing more frequent assessment waves (such as intensive longitudinal designs) is required to accurately model the true inter‐individual variability in the developmental rates of these specific adjustment domains.

Finally, there remains considerable room for refinement regarding variable control and the exploration of potential heterogeneity. The current study primarily relied on static baseline covariates, failing to adequately capture the dynamic transitions of left‐behind status (e.g., parents returning home or re‐migrating) and the cumulative effects of separation duration throughout the tracking period. Moreover, the current study did not assess time‐varying factors such as parenting styles, guardianship quality, teacher support, or peer support across the measurement waves. This omission may partly explain why baseline family factors demonstrated minimal effects on the developmental slopes, as reported in our results. Future research should strongly consider employing person‐centered approaches, such as Growth Mixture Modeling (GMM) or Latent Class Growth Analysis (LCGA), to identify latent subpopulations with distinct developmental profiles, and incorporating time‐varying covariates into these models to provide a robust empirical basis for designing precise, stratified intervention strategies.

6. Practical Implications and Conclusions

The current study makes substantial theoretical contributions to the child development literature and offers critical practical implications for the safeguarding and optimal development of left‐behind children.

First, on a theoretical and methodological level, this study advances the field by employing a three‐wave longitudinal design to systematically map the multidimensional trajectories of social adjustment among left‐behind children. This approach transcends the limitations of prior unidimensional or cross‐sectional investigations, thereby bridging critical knowledge gaps regarding their long‐term developmental dynamics. Second, at the intervention level, the study elucidates the consistent protective effects of parent‐child cohesion across multiple domains of social adjustment. By substantiating its critical role within the family microsystem, these findings position parent‐child cohesion as a pivotal, actionable target for intervention. Consequently, intervention strategies should center on fortifying and sustaining high‐quality parent‐child emotional bonds. Given the elevated risk profile of children experiencing both‐parent migration, it is imperative that schools and communities provide targeted, remote family education guidance for migrant parents. This guidance should empower parents to compensate for the deficit in instrumental support by augmenting the frequency of affective interactions, effectively navigating the constraints of physical spatial separation. Third, at the policy and social service levels, left‐behind status and gender emerged as salient moderators of the developmental trajectories of social adjustment. This empirical evidence compellingly advocates for a paradigm shift from uniform, “one‐size‐fits‐all” approaches toward stratified, precision‐based interventions. Policymakers and school practitioners should leverage this evidence‐based differentiation to construct dynamic risk‐screening mechanisms. Such systems would facilitate the targeted allocation of services, such as prioritizing psychological resources for high‐risk subgroups (e.g., children experiencing both‐parent migration) to effectively foster their normative and healthy development.

In summary, this study reveals a domain‐differentiated developmental pattern among left‐behind children. Behavioral indicators, including problem behavior, prosocial behavior, and school engagement, exhibit normative improvements across the three waves, suggesting progressive contextual adjustment. Depressive symptoms, by contrast, remain stable at elevated levels with no evidence of age‐normative decline, indicating that emotional difficulties become progressively consolidated across this developmental period. This behavioral‐emotional divergence underscores that the psychological costs of parental migration are not self‐limiting and thus demand structured, developmentally informed intervention.

Author Contributions

Yadong Sun contributed to the article's conceptualization, methodology, software, data curation, validation, formal analysis, visualization, writing of the original draft, and editing & reviewing the final draft. Yile Li, Jiamei Li contributed to the article's methodology, investigation, and editing & reviewing the final draft. Zhi Wang contributed to the article's conceptualizing and editing & reviewing the final draft. Jingjin Shao contributed to the article's conceptualization, supervision, funding acquisition, project administration, resources, and editing & reviewing the final draft. The authors read and approved the final manuscript.

Ethics Statement

This study was reviewed and approved by the Committee of the Faculty of Psychology, Southwest University on August 20, 2016. The approval number was 201616.

Consent

This study’ informed consent was obtained from students and their guardians.

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request. The data are not publicly available due to privacy or ethical restrictions.