How do cumulative family risks influence the trajectory of problematic social media use among Chinese adolescents: A three-year longitudinal study

Abstract

Background and aims

Problematic social media use (PSMU) is a concerning public health issue among adolescents. Existing literature has paid attention to the role of singular family risk on PSMU, but how cumulative family risks affect the trajectory of PSMU needs to be further explored.

Methods

The current study employed a five-wave longitudinal design that lasted for three years (each time interval: 6, 6, 12, and 12 months) to reveal the longitudinal mechanism between cumulative family risks and the trajectory of PSMU, examining the mediating roles of escape and relationship motivations. This study investigated 1,973 adolescents (M_age = 14.51, SD_age = 1.49; age range: 11.95–17.45 years old; 47.4% females; 40% middle school) from southern China at wave 1, and the final sample size was 882 at wave 5.

Results

PSMU among Chinese adolescents presented with a stable tendency. In addition, cumulative family risks positively predicted the initial level (B = 0.21, SE = 0.02, p < 0.001) but not the slopes of PSMU. Moreover, escape motivation mediated the association between cumulative family risks and the trajectory of PSMU (B_intercept = 0.10, SE = 0.01, 95%CI = [0.076, 0.118]; B_{linear slope} = −0.03, SE = 0.01, 95%CI = [–0.040, −0.019]; B_{quadratic slope} = 0.004, SE = 0.001, 95%CI = [0.002, 0.006]).

Conclusions

Findings suggest that adolescents who experience cumulative family risks may be more likely to develop PSMU, potentially via the drive to escape from real life. A favorable family environment may be conducive to mitigating adolescent escape motivation and PSMU.

Introduction

In recent decades, social media (e.g. Facebook, Instagram, and WeChat) have gained popularity among youth, which serves a variety of functions such as recreation, communication, and information-seeking (Al-Menayes, 2015; Throuvala, Griffiths, Rennoldson, & Kuss, 2019). While adolescents benefit from using social media, some of them may develop problematic social media use (PSMU). PSMU refers to the addictive-like behavior characterized by inability to regulate the impulse of using social media, feeling stress when without social media, and continuously being occupied by social media use in mind (Boer et al., 2020, 2022), which is negatively related to adolescents' academic performance, and positively associated with sleeping disorders, depression, and psychological distress (Chen et al., 2020; Gingras et al., 2023; Homaid, 2022; Huang, 2022). A recent survey across 44 countries in Europe, Central Asia, and Canada indicated that the prevalence of adolescent PSMU was 11% in 2022 (Boniel-Nissim et al., 2024). In mainland China, 44.9% of adolescents reported higher risks of PSMU (Tang et al., 2018), implying that PSMU is a more noticeable phenomenon among Chinese adolescents. Given the significant negative impact of PSMU on adolescents' development and its increasing prevalence, it is imperative to investigate the developmental trajectory of PSMU and its potential risk factors, which can beneficially guide the well-directed intervention for alleviating adolescents' PSMU.

So far, several studies have explored the trajectory of PSMU in adolescents (e.g., Boer, Stevens, Finkenauer, & Van den Eijnden, 2022; Raudsepp, 2019; Raudsepp & Kais, 2019; Xiong, Xu, Chen, & Zhang, 2025). However, Boer et al. (2022) study focused primarily on younger adolescents in the Netherlands; Raudsepp and Kais (2019) exclusively explored the trajectory of PSMU among female adolescents in Estonia; Xiong et al. (2025) used a relatively small sample of Chinese adolescents ranging in a limited age range. As a result, this study attempted to depict a comprehensive picture of how PSMU evolves among a broader age range and a larger sample of adolescents, as well as over a longer period.

Moreover, it is also essential to explore the factors that influence the development of PSMU. Based on the ecological systems theory (Bronfenbrenner, 1979), the family, one of the essential mesosystems, has a significant influence on both adolescents' mental and behavioral developments. Prior studies have well documented that parental conflict (He, Liu, & Shen, 2021; Wang, Xu, & He, 2021), negative parenting style (Onyekachi, Egboluche, & Chukwuorji, 2022; Vossen, van den Eijnden, Visser, & Koning, 2024), and parental psychological control (Khodarahmi, Amanelahi, & Abaspour, 2023; Yao et al., 2022) were positively related to PSMU. However, these studies primarily examine the individual effects of specific family risk factors on PSMU. The cumulative impact of family risks on adolescents' PSMU over time remains largely unexplored, despite evidence suggesting a stronger deleterious effect from multiple risks than from any single risk (Li, 2017), and the reality that adolescents often face numerous family-related challenges simultaneously. Since the cumulative risk model is a widely used approach to combine risk factors into a comprehensive indicator in the research realm of addiction (Li, Zhou, Zhao, Wang, & Sun, 2016; Wang et al., 2022), this study adopted this method to compute the cumulative risks from family.

Finally, the underlying mechanism between cumulative family risk and the trajectory of PSMU remains underexplored. The motivational framework of problematic internet-related use suggests that individuals can fulfill psychological needs via the internet, emphasizing the importance of motivations in understanding problematic internet use (Schimmenti, 2023). Additionally, based on the self-determination theory (Deci & Ryan, 2008) and the compensatory internet use theory (Kardefelt-Winther, 2014), cumulative family risks may impede the satisfaction of adolescents' basic psychological needs, prompting them to turn to social media as a means of compensating for these unmet needs, which may in turn increase their vulnerability to PSMU. Among the internet use motivations, relationship and escape motivations are strongly linked to PSMU (Sever & Özdemir, 2022; Zhen, Liu, Hong, & Zhou, 2019). And as such, we explored how cumulative family risks influence PSMU over time in this study, with relationship and escape motivations as potential mediators.

Based on the foregoing, with a five-wave longitudinal design, the present study aimed to explore how cumulative family risks influence the trajectory of PSMU. Specifically, this study examined the mediating role of escape and relationship motivations in this association. Findings contribute to raising the awareness of the significant influence of family on adolescents' behavioral development, providing targeted interventions in adolescent PSMU rooted in the perspective of the family system.

Cumulative family risks and the trajectory of problematic social media use

Cumulative risk is defined as the simultaneous exposure to multiple risks, which collectively exert a more substantial negative impact than individual risks alone (Evans, Li, & Whipple, 2013; Rutter, 1979). This perspective emphasizes the quantity rather than the intensity of co-occurring risks (Evans, 2004), reflecting the reality that adversities often cluster together. As such, cumulative risk is an important perspective for understanding the complex origins of developmental challenges (Li, 2017).

The significance of the family environment as a core microsystem in shaping youth development is underscored by Bronfenbrenner's ecological systems theory (Bronfenbrenner, 1979). As a result, among the various sources of cumulative risk, family-related adversities are especially salient in shaping adolescents' social media use. Cumulative family risk refers to the co-occurrence of multiple risk factors within the family system. Indeed, a growing body of literature has documented associations between specific family-related risk factors and adolescents' PSMU (Khodarahmi et al., 2023; Vossen et al., 2024; Wang, Xu, & He, 2021; Yao et al., 2022). For instance, parental conflict is linked to higher PSMU among Chinese adolescents (Wang, Xu, & He, 2021), while low parental monitoring is associated with increased PSMU in Finland (Paakkari, Tynjälä, Lahti, Ojala, & Lyyra, 2021). Similarly, Khodarahmi et al. (2023) reported that parental psychological control predicted PSMU among female adolescents in Iran. However, limited empirical studies focused on the impact of cumulative family risks on adolescents' PSMU. Therefore, it is crucial to explore the risk factors of adolescents' PSMU from the perspective of cumulative family risk.

Research has shown that adolescents exposed to multiple family-related challenges are at increased risk for a range of problematic outcomes (Benito-Gomez, Fletcher, & Buehler, 2019; Li et al., 2016; Sameroff & Fiese, 2000). Self-determination theory suggests that such risks may frustrate adolescents' basic psychological needs for autonomy, competence, and relatedness, which are essential for well-being and self-regulation (Deci & Ryan, 2008). Consequently, adolescents may resort to social media to compensate, potentially increasing the risk of PSMU. As elaborated above, prior studies have demonstrated that specific family risks are associated with elevated PSMU (Khodarahmi et al., 2023; Vossen et al., 2024; Wang, Xu, & He, 2021; Yao et al., 2022), and cumulative risk has been linked to problematic internet use (Li et al., 2016; Wang et al., 2022).

In addition, several studies have explored the trajectories of PSMU among adolescents (Boer et al., 2022; Raudsepp, 2019; Raudsepp & Kais, 2019; Xiong et al., 2025). Boer et al. (2022) used a four-wave longitudinal investigation and identified the nonlinear trajectory of PSMU and social media use frequency among 1,419 Dutch adolescents, with levels initially increasing and then decreasing over time. Raudsepp and his colleague examined the developmental trend of PSMU among 397 adolescent girls in Estonia across two years, revealing a linear increase in PSMU over time (Raudsepp, 2019; Raudsepp & Kais, 2019). A latest three-wave survey that lasted for one year showed a linear increase in PSMU among 357 Chinese middle school students in general, and revealed three developmental trajectory classes of PSMU: high risk-gradual increase group, low risk-sharp increase group, and low risk-stable group (Xiong et al., 2025). However, findings regarding the developmental trajectory of PSMU remain inconsistent, and no clear consensus has been reached. Thus, we did not put forward any hypothesis about the developing trend of PSMU among Chinese adolescents. In terms of the association between cumulative family risks and the trajectory of adolescents' PSMU, based on the previous evidence (Li et al., 2016; Wang et al., 2022), we proposed the following hypothesis:

Hypothesis 1:

Cumulative family risks predict the initial level (H1a) and the slope of PSMU (H1b).

The mediating roles of escape motivation and relationship motivation

Drawn from the motivational framework (Schimmenti, 2023), the internet can be seen as a genuine environment where individuals can fulfill psychological needs and seek emotion regulation strategies. In particular, the use and gratification theory (UGT; Katz, Blumler, & Gurevitch, 1974) identifies five key media-related needs, namely cognition, emotion, relationship, integration, and escape. Especially, escape motivation and relationship motivation are two significant antecedents leading to PSMU (Li, Zhan, Zhou, & Gao, 2021; Wongpakaran, Wongpakaran, Pinyopornpanish, Simcharoen, & Kuntawong, 2021; Zhen et al., 2019). Escape motivation refers to the tendency to avoid real-life problems and negative emotions by immersing oneself in digital environments (Kim, 2017). Relationship motivation involves using social media to establish or strengthen social bonds, often as a way to alleviate emotional distress (Kim, 2017). Empirical research supports the link between these motivations and a range of problematic internet-related behaviors, including internet gaming disorder (Kircaburun et al., 2020) and problematic mobile phone use (Kim, 2017; Li et al., 2021; Shen & Wang, 2019; Young, Kuss, Griffiths, & Howard, 2017).

Drawing on compensatory internet use theory (Kardefelt-Winther, 2014), we proposed that escape and relationship motivations would mediate the relationship between cumulative family risks and the development of PSMU in adolescents. When adolescents encounter cumulative family risks, their psychological needs might be undermined. In turn, adolescents may turn to social media as a compensatory strategy to fulfill their unmet needs (Barzeva, Richards, Meeus, & Oldehinkel, 2020; Zhai et al., 2019), and some of them may excessively use social media, finally falling into addictive-like behaviors on social media. That is to say, PSMU could be a negative consequence of the maladaptive strategy for compensation. Accordingly, we expected two indirect pathways linking cumulative family risks to the development of PSMU. On the one hand, adolescents in adverse family environments may turn to social media to escape real-life challenges and alleviate negative emotions like anxiety. On the other hand, family risks may hinder the fulfillment of adolescents' need for relatedness, prompting them to seek social connections on social media for compensation. Driven by these two motivations, adolescents are more likely to use social media, and increase the risk of developing PSMU. A handful of studies have documented that internet use motivations mediated the connection between family risk factors and problematic internet-related behaviors (Sun et al., 2020; Zhen et al., 2019). For example, Zhen et al. (2019) found that positive parent-child relationships relieved the escape and relationship motivations and, in turn, mitigated the risk of problematic mobile phone use. Based on above, we proposed the following hypotheses:

Hypothesis 2:

Escape motivation would mediate the relationship between cumulative family risks and the intercept (H2a) and the slope (H2b) of PSMU.

Hypothesis 3:

Relationship motivation would mediate the link between cumulative family risks and the intercept (H3a) and the slope (H3b) of PSMU.

The present study

Taken together, to address the research gaps, the current research aimed to answer three research questions: First, how does the PSMU among Chinese adolescents evolve over time? Second, to what extent do cumulative family risks influence the development of PSMU? Third, whether and to what extent cumulative family risks influence the development of PSMU via escape and relationship motivations? Specifically, we employed a five-wave longitudinal investigation that lasted for three years and the cumulative risk model to reveal whether and how cumulative family risks influence the developmental trajectory of PSMU over time, examining the parallel mediating effect of escape motivation and relationship motivation (see Fig. 1).

Fig. 1.

Conceptual longitudinal mediation model

Note: T1 = Time 1, T2 = Time 2, T3 = Time 3, T4 = Time 4, T5 = Time 5.

Methods

Participants

This study employed a five-wave longitudinal design that lasted for three years. We distributed questionnaires to 2,327 students from middle and high schools located in southern China at Time 1 (T1: September, 2020) by cluster sampling. Due to the unavailability of seven classes during the data collection period and the exclusion of 39 participants for incomplete or invalid responses, the final valid sample consisted of 1,973 adolescents (M_age = 14.51, SD_age = 1.49; age range: 11.95–17.45 years old; 47.4% females; 40% middle school), with a response rate of 84.79%. Six months later, there remained 1,801 participants (47.9 % females) at Time 2 (T2), with an attrition rate of 8.72%. After another six months (Time 3, T3), 1,708 adolescents (48.6% females) continually participated in the data collection, with 5.16% dropping out. The fourth wave was conducted a year later (Time 4, T4), with a sample of 1,589 adolescents (48.6% females; attrition rate = 6.96%). Finally, the fifth wave of data was collected a year later, with the final sample of 882 adolescents (50.5% females; attrition rate = 44.49%). The dropout of several participants was due to their absence on the day of investigation (e.g., taking sick leave, dropping out of school, transferring to another school) or unwillingness to participate in the follow-up surveys at T2, T3, T4, and T5.

Procedures

The investigation involved three middle schools and two high schools in a southern Chinese metropolis. Schools were selected through the first author's established network of professional contacts with key individuals within the educational institutions. The investigation was administered in classroom settings using paper and pen, whereby trained assistants monitored and assisted participants where necessary. Participants were informed about the purpose of the study, and that they could withdraw their participation at any moment without consequences. Participants volunteered to take part in the survey with no incentive. The procedures were standardized and identical across all waves.

Measures

Cumulative family risks questionnaire

The cumulative risk model is a practical approach to combing the overlapping risk factors, which is a widely accepted method in the field of problematic or addictive behaviors (Li et al., 2016; Wang et al., 2022), with the advantages of reducing measurement error (Ghiselli, Campbell, & Zedeck, 1981) and enhancing validity (Brinberg & Kidder, 1982). Based on the prior research, interparental conflict, negative parenting style, and parental psychological control were considered the family risks for PSMU in this study (Khodarahmi et al., 2023; Paakkari et al., 2021; Wang, Xu, & He, 2021), and as such they were selected as the studied variables. The cumulative risk indices were created according to previous literature regarding cumulative risks (Li, 2017; Li et al., 2016; Wang et al., 2022). According to the cumulative risk model, each risk factor is dichotomized (0 = no risk; 1 = risk) to standardize the metric, and multiple risk factors are summed up to create the additive index of cumulative risks (Evans et al., 2013).

Children's Perception of Interparental Conflict Scale (CPIC). The Children's Perception of Interparental Conflict Scale (CPIC) was initially developed by Grych, Seid, and Fincham (1992) and the Chinese version was revised by Jin, Zhao, and Zou (2019). Using the Chinese version, adolescents reported the intensity of interparental conflict at T1. This scale consists of 5 items, rating on a 5-point scale from 1 (Completely false) to 5 (Completely true). Example items are “My parents have broken or thrown things during an argument” and “When my parents have an argument, they yell a lot”. The mean score was calculated, with the higher score implying the more severe interparental conflict. The score in the bottom 75th percentile was encoded as 1 and the rest was encoded as 0. The Cronbach's α of this scale was 0.87.

Alabama Parenting Questionnaire (APQ). Negative parenting at T1 was measured by the sub-scales (inconsistent discipline and poor supervision) of the Alabama Parenting Questionnaire - 9 items (APQ-9), which was adapted by Elgar, Waschbusch, Dadds, and Sigvaldason (2007). This scale was translated into Chinese and demonstrated good reliability among Chinese adolescents (Jiang, Jiang, Ren, & Wang, 2021). The inconsistent discipline scale consists of 3 items, such as “Your parents threaten to punish you and then do not do it”. The poor supervision scale includes 3 items, such as “Your dad/mom do not know the friends you go out with”. Items were rated on a 5-point scale (from 1 = Never to 5 = Always). The mean score was calculated, with the higher score showing more negative parenting. We encoded the score in the bottom 75th percentile as 1 and the rest was coded as 0. The Cronbach's α of inconsistent discipline scale was 0.74 and the Cronbach's α of poor supervision scale was 0.79.

Parental Psychological Control Scale (PPCS). The degree of parental psychological control at T1 was assessed with Parental Psychological Control Scale (PPCS; Shek, 2006). This scale was translated into Chinese and demonstrated good reliability among Chinese adolescents (Cao, Li, Ye, Xie, & Lin, 2020). This scale is unidimensional and consists of 10 items. Adolescents answered items on a 5-point scale from 1 (Never) to 5 (Always). Example items are “My parents always want to change my thoughts.” and “When I disappoint my parents, they stop talking to me”. The average score was computed, with the high score implying a high level of parental psychological control. The score in the bottom 75th percentile was coded as 1 while that in the upper was assigned as 0. The Cronbach's α in our study was 0.94.

Motivations for internet (smartphone) use scale

Participants' escape and relationship motivations for internet use at T2 were assessed by the motivations for internet (smartphone) use scale (Kim, 2017). This scale was translated into Chinese and demonstrated good reliability among Chinese adolescents (Zhen et al., 2019). It should be noted that we adopted the generalized word “Internet” instead of “smartphone” in the present study. This scale consists of 21 items, which are classified into four dimensions (escape motivation, pass-time motivation, relationship motivation, and information motivation). Since we focused on escape motivation and relationship motivation, the specific sub-scales were drawn. There are 6 items in the escape motivation scale (e.g. “To forget about worries and concerns”) and 3 items in the relationship motivation scale (e.g. “To communicate with others”). Participants rated their feelings of internet use on a 7-point scale (1 = Strongly disagree, 7 = Strongly agree). The Cronbach's α of escape motivation scale and relationship motivation scale were 0.89 and 0.87, respectively.

Problematic social media use scale

The adapted version of the C-VAT instrument (Van Rooij, Ferguson, Van de Mheen, & Schoenmakers, 2017) was used to evaluate PSMU from T2 to T5 among Chinese adolescents. This scale was adapted from the CIUS scale developed and validated by Meerkerk, Van den Eijnden, Vermulst, and Garretsen (2009) and Meerkerk (2007). This scale was translated into Chinese and demonstrated good reliability among Chinese adolescents (Dou, Wang, Li, Yuan, & Wang, 2024). There are seven items rated on a 5-point scale (0 = Never, 4 = Always). Example items are “How frequently do you use social media because you feel unhappy?” and “How frequently do you lack sleep because you spent the night using social media?”. The mean score of all items was considered as the indicator of PSMU, with a high score indicating a high level of PSMU. The Cronbach's α of this scale at T2, T3, T4, and T5 were 0.88, 0.89, 0.89, and 0.92 respectively.

Covariates

Previous studies have documented that age, sex, and socioeconomic status (as indicated by parents' educational levels) are associated with adolescents' risk for PSMU (Andreassen et al., 2016, 2017; Kircaburun et al., 2019; Sun et al., 2021). Specifically, younger adolescents, girls, and those from lower socioeconomic backgrounds are more susceptible to PSMU. Hence, adolescents' age, sex (1 = male, 2 = female), and parents' educational level (1 = primary school, 2 = middle school, 3 = undergraduate, 4 = postgraduate) were measured and included as covariates to control for potential confounding effects. Age was calculated based on participants' reported birthdate (month and year).

In addition, independent-samples t-tests revealed significant group differences in T2 escape motivation, T2 relationship motivation, and PSMU (T2–T5) based on residence (1 = city, 2 = countryside) and only-child status (1 = yes, 2 = no). As such, these two variables were also included as covariates in the analyses.

Statistical analyses

The calculation of the independent variable employed the cumulative risk model. Attrition analysis, descriptive statistics, bivariate correlations analysis, and common method bias were conducted with Statistical Package for Social Science (SPSS) 27.0. Measurement invariance and structural equation modeling (SEM) such as latent growth curve model and mediation analysis were performed using Mplus 8.0.

In preliminary analyses, we first conducted the descriptive statistics to describe the means and standard deviations of studied variables (i.e., T1 cumulative family risks, T2 escape motivation, T2 relationship motivation, PSMU at T2, T3, T4, and T5) and covariates (T1 adolescents' age, sex, the only child status, residence, parents' educational levels). Additionally, the bivariate correlations between studied variables and covariates were detected with Person's correlation analysis. Second, to ensure the measurement was longitudinally consistent across waves, the measurement invariance (i.e., configural, metric, and scalar invariances) was examined. According to Hu and Bentler (1999), the model fit indices included the comparative fit index (CFI; acceptable >90), root mean square error of approximation (RMSEA; acceptable <0.08), and standard root mean square residual (SRMR, acceptable <0.08). Given that the chi-square is sensitive to large sample size, we did not use the chi-square test to evaluate the measurement invariance (Liu, Yue, & Yang, 2024; Wu & Becker, 2023). Based on the previous study (Chen, 2007), changes in CFI, RMSEA, and SRMR were considered as indicators of measurement invariance. The changes in CFI ≤ |0.010|, RSMEA ≤ | 0.015|, and SRMR ≤ | 0.030| show that the metric invariance is valid. In addition, the changes in CFI ≤ |0.010|, RSMEA ≤ | 0.015|, and SRMR ≤ |0.010| show that the scalar invariance is valid. Next, due to the dropout in the longitudinal study, attrition analyses of the key variables and covariates at T1 between participants who took part in all investigations across five waves (i.e., the complete group) and those who dropped out (i.e., the dropout group) were performed. Specifically, continuous variables (i.e., T1 cumulative family risks, T1 PSMU, T1 adolescents' age) were utilized independent samples T-test to test if there were differences between the two groups, while categorical variables (i.e., adolescents' sex, the only child status, residence, parents' educational levels) were used the chi-square test to detect the differences. To examine whether the data was biased by self-reports, Harman's single-factor test (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003) was employed to test the common method bias.

Subsequently, the latent growth curve model was performed to reveal the developmental trajectory of PSMU among Chinese adolescents. This model is a widely accepted technique to explore the change over time, which provides information on the initial level (intercept) and slope regarding the intra-individual and inter-individual overall population and their differences (Duncan & Duncan, 2004; Rioux, Stickley, & Little, 2021). The full information maximum likelihood estimation (Acock, 2005) was utilized to handle the missing data in the latent growth curve models. First, the unconditional latent growth curve models were employed to reveal the developmental trajectory of PSMU among adolescents. Both linear and nonlinear (i.e. quadratic) latent growth models of PSMU were tested. According to the interval of measurement across four waves (the interval between T2 and T3, T3 and T4, and T4 and T5 were 6 months, 12 months, and 12 months, respectively), the loadings of the latent intercept were fixed as 1, 1, 1, 1 and that of the slope were fixed as 0, 1, 3, 5 in the linear growth model. In the quadratic growth model, the loadings of the intercept and the linear slope were as the same in the linear model, but the loadings of the quadratic slope were fixed as 0, 1, 9, and 25. The mean of the intercept represents the average initial level while the mean of the slope represents the line of trajectory over time (Burant, 2016). The significant mean indicates there are changes in the studied variables. In addition, the significant variances of intercept and slope reflect that there are individual differences in initial status and change. Second, conditional latent growth models were utilized to examine the direct links between cumulative family risks and initial levels and slopes of PSMU, testing the H1. Third, to explore the longitudinal mechanism between T1 cumulative family risks and the developmental trajectory of PSMU from T2 to T5, structural equation modeling was performed to examine the parallel mediating effect of T2 escape motivation and T2 relationship motivation, which test the H2 and H3. Escape and relationship motivations were specifically selected for the second wave in order to follow the chronological order and better reveal the mechanism across time. The standardized regression coefficient is used as an index of effect size (Nieminen, 2022). According to Cohen's criteria (1969), the effect size is at the small magnitude when |β| = 0.10, the medium effect size denotes when |β| = 0.30, and the effect size is considered large when |β| = 0.50. In addition, the effect proportion mediated (P_M) is used to quantify the strength of the mediation effect, which computes the ratio of the indirect effect over the total effect of the independent variable to the dependent variable (Shrout & Bolger, 2002; P_M = $\frac{ab}{c}$). The mediating effect is considered meaningful only if the estimate of P_M surpasses 0.2. Of note, P_M is interpretative only when the indirect effect and the direct effect are the same sign; otherwise, the values of P_M may exceed 1 or be with negative sign, implying the suppression effect (Shrout & Bolger, 2002).

Finally, to test the robustness of the results, we used listwise deletion to conduct all the latent growth curve models in this study, examining whether the results with the sample of participants who did not have missing data were consistent with the primary results.

Ethics

The procedures involved in this study obtained the approval of the research ethics committee in Guangzhou University before the data collection. The study was carried out in accordance with the Declaration of Helsinki. Participants' parents signed the informed consent and adolescent participants also assented to fill out the questionnaires. Participants were informed that they reserved the right to withdraw from the study at any time and that their information and answers would be confidential. Due to the longitudinal design, names and student ID numbers were collected to match data across waves. Participants were fully informed that their personal information, along with their responses, would be kept confidential and used only for research purposes.

Results

Descriptive statistics and correlations

The characteristics of participants' demographic variables are displayed in Table 1. The means, standard deviations, and bivariate correlations among studied variables and covariates are illustrated in Table 2. T1 cumulative family risks were positively associated with T2 escape motivation, T2 relationship motivation, and PSMU from T2 to T5. T2 escape motivation was significantly and positively related to T2 relationship motivation and PSMU from T2 to T5. T2 relationship motivation was positively related to PSMU across four waves.

Table 1.

Participants' demographic characteristics

Demographic variables at Time 1	N = 1,973	Percentage (%)
Adolescent's sex
Male	1,038	52.61
Female	935	47.39
The only child status
Yes	679	34.4
No	1,289	65.4
Missing data	5	0.3
Residence
City	1,454	73.7
Countryside	513	26.0
Missing data	6	0.3
Father's educational level
Primary school degree	67	3.4
Middle school degree	1,029	52.2
Bachelor's degree or the equal education	652	33
Master's degree or above	216	10.9
Missing data	9	0.5
Mother's educational level
Primary school degree	134	6.8
Middle school degree	1,049	53.2
Bachelor's degree or the equal education	576	29.2
Master's degree or above	204	10.3
Missing data	10	0.5

Note: N = Sample size. There are missing values due to omissions or the participants' unwillingness to report.

Table 2.

Descriptive statistics and correlations of the study variables

	1	2	3	4	5	6	7	8	9	10	11	12	13
Key variables
1.T1 Cumulative family risks	–
2.T2 Escape motivation	0.24***	–
3.T2 Relationship motivation	0.05*	0.41***	–
4.T2 Problematic social media use	0.22***	0.54***	0.29***	–
5.T3 Problematic social media use	0.23***	0.41***	0.22***	0.66***	–
6.T4 Problematic social media use	0.23***	0.33***	0.20***	0.51***	0.57***	–
7.T5 Problematic social media use	0.16***	0.28***	0.17***	0.39***	0.42***	0.43***	–
Covariates
8.T1 Adolescent's age	0.04	0.28***	0.21***	0.31***	0.31***	0.18***	0.13***	–
9.Adolescent's sex	−0.03	0.09***	0.01	0.14***	0.23***	0.14***	0.09**	0.05*	–
10.The only child status	−0.01	0.06*	0.01	0.10***	0.09***	0.06**	0.08**	0.10***	−0.19***	–
11.Residence	0.03	0.16***	0.11***	0.15***	0.15***	0.09***	0.08*	0.41***	−0.06**	0.12**	–
12.Father's educational level	−0.02	−0.20***	−0.09***	−0.18***	−0.17***	−0.13***	−0.07*	−0.41***	−0.05*	−0.19***	−0.37***	–
13.Mother's educational level	−0.04	−0.21***	−0.07**	−0.18***	−0.20***	−0.14***	−0.06	−0.42***	−0.06**	−0.20***	−0.34***	0.69***	–
M	0.81	3.81	3.51	1.29	1.27	1.28	1.31	14.51	47.39a	1.65	1.26	2.52	2.43
SD	0.86	1.42	1.53	0.88	0.87	0.87	0.89	1.49	0.50	0.48	0.44	0.73	0.77

Note: The initial sample size was 1,973 and the final sample size was 882. ^*p < 0.05, ^**p < 0.01, ^***p < 0.001. Abbreviations: M = Mean, SD = Standard deviation. Adolescent's gender: 1 = male, 2 = female. ^a the percentage of female adolescents. Parents' educational levels: 1 = primary school, 2 = middle school, 3 = undergraduate, 4 = graduate student. Residences: 1 = city, 2 = countryside. The only child status: 1 = yes, 2 = no. T1 = Time 1, T2 = Time 2, T3 = Time 3, T4 = Time 4, T5 = Time 5.

Measurement invariance

We constructed a series of models to evaluate the measurement invariance from T2 to T5. First, an unconstrained model (Model 1) was used to test the configural invariance. The model indices demonstrated a good fit (see Table 3), showing that the configural invariance was valid. Second, we constructed a model that constrained the factor loadings to be identical at four waves (Model 2) and compared it with M1 to test the metric invariance. As the absolute values of changes in CFI, RSMEA, and SRMR were less than 0.010, 0.015, and 0.030 respectively, results indicated that the metric invariance was valid. Finally, the model sequentially constrained the intercepts across four waves (Model 3) was assessed and it was compared to M2. Results showed that the scalar invariance was denied. We then tested the partial scalar measurement model that freely estimated the intercepts of two items (Model 4) and compared it with M2. As shown in Table 3, results suggest that the measurement invariance reaches the partial scalar invariance, representing a relatively strong level.

Table 3.

Model fit indices for measurement models and tests of measurement invariance

Model	RMSEA	CFI	TLI	SRMR	△CFI	△RMSEA	△SRMR
Model 1: Configural invariance	0.042	0.951	0.939	0.039	–	–	–
Model 2: Metric invariance	0.043	0.946	0.936	0.043	−0.005	+0.001	+0.004
Model 3: Scalar invariance	0.046	0.933	0.925	0.046	−0.013	+0.003	+0.003
Model 4: Partial scalar invariance	0.044	0.939	0.931	0.046	−0.007	+0.001	+0.003

Note: RMSEA = Root mean square error of approximation, CFI = Comparative fit index, TLI = Tucker-Lewis index, SRMR = Standardized root mean square residual.

Attrition analysis and common method bias

To examine possible bias between participants who took part in surveys across all time points (Group 1: the complete group) and those who dropped out at T2 and/ or T3 and/ or T4 and/ or T5 (Group 2: the attrition group), attrition analyses were conducted. Independent-samples T-test indicated that cumulative family risks (t (1,971) = −2.00, p = 0.046, d = −0.09), T1 PSMU (t (1,971) = 2.47, p = 0.014, d = 0.11), and age (t (1964.40) = 11.38, p < 0.001, d = 0.51) showed a significant difference. Compared to the attrition group, the participants in the complete group were older and had lower levels of cumulative family risks and T1 PSMU. The chi-square test revealed that the two groups significantly differ in sex (χ²(1) = 6.01, p = 0.014, Cramer's φ = 0.06), the only child in the family (χ²(2) = 11.34, p = 0.001, Cramer's V = 0.08), residence (χ²(2) = 13.70, p < 0.001, Cramer's V = 0.08), father's educational level (χ²(4) = 70.30, p < 0.001, Cramer's V = 0.19), and mother's educational level (χ²(4) = 88.33, p < 0.001, Cramer's V = 0.21). Specifically, compared to the attrition group, the complete group held a lower proportion of females and a higher proportion of males. A lower percentage of participants in the complete group were the only child in their family and lived in the city, compared to the attrition group. Furthermore, participants' fathers in the complete group had a higher proportion of primary school degrees and middle school degrees while a lower proportion of undergraduate, postgraduate, and missing data than the attrition group. In contrast to the attrition group, participants' mothers in the complete group presented a higher percentage of primary and middle school degrees whereas the lower percentage of undergraduate, postgraduate, and missing data. Little's Missing Completely at Random (MCAR) test indicated that the data were not missing completely at random (χ²(97) = 272.21, p < 0.001). Although the two groups differed in several variables, most of the effect sizes ranged from small to medium effect. Hence, we supposed that the abovementioned discrepancies might not impact the current study.

Given that the data were basically collected from participants' self-reports, which may generate the common method bias. Harman's single-factor test (Podsakoff et al., 2003) showed that 16 characteristic roots were greater than 1 and the first factor accounted for 20.38% of the variance, below the threshold of 40%.

The trajectory of PSMU among adolescents

Tables 4 and 5 present the fit indices, intercept, and slope statistics of linear and nonlinear growth models. According to the results, all growth models demonstrated a good fit. Results indicated that the PSMU among adolescents showed a stable trend over 2.5 years. Given that the quadratic growth model showed the best model indices and the smaller BIC value denotes the more parsimonious model (Wu & Becker, 2023), the trajectory was treated as the quadratic growth model in the following analyses. In addition, the variances of intercept and linear slope were significant while the variance of quadratic slope was marginally significant, indicating that there was an inter-individual difference in the initial level and the linear slope.

Table 4.

Model fit indices of the latent growth curve models

Model	X2(df)	p	CFI	RMSEA (90% CI)	BIC	SRMR
Model 1	18.25 (5)	0.003	0.993	0.037 ([0.020, 0.056])	14845.26	0.025
Model 2	0.48 (1)	0.489	1.00	0.000 ([0.000, 0.052])	14857.81	0.003
Model 3	23.26 (8)	0.003	0.994	0.031 ([0.017, 0.046])	14465.61	0.014
Model 4	53.63 (16)	<0.001	0.989	0.035 ([0.025, 0.045])	26501.15	0.013

Note: Unstandardized regression coefficients are reported. Model 1 = the unconditional linear latent growth model, Model 2 = the unconditional quadratic latent growth model, Model 3 = the conditional latent growth model with cumulative family risks at Time 1 as the predictor, sex, age, residence, the only child status, and parents' educational levels as covariates, Model 4 = the conditional latent growth model with cumulative family risks at Time 1 as the independent variable, escape and relationship motivations at Time 2 as mediators, sex, age, residence, the only child status, and parents' educational levels as covariates. Abbreviations: χ² = chi-square statistic, df = degrees of freedom, CFI = Comparative fit index, RMSEA = Root mean square error of approximation, 90% CI = 90% Confidence interval, BIC = Bayesian information criteria, SRMR = Standardized root mean square residual.

Table 5.

Intercept and slope statistics of the latent growth curve models

Model	Means						Variances
	I	p (95% CI)	S1	p (95% CI)	S2	p (95% CI)	I	p (95% CI)	S1	p (95% CI)	S2	p (95% CI)
1	1.28	<0.001 ([1.243, 1.319])	0.001	0.888 ([–0.009, 0.010])	–	–	0.54	<0.001 ([0.499, 0.593])	0.01	<0.001 ([0.006, 0.016])	–	–
2	1.28	<0.001 ([1.244, 1.323])	−0.01	0.670 ([–0.036, 0.022])	0.002	0.606 ([–0.005, 0.008])	0.59	<0.001 ([0.517, 0.653])	0.09	0.001 ([0.039, 0.147])	0.002	0.055 ([0.000, 0.005])
3	−1.60	<0.001 ([–2.127, −1.091])	0.49	0.022 ([0.060, 0.903])	−0.04	0.386 ([–0.123, 0.052])	0.43	<0.001 ([0.369, 0.496])	0.07	0.015 ([0.017, 0.125])	0.002	0.107 ([0.000, 0.004])
4	−1.75	<0.001 ([–2.198, −1.313])	0.46	<0.001 ([0.294, 0.623])	−0.03	0.005 ([–0.047, −0.008])	0.33	<0.001 ([0.274, 0.388])	0.08	0.002 ([0.035, 0.136])	0.002	0.065 ([0.000, 0.005])

Note: Unstandardized regression coefficients are reported. Model 1 = the unconditional linear latent growth model, Model 2 = the unconditional quadratic latent growth model, Model 3 = the conditional latent growth model with cumulative family risks at Time 1 as the predictor, sex, age, residence, the only child status, and parents' educational levels as covariates, Model 4 = the conditional latent growth model with cumulative family risks at Time 1 as the independent variable, escape and relationship motivations at Time 2 as mediators, sex, age, residence, the only child status, and parents' educational levels as covariates. Abbreviations: I = Intercept, S₁ = Linear slope, S₂ = Quadratic slope.

The effect of cumulative family risks on the trajectory of PSMU

Cumulative family risks and covariates were treated as time-invariant variables and incorporated in the unconditional growth model, to examine their effects on the changing process of PSMU. As displayed in Table 4, the model fit indices demonstrated good. As shown in Fig. 2, after controlling for the demographic variables, T1 cumulative family risks positively predicted the intercept of PSMU (B = 0.21, β = 0.24, SE = 0.02, p < 0.001, 95%CI = [0.166, 0.253]) but not the linear slope (B = 0.02, β = 0.07, SE = 0.02, p = 0.180, 95%CI = [–0.011, 0.057]) and the quadratic slope of PSMU (B = −0.006, β = –0.12, SE = 0.004, p = 0.085, 95%CI = [–0.014, 0.001]). In addition, the intercept was significantly and negatively associated with the linear slope (B = −0.57, β = –0.33, SE = 0.03, p = 0.030, 95%CI = [–0.111, −0.008]), which indicated that those who reported a higher initial level of PSMU may present with a slower linear slope. However, the intercept was not significantly related to the quadratic slope (B = 0.004, β = 0.13, SE = 0.01, p = 0.421, 95%CI = [–0.005, 0.013]).

Fig. 2.

The conditional latent growth model

Note: Unstandardized regression coefficients are reported before the slash and standardized results are presented after the slash. ^*p < 0.05, ^**p < 0.01, ^***p < 0.001. T1 = Time 1, T2 = Time 2, T3 = Time 3, T4 = Time 4, T5 = Time 5. In addition to cumulative family risks, sex, age, residence, the only child, father's educational level, and mother's educational level were considered as covariates in this model. However, to simplify the figure, we only presented the key studied variables.

The longitudinal mediation of escape motivation and relationship motivation

To unravel the underlying meditating effect of escape and relationship motivations, we constructed a parallel mediation model. As shown in Table 4, the model provided good fit indices. Results are illustrated in Fig. 3. After controlling for demographic variables, T1 cumulative family risks positively predicted the intercept (B = 0.11, β = 0.13, SE = 0.02, p < 0.001, 95%CI = [0.072, 0.152]), and the linear slope of PSMU (B = 0.05, β = 0.15, SE = 0.02, p = 0.003, 95%CI = [0.018, 0.087]), but negatively predicted the quadratic slope of PSMU (B = −0.01, β = –0.18, SE = 0.04, p = 0.005, 95%CI = [–0.018, −0.003]). T1 cumulative family risks positively predicted T2 escape motivation (B = 0.38, β = 0.23, SE = 0.04, p < 0.001, 95%CI = [0.313, 0.452]). Nevertheless, the relationship between T1 cumulative family risks and T2 relationship motivation was not significant (B = 0.07, β = 0.04, SE = 0.04, p = 0.068, 95%CI = [–0.004, 0.153]). T2 escape motivation positively predicted the intercept (B = 0.25, β = 0.47, SE = 0.02, p < 0.001, 95%CI = [0.221, 0.281]) as well as the quadratic slope of PSMU (B = 0.01, β = 0.30, SE = 0.003, p < 0.001, 95%CI = [0.005, 0.015]), and negatively predicted the linear slope of PSMU (B = −0.07, β = –0.35, SE = 0.012, p < 0.001, 95%CI = [–0.097, −0.051]). T2 relationship motivation positively predicted the intercept (B = 0.04, β = 0.08, SE = 0.01, p = 0.001, 95%CI = [0.015, 0.063]) but had no significant association with the linear slope (B = 0.003, β = 0.01, SE = 0.01, p = 0.818, 95%CI = [–0.019, 0.024]) as well as the quadratic slope of PSMU (B = −0.001, β = –0.02, SE = 0.002, p = 0.815, 95%CI = [–0.005, 0.004]). As summarized in Table 6, the mediating effect of T2 escape motivation between T1 cumulative family risks and the intercept of PSMU was significant (estimate = 0.10, standardize estimate = 0.11, SE = 0.01, p < 0.001, 95%CI = [0.076, 0.118], P_{M =} 0.45). In addition, T2 escape motivation also acted as a significant mediator of the link between T1 cumulative family risks and slopes of PSMU (linear slope: estimate = −0.03, standardize estimate = −0.08, SE = 0.01, p < 0.001, 95%CI = [–0.040, −0.019], P_{M =} −1.17; quadratic slope: estimate = 0.004, standardize estimate = 0.07, SE = 0.001, p < 0.001, 95%CI = [0.002, 0.006], P_{M =} −0.57). However, T2 relationship motivation could not mediate the link between T1 cumulative family risks and the trajectory of PSMU.

Fig. 3.

The longitudinal mediation model

Note: Unstandardized regression coefficients are reported before the slash and standardized results are presented after the slash. ^*p < 0.05, ^**p < 0.01, ^***p < 0.001. T1 = Time 1, T2 = Time 2, T3 = Time 3, T4 = Time 4, T5 = Time 5. Sex, age, residence, the only child, father's educational level, and mother's educational level were considered as covariates in this model. However, to simplify the figure, we only presented the key studied variables.

Table 6.

The indirect pathways of the longitudinal mediation model

Pathway	Unstandardized estimates	Standardized estimates	SE	p	95%CI
T1 CFR→T2 EM→PSMU intercept	0.10	0.11	0.01	< 0.001	[0.076, 0.118]
T1 CFR→T2 EM→PSMU linear slope	−0.03	−0.08	0.01	< 0.001	[–0.040, –0.019]
T1 CFR→T2 EM→PSMU quadratic slope	0.004	0.07	0.001	< 0.001	[0.002, 0.006]
T1 CFR→T2 RM→PSMU intercept	0.003	0.003	0.002	0.126	[0.000, 0.008]
T1 CFR→T2 RM→PSMU linear slope	0.000	0.001	0.001	0.841	[–0.001, 0.003]
T1 CFR→T2 RM→PSMU quadratic slope	0.000	−0.001	0.000	0.839	[–0.001, 0.000]

Note: SE = Standard error. T1 = Time 1, T2 = Time 2. CFR = Cumulative family risks, EM = Escape motivation, RM = Relationship motivation, PSMU = Problematic social media use. The significant results are in bold.

Sensitivity analyses

To test the robustness of the results, the listwise deletion was employed to run the models with the sample who did not have missing data, including the unconditional latent growth curve model, the conditional latent growth curve model with T1 cumulative family risks as the predictor, and the conditional latent growth curve model with T1 cumulative family risks as the independent variable and T2 escape and relationship motivations as mediators. As summarized in Tables S1–S3 (Online supplementary materials), all the models fitted the data well and most of the findings were consistent with the results using the overall sample, except that T2 escape motivation could not significantly mediate the relationship between T1 cumulative family risks and the quadratic slope of the trajectory of PSMU. To sum up, these findings indicated that the main results were robust in the current research.

Discussion

As a noticeable issue among adolescents, PSMU has gained attention from numerous researchers. In the present study, grounded in the ecological systems theory, the use and gratification theory, and compensatory internet use theory, a five-wave longitudinal design was adopted and the latent growth curve model indicated that PSMU among Chinese adolescents presented with a stable tendency. Results also found that cumulative family risks could be a significant predictor of the initial level of PSMU but not the slope. Furthermore, escape motivation rather than relationship motivation mediated the association between cumulative family risks and the developmental trajectory of PSMU. Findings beneficially extend the existing literature that explores family predictors of PSMU from the singular risk to cumulative risks and reveal the longitudinal mechanism between family risks and the trajectory of PSMU from the motivational aspect.

The developmental trajectory of PSMU among Chinese adolescents

Previous studies have recognized the trajectory of PSMU among adolescents (Boer et al., 2022; Raudsepp, 2019; Raudsepp & Kais, 2019; Xiong et al., 2025). Different with the findings that stated adolescents' PSMU changed as time went on, the current study found that the average level of PSMU was relatively low and the trajectory of PSMU among adolescents remained steady over almost three years. Hence, results may imply that the developmental trajectory of PSMU in Chinese adolescents differs from that of adolescents in the Netherlands and Estonia. The above findings may be ascribed to the restriction of mobile phones among Chinese adolescents. Adolescents in China are not allowed to take their phones to school and as such they only have limited time to use their phones after school. In addition, parents in China usually exert restrictive practices as a way of support (Hawk, Wang, Wong, Xiao, & Zhang, 2023; Pomerantz & Wang, 2009), which means they tend to conduct control functions in children's time of being on the phone. Nevertheless, it should be noted that there are other differences between prior research (Boer et al., 2022; Raudsepp & Kais, 2019) and this study in addition to cultural context. Specifically, prior research mainly explored the trajectory of PSMU in the sample of early adolescents whereas this study paid attention to both early and middle adolescents. Besides, Boer et al. (2022) traced the development of PSMU for four years, whereas this study only observed the development of PSMU for about three years. Therefore, Boer et al.’s study might reflect the change of adolescent PSMU from a longer-term perspective than this study. Newly published research indicated a linear increase in PSMU among Chinese adolescents and identified three different trajectories (Xiong et al., 2025). Although the present study shares the same culture as Xiong et al.’s study, we did not observe a similar trend to theirs. It may be attributed to different magnitudes and age ranges of participants and distinct tracking durations. Overall, we utilized a longer time interval lasting for three years and investigated a larger sample size of participants covering a larger age range (i.e., from 11.95 to 17.45 years old), whereas Xiong et al. employed a one-year trace with 357 students in the middle adolescence stage (i.e., from 12 to 15 years old). Future research can consider examining the development of PSMU among adolescents in various stages and revealing the heterogeneity between different stages.

Additionally, the present study found that adolescents' initial levels and slopes of PSMU had an inter-individual difference, suggesting that although individuals show a general pattern of PSMU, they would have their own developmental characteristics. We also found the initial level of PSMU was negatively associated with the slope, which implies that the initial status and the slope might influence each other. For instance, adolescents with a higher baseline of PSMU may develop PSMU more slowly.

The effect of cumulative family risks on the development of PSMU

Cumulative family risks could predict the initial level of PSMU with an effect close to medium-size but showed no significant effect on the slope, supporting H1a but not H1b. This finding is also similar to previous studies related to problematic internet use (Li et al., 2016; Wang et al., 2022), showing that cumulative family risks also affect adolescents' PSMU. Corresponding to ecological systems theory, results prove that the family system is possibly fundamental for adolescents' development, especially problematic internet-related behaviors in this study. Drawing from compensatory internet use theory, adolescents who suffer from family risks are prone to use social media platforms to relieve their negative emotions and satisfy unmet psychological needs. Moreover, problematic internet-related use could be an approach for children to catch their parents' attention (Liu, Fang, Zhou, Zhang, & Deng, 2013). For adolescents who live in an unfavorable family environment, their psychological needs may be ignored by their parents, which may lead them to misbehave in order to seek care and attention from their parents.

Albeit with the pathway from cumulative family risks to adolescent PSMU revealed in the current research, in fact, PSMU could reversely accelerate family risks such as family dysfunction and parental phubbing (Yin et al., 2024; Zhou, Zhao, Wang, & Zhu, 2023). For instance, Zhou employed the cross-lagged panel model and evidenced that family dysfunction predicted increased problematic gaming while problematic gaming could foster more family dysfunction and vice versa. Therefore, the relationship between family risks and PSMU may not be unidirectional, but a potentially vicious circle.

The escape motivation as a core mediator

Findings demonstrated that escape motivation was positively related to the initial level with a large effect and slopes of adolescents' PSMU with at least medium effect size, implying a meaningful impact of escape motivation on adolescents' trajectories of PSMU. However, it should be noted that there were suppressions of escape motivation on the relationship between cumulative family risks and slopes of PSMU (i.e. P_M for linear and quadratic slopes was higher than 1 and the sign was negative), which implies that escape motivation as a mediator may suppress the true links between cumulative family risks and slopes of the trajectory of PSMU. This explains why the inconsistency exists in the relationships between cumulative family risks and slopes of PSMU in the model that considers only the direct pathway and in the model that integrates mediators. In addition, similar to previous studies (Shen & Wang, 2019; Zhen et al., 2019), escape motivation played the mediating role between cumulative family risks and the developmental trajectory of PSMU, supporting the H2. The current results correspond to the motivational framework, which points out that the development of internet-related addiction lies in individuals' specific needs (Schimmenti, 2023). Aligning with compensatory internet use theory (Kardefelt-Winther, 2014), seeking social media platforms may mitigate their negative emotions, where, for instance, adolescents can be occupied by entertainment as well as a constant row of updated information, and in this case, their escape motivations are gratified. Since adolescents may use social media as a compensatory strategy to cope with unmet psychological needs, the temporary satisfaction they experience can act as positive reinforcement, increasing the likelihood of excessive and habitual use, ultimately leading to PSMU.

Although relationship motivation was significantly associated with the initial level of PSMU with a very small effect, it did not mediate the association between T1 cumulative family risks and the trajectory of PSMU, which contradicts H3. The significant pathway from relationship motivation to PSMU aligned with prior research (Zhen et al., 2019), but was inconsistent with other studies that noted that motivation to socialize was not significantly correlated to problematic internet-related use (Chang, Hsieh, & Lin, 2018; Kircaburun et al., 2020). Moreover, Kim found that relationship motivation could not significantly predict face-to-face interaction (Kim, 2017). The present study extended Kim's study and proved that although people with relationship motivation might not turn to interaction in real life, they were prone to seek online interaction on social media platforms. Nevertheless, it should be cautious that the effect size of this result was merely at a small magnitude, reflecting that the significant link might be unreliable to some extent.

In terms of the insignificant pathway from cumulative family risks to relationship motivation, adolescents who are exposed to cumulative family risks may perceive high uncertainty of building bonds with others (Feldman & Downey, 1994), in which case, it may be hard for them to start a new relationship either in real life or in the virtual world due to the fear of feeling rejection. For instance, considerable empirical evidence indicated that family risks such as interparental conflict and negative parenting could facilitate adolescent social anxiety and withdrawal (Choi, Choi, & Kim, 2019; Ran et al., 2021). Furthermore, the deleterious impact of family risks on socially maladaptive behaviors among adolescents might function through increasing attachment insecurity with parents (Wang, Wu, & Wang, 2021). This may explain why cumulative family risks could not predict relationship motivation for internet use.

To sum up, the present findings demonstrated that adolescents who experience cumulative family risks tend to foster PSMU due to the motivation of escaping from real life rather than seeking for relationship.

Limitations and implications

The findings should be interpreted in light of several limitations. First, even if the cumulative risk approach provides the index of the additive quantity of risks, it has several weaknesses. For instance, the inclusion of risk factors is vague and unstandardized; the information on risk factor intensity is lost with this method; the designation of risk in most cumulative risk models is arbitrary (Evans et al., 2013). Moreover, the indicators of cumulative family risks might not be comprehensive in this study for the reason that we only took negative parenting, interparental conflict, and parental psychological control into account. Researchers should integrate more relevant and significant family factors into cumulative family risks, such as parent-child relationship (Zhen et al., 2019) and family functioning (Shi, Wang, & Zou, 2017). Thirdly, there were significant differences between the complete group and the attrition group in T1 cumulative family risks and all demographic variables, resulting from the massive attrition of the fifth wave of data, which might bias the dataset. Fourthly, although internet use motivations contain or overlap the motives for using social media, using the scale for internet use motivations rather than that for social media use may blur the distinction between these two concepts, which further limits the explanation of results. Future research is encouraged to measure specific social media motivations rather than general internet motivations to specify how various urges foster problematic social media use. Last but not least, this study only controlled a small number of covariates, other factors such as attachment styles (Musetti, Manari, Billieux, Starcevic, & Schimmenti, 2022) and narcissism (Casale & Banchi, 2020) may also affect adolescent PSMU and should be considered as covariates in future research.

Regardless of the limitations above, this study elucidates the developmental trajectories of PSMU among Chinese adolescents and unveiled the longitudinal mechanism between cumulative family risks and PSMU, which supplements the literature regarding how multiple family risks foster the development of PSMU from the perspective of motivation and contributes to providing scientific guidance for intervening adolescent PSMU rooted in the family system. In addition, different from cross-sectional or short-term longitudinal designs, the present study examined the association across three years, which makes the results more compelling and robust. With the latent growth curve model, findings indicated not only how family risks impact the initial level of adolescent PSMU, but also their negative effects on the developing slope of PSMU, tracking the underlying influence in depth. Since parenting styles, interparental conflict, and parental psychological control were vital risk factors contributing to adolescents' PSMU, this could inspire several educational implications. First, it is beneficial to adopt the positive parenting style, which is characterized by reciprocal attachment, respect, and proactive as well as empathetic parenting (Vossen et al., 2024), so as to reduce the escape motivation for using social media among adolescents. Also, in the context of positive parenting, parents can set rules suitable for adolescents regarding social media usage (Vossen et al., 2024). Besides, conflict between parents has a detrimental impact on adolescents' behavioral development, so measures must be conducted to diminish interparental conflict such as family and marriage counseling. Thirdly, parents should decrease the psychological control of children, be respectful of their autonomy and thoughts, and encourage them to express and think. Fourthly, it is conducive for the government or the community to hold the family intervention program for PSMU. For example, a family-centered group program in Sweden offers theoretical and empirical knowledge on parenting and child development, aimed at helping families with adolescents who have problematic screen use (Werner, Kapetanovic, & Claesdotter-Knutsson, 2024). Finally, in addition to family factors, a variety of factors from other environments can make a difference in adolescent PSMU, such as personal factors (Saladino, Verrastro, Cannavò, Calaresi, & Barberis, 2024) and school factors (Jia et al., 2024), and hence, interventions targeting other factors should be considered. For example, the positive metacognition about the utility of social media usage is a factor motivating individuals to engage in PSMU (Casale, Rugai, & Fioravanti, 2018). As such, changing the perceptions about the belief that using social media is an efficient strategy to cope with emotions may helpfully reduce the motivation to use social media platforms. Additionally, school factors such as positive teacher-student relationships could act as a protective role in mitigating adolescent PSMU (Jia et al., 2024). If teachers can establish harmonious and equal relationships with students, adolescents tend to seek assistance from teachers when they feel upset or confused, rather than turning to social media.

Conclusion

With five-wave data that lasted for three years, results highlight the potentially deleterious role of cumulative family risks on PSMU. Furthermore, adolescents who experience cumulative family risks tend to engage in PSMU as a means of escaping from real life rather than seeking social relationships. Overall, this study traced the developmental trajectory of PSMU among Chinese adolescents, supplementing the knowledge gap in the literature regarding how cumulative family risks influence the trajectory of PSMU. The findings reflect the need for interventions aimed at mitigating family risks in order to reduce PSMU among adolescents.

Supplementary data

jba-14-1394-s001.pdf ^{(356.9KB, pdf)}

Funding Statement

Funding sources: This work has supported a grant by the National Social Science Fund of China (No. 23BSH138) to Kai Dou.

Data availability

Inquiries about the data that support the findings of this study may be requested from the corresponding author.