Abstract
Objectives
Previous psychometric network models have been limited to cross-sectional designs, while current longitudinal models also suffer from methodological shortcomings. This study challenges and refines previous longitudinal psychometric network models, addressing the temporal and static correlational relationships between mobile phone addiction and mental health symptoms, as well as gender differences.
Methods
We conducted a one-year follow-up study involving 1,479 adolescents (Mage=15.28±1.51, 786 males), and measured mobile phone addiction (inability to control cravings, feeling anxious and lost, withdrawal and escape, productivity loss) and mental health (obsessive compulsive tendencies, paranoid, hostility, interpersonal sensitivity, depression, anxiety, academic stress, maladaptation, emotional disturbance, psychological imbalance). Panel graphical vector autoregression network analysis was used to establish temporal, contemporaneous, and between-subjects networks, and estimate node centrality. Network comparisons were additionally performed between gender.
Results
The majority of mobile phone addiction predicted subsequent mental health symptoms, especially inability to control cravings, productivity loss, and anxiety. Hostility, emotional disturbance, or academic stress were also involved in temporal, contemporaneous, and between-subjects associations, respectively. There were no significant differences in the global metrics of the gender subgroup network, but these differences were reflected at the micro-level. Primarily, it revealed a temporal link between productivity loss and anxiety for females, whereas for males, it highlighted inability to control cravings and academic stress.
Conclusions
These findings supplement the longitudinal evidence from a psychometric perspective, revealing the complex network interactions between mobile phone addiction and mental health, and highlighting the key associations between gendered mobile phone use and psychological problems as intervention targets.
1. Introduction
Adolescents face critical developmental tasks that involve personality formation and rapid physical and psychological growth, making them a key focus for identifying and intervening in mental health issues. Reports indicate that 50% of mental health problems emerge before age 14. 1 Prominent mental health concerns among middle school students included depression, anxiety, sleep disorders, and self-harm.2,3 By 2025, the prevalence of severe mental health issues among Chinese adolescents had reached 10%, 4 surpassing previously-reported numbers. 5 Additionally, Ghafari et al. found that the global prevalence rate among adolescents with unmet mental health care needs was as high as 54%. 6 Unresolved mental health issues often persist into adulthood, leading to subsequent morbidity and disability. 7
In the digital information era, the impact of digital media on physical and mental health has drawn particular attention. Statistics from the China Internet Network Information Center indicate a trend toward younger internet users, with those aged 10-19 accounting for 13.7% of the total in 2025. The latest survey on minors’ internet usage revealed that 20% self-report high levels of internet dependency, with mobile phones becoming the most commonly used device for accessing the internet among young internet users. 8 Research also indicates that predictors of worsening mental health among adolescents include social isolation, excessive screen time and social media use, parental pressure and poor parent-child relationships, low socioeconomic status, and pre-existing mental health conditions or disabilities. 9 One review found that screen time positively correlates with mental health issues and that social media use increases depression risk among girls. 10 Thus, the role of mobile phone dependency in adolescent mental health symptoms warrants attention.
1.1. Mobile phone addiction
Mobile phone addiction is characterized by compulsive mobile phone use that persists despite adverse effects on physical, psychological, and social functioning.11,12 Among adolescents it typically manifests as excessive and compulsive phone usage accompanied by cravings and dependency; avoidance of social activities while relying on phones as the primary medium for information access, entertainment, and social relationship maintenance; and pronounced anxiety when separated from phones, to the extent that it disrupts normal learning and daily life, jeopardizing physical and mental health. 13
The concept of mobile phone addiction can be traced back to 2005. Bianchi et al. proposed the term “Problem Mobile Phone Use” to describe the adverse effects on physical, mental, and social functioning resulting from excessive mobile phone use. 14 Based on the Diagnostic and Statistical Manual of Mental Disorders (DSM)-4, they developed a 27-item Problem Mobile Phone Problem Use Scale (MPPUS) to measure characteristics such as tolerance, escape behavior, and negative impact on daily life among Australians.
The MPPUS has since become a powerful tool for assessing mobile phone addiction. 15 Leung revised the MPPUS into a 17-item, four-dimensional mobile phone addiction index scale. 11 The addictive characteristics of mobile phone addiction were categorized into inability to control cravings (inability to regulate phone usage time), productivity loss (reduced academic or work performance due to excessive phone use), feeling anxious and lost (negative emotional reactions upon separation from the phone), and withdrawal and escape (immersing oneself in the mobile world to escape reality). These dimensions aligned with the DSM-5 criteria for internet addiction, 16 and capture the behavioral pattern of mobile phone addiction without requiring biologically active substances. 17
Given the comprehensiveness of Leung’s framework and the cross-cultural validation of his measurement tool, this study adopts Leung’ s definition: mobile phone addiction refers to excessive mobile phone use that causes impairment in physiological, psychological, and social functioning that remains difficult to control.
1.2. Mental health
An individual’s mental health, as a comprehensive concept, denotes his overall state of psychological well-being. According to the World Health Organization (WHO), good mental health is a state of well-being in which an individual realizes his own abilities, can effectively cope with the normal stresses of life, productively use his abilities, and is able to make a contribution to his community. 18 From a positive psychology perspective, Liu posited that mental health constitutes a sustained psychological state where individuals exhibit vitality, positive inner experiences, and effective social adaptation, thereby enabling the realization of personal potential and positive social functioning. 19 Despite variations in these definitions, they consistently emphasize emotional stability, adaptive capacity, and positive impacts on oneself and the environment.
However, empirical research on mental health has adopted diverse measurement approaches. Some studies assess mental health through positive indicators such as well-being and life satisfaction,20,21 while others focus on mental health symptoms including depression, interpersonal sensitivity, and anxiety.22,23 A third approach measures both positive and negative dimensions simultaneously.24,25
Given the developmental challenges faced by adolescents and the potential negative psychological impacts of mobile phone addiction, this study focuses on mental health symptomatology, specifically examining obsessive-compulsive tendencies, paranoid, depression, and anxiety, to better identify and understand potential mental health risks in this population.
1.3. The relationship between mobile phone addiction and mental health: From variable-centered to network perspectives
Substantial research has examined the association between mobile phone addiction and mental health, though findings remain inconsistent. Traditional theoretical frameworks have predominantly adopted a variable-centered perspective, focusing on specific relationships between selected variables. Early frameworks, such as the interpersonal relationship model, primarily proposed a unidirectional association, suggesting that inadequate social support contributes to poorer mental health and increases adolescents’ reliance on mobile phones and the internet as sources of virtual support.26,27 Subsequent theories emphasized reciprocal relationships. The Interaction of Person-Affect-Cognition-Execution (I-PACE) model proposes that individual characteristics, affective and cognitive responses, and executive processes interact in the development and maintenance of problematic Internet use.28,29 More recently, the “Digital Goldilocks” hypothesis has suggested that the association between digital engagement and mental health may not be purely linear, proposing that moderate levels of digital use are not necessarily detrimental and that the relationship may vary according to the degree and context of use. 30
Taken together, these theoretical developments suggest an increasingly complex relationship between problematic mobile phone use and mental health, progressing from unidirectional associations to reciprocal and potentially nonlinear relationships. However, despite these advances, existing theories primarily operate at the construct level and provide limited insight into how specific symptoms of mobile phone addiction and mental health interact with one another.31,32 Consequently, it remains unclear which symptoms are most influential, how symptoms may reinforce one another, and whether these relationships differ by gender. These unresolved problems motivated the present study and provided the rationale for adopting a psychometric network approach.
To address these limitations, Borsboom proposed the psychometric network model, 33 which conceptualizes mental disorders not as latent variables but as emergent phenomena arising from the interaction of observable symptoms. In this framework, symptoms correspond to nodes in a network structure, and edges between nodes represent associations between symptoms. This network perspective bridges two dominant approaches in mental health research: the DSM and ICD diagnostic classifications, 34 and the multidimensional continuum model of mental health. 35 Critically, network models eliminate assumptions about causal directionality from disorders to symptoms, instead revealing how symptoms interact to maintain psychopathological states. 36 For example, rather than treating flashbacks and nightmares as mere consequences of PTSD, network models examine how these symptoms might directly trigger and reinforce each other. 37
Emerging research has begun applying network analysis to understand mobile phone addiction and mental health. Panayiotou et al. constructed a network model integrating social media use, psychological symptoms, 38 school and peer relationships, and positive functioning, revealing that social media exerted a relatively minor impact on adolescent mental health compared to bullying and lack of family support. Yang et al. found that mental health problems served as precursors to mobile phone addiction in their network model, 39 though the reverse predictive pathway did not reach statistical significance. However, these studies predominantly employed longitudinal designs, which reflect static between-person differences but cannot distinguish individual characteristics from group patterns. 40
1.4. The necessity and advantages of longitudinal panel network analysis
Network analysis methods are increasingly recognized as powerful tools in psychological science. 41 Based on data structure, network models can be categorized into cross-sectional, time-series, and panel networks. Cross-sectional networks construct symptom networks from single-timepoint measurements across multiple individuals, revealing common associations among symptoms but providing no information about temporal associations or within-person processes. 42 Networks using time-series data, derived from intensive repeated measurements of single individuals or groups, can analyze both contemporaneous associations (conditional relationships between variables at the same timepoint) and temporal associations (lagged effects across timepoints), but findings from single individuals are difficult to generalize to populations. 43
Instead, panel networks decompose variance into three distinct structures that address different research questions. The temporal network captures lagged effects of variables across timepoints within individuals, revealing predictive relationships essential for examining whether mobile phone addiction prospectively predicts mental health symptoms. The contemporaneous network represents conditional associations between variables at the same timepoint within individuals, illuminating how symptoms co-occur and interact while controlling for temporal effects. The between-person network reflects stable individual differences independent of time, revealing trait-like associations and allowing examination of whether network patterns differ across subgroups such as gender.
Thus, for understanding the relationship between mobile phone addiction and mental health in adolescents, longitudinal panel networks offer several critical advantages. First, adolescents exhibit multidimensional and complex mental health symptoms that interact in temporal ways, making network analysis particularly well-suited for capturing these reciprocal influences. Second, to determine whether mobile phone addiction represents transient physiological dependence or a problematic behavior with enduring mental health consequences, tracking psychological and behavioral characteristics across time is essential. Third, panel networks can distinguish whether observed associations reflect stable between-person differences (e.g., some individuals consistently show higher levels of both addiction and depression) versus within-person associations (e.g., increases in addiction predict subsequent increases in depression within the same individual). This distinction is crucial for developing targeted interventions.
1.5. The present study
Despite substantial evidence regarding mobile phone addiction and mental health symptoms,10,44,45 prior research has predominantly employed theory-driven, variable-by-variable analytical approaches that validate specific hypothesized relationships while overlooking the broader network architecture and temporal associations. To address these gaps, this study constructs a longitudinal panel network model of adolescent mobile phone addiction and mental health symptoms, as illustrated in Figure 1. This approach aims to reveal the salient characteristics and cross-temporal stability of these phenomena both within and across individuals, providing scientific insights for public health prevention strategies. Longitudinal panel network model. Note. The model illustrates the three-layer structure of relationships between mobile phone addiction and mental health symptoms: temporal network (cross-lagged effects from T1 to T2), contemporaneous network (within-associations), and between-person network (stable individual differences).
Based on the longitudinal panel network framework, this study addresses three core research questions corresponding to the three network structures as well as their gender differences: (1) Temporal network: Can adolescent mobile phone addiction prospectively predict the emergence of subsequent mental health symptoms? Which symptoms exhibit the strongest temporal effects across time? (2) Contemporaneous network: What patterns of contemporaneous associations exist between mobile phone addiction and mental health symptoms at the same timepoint? Which symptoms serve as central nodes that bridge different symptom clusters? (3) Between-person network: What stable characteristics emerge in the between-person network structure of mobile phone addiction and mental health symptoms?
2. Methods
2.1. Participants
Descriptive statistics of participant.
To prevent secondary trauma of mental health problems, or reduced reliability of self-reported questionnaires, we excluded groups with previous or current mental disorders. According to previous studies, this study was conducted on a non-clinical sample of adolescents.49–52 The written informed consent of the participants’ parents or teachers was obtained before the formal investigation began, and this study was approved by the medical ethics review committee of the Guangzhou Medical University.
2.2. Measurements
2.2.1. Mobile phone addiction index
Our study used the Mobile Phone Addiction Index (MPAI) developed by Leung, 11 and the Chinese version revised by Huang et al. to measure adolescents’ mobile phone addiction. 15 The scale consists of 17 items, divided into 4 dimensions: inability to control cravings (7 items), productivity loss (4 items), feeling anxious and lost (3 items), and withdrawal and escape (3 items). The scale adopts a Likert 5-point rating, where 1 means “never” and 5 means “always”. Higher scores indicate a greater degree of mobile phone addiction. The Cronbach’s α values for the MPAI were 0.926 at T1 and 0.906 at T2.
2.2.2. Middle School Student Mental Health Scale
Our study also employed the Middle School Student Mental Health Scale (MSSMHS) developed by Wang to assess mental health symptoms among adolescents. 53 This scale comprises 60 items across 10 dimensions: obsessive-compulsive tendencies, paranoid, hostility, interpersonal sensitivity, depression, anxiety, academic stress, maladaptation, emotional disturbance, and psychological imbalance, each with 6 items. It also uses a 5-point rating: “None” gets 1 point, “Mild” 2 points, “Moderate” 3 points, “Slightly Severe” 4 points, and “Severe” 5 points. The total score for each dimension is divided by 6 to obtain the average score for that dimension. Dimension averages correspond to mild (2-2.99), moderate (3-3.99), significant (4-4.99), and severe (5) symptoms. The Cronbach’s α values for the MSSMHS were 0.975 at T1 and 0.974 at T2.
2.2.3. Demographic variables and covariates
Demographic characteristics of participants, including gender, age, only-child status, family economic status, and family structure, were measured through self-reported questionnaires. Drawing on prior research,38,54 our study considered only-child status, family economic status, and family structure as covariates. Linear regression equations were estimated separately for each network model’s primary variable and these three covariates. Residuals were obtained after controlling for covariates and then used as input data to form network models. This ensured that network analysis remained unaffected by the aforementioned confounding variables.
2.3. Data analysis
All data were cleaned and coded using R Studio 4.5.2. 55 Categorical variables were converted to factor types prior to analysis: gender (male=1, female=2), only-child status (yes=1, no=0), family economic status (tight=1, average=2, good=3), and family structure (single-parent=1, reconstituted=2, skip-generation=3, normal=4). Node variables were standardized by Z-score to avoid the difference in network structure and centrality index caused by the different scoring scales, and covariates were controlled for obtaining variable residuals, 38 followed by model fitting and selection, longitudinal network and node centrality estimations, and network comparisons.
2.3.1. Model selection
An optimal panel network model was estimated using the psychonetrics package in R Studio with full information maximum likelihood (FIML) to handle missing data. 43 Model selection followed a combination of pruning and step-up estimation procedure aimed at obtaining an interpretable and parsimonious network structure. 56 First, a baseline model was estimated that included all temporal, contemporaneous, and between-person parameters. Second, non-significant edges were removed using a recursive pruning procedure based on statistical significance (α=0.01), and the model was re-estimated with the pruned parameters fixed to zero. This conservative threshold was adopted to improve model fit and stability. Third, starting from the pruned model, step-up estimation was applied to iteratively add parameters that improved model fit according to the Bayesian Information Criterion (BIC). The final model was selected based on overall model fit and BIC, and model comparisons among the baseline, pruned, and step-up models were conducted to document the model selection process. To ensure convergence and numerical stability during estimation and model search, approximate standard errors were used throughout the modeling procedure.
Besides, as the primary goal of the subgroup analyses was to compare parameters under a common model specification rather than to discover structure, independent model selection in smaller subsamples may have led to unstable or degenerate solutions. Thus, the final network structure was held constant and re-estimated separately in female and male subsamples using FIML, allowing parameters to vary while preserving network topology. This approach prevented model overfitting in the gender subsample and ensured that differences in network structure between the two samples were not due to differences in model selection.
2.3.2. Estimation of the longitudinal network
Longitudinal network models were constructed using the psychonetrics package in R Studio. 43 Fourteen dimensions representing mobile phone addiction and mental health symptoms were selected as node variables. All node variables were regressed on relevant covariates (age, only-child status, family economic status, and family structure) prior to network construction. The resulting standardized residuals were then utilized as input for the psychonetrics models. This residual-based approach effectively partials out the potential confounding effects of demographic factors, ensuring that the identified edges reflect the pure psychological temporal associations between mobile phone addiction and mental health symptoms rather than spurious correlations driven by shared external variance. Network model construction, node centrality calculations, and visualization were implemented using the ggplot2 57 and qgraph packages. 58 Edge thickness in the network model was estimated using partial correlation coefficients, where edges between nodes represented associations between them, with positive associations depicted as blue edges and negative associations as red edges. When the edge weight value exceeded 0.03, the relationship between nodes was considered meaningful. 59
To disentangle within-person state associations from stable between-person trait differences, the longitudinal network analysis was conceptualized into three distinct network structures using the panel graphical vector autoregression (gVAR) framework. Traditional cross-lagged panel models often conflate within-person processes with between-person stable differences, which can lead to ecological fallacies. The Panel gVAR framework overcomes this limitation by disaggregating the data variance into three distinct, orthogonal network structures. To ensure a parsimonious and highly replicable model, a strict step-down model pruning strategy was applied prior to final network construction, retaining only edges that were statistically significant at a conservative threshold of α=0.01: (1) Contemporaneous Network (within-person): This network captures the conditional dependencies between mobile phone addiction and mental health symptoms at the same time point, controlling for temporal effects. It was estimated as a Gaussian Graphical Model (GGM), where edges represent undirected within-person partial correlations. 58 (2) Temporal Network (within-person): This network estimates the cross-lagged predictive relationships of variables from one time point to the next. Using a vector autoregression structure, 60 it calculates standardized directed coefficients to reveal directional pathways over time, simultaneously controlling for the autoregressive and cross-lagged effects of all other variables. (3) Between-Person Network: This network explores stable, trait-like inter-individual differences. It was estimated via a GGM based on the stationary means (random intercepts) of the individuals, forming an undirected network of partial correlations that demonstrates how average levels of node variables are stably associated across the sample. 43
2.3.3. Estimation of node centrality
Expected Influence (EI) was employed as the centrality metric to capture node characteristics and variation patterns of node variables in network models. EI, 61 proposed by Robinaugh et al. overcomes limitations of traditional strength-based centrality by appropriately handling networks with both positive and negative connections, making it a more reliable indicator of node importance in complex symptom networks. In temporal networks, EI was decomposed into directional metrics: In Expected Influence (IEI) represents the incoming influence from other nodes at earlier timepoints, while Out Expected Influence (OEI) reflects the outgoing influence on other nodes at later timepoints. 38 Nodes with high EI values represent significant influence across temporal associations and may serve as critical factors sustaining the symptom network.
2.3.4. Comparing gender subgroup networks
We used the NetworkComparisonTest package to compare networks across gender subgroups at the same time point. 62 Two networks’ overall structures were compared using the network invariance test, and their global strengths were compared using the global strength invariance test. The edge and centrality invariance tests allow us to compare the variations in the EI centrality of each edge and each node. The Benjamini-Hochberg procedure was used to adjust p-values for multiple comparisons.
3. Results
3.1. Descriptive statistical results and model fitting
Descriptive statistical results of node variables.
Model fitting parameter results in the full sample.
3.2. Network estimation
3.2.1. Network structures
Three network structures were estimated, including temporal, contemporaneous, and between-subjects. For the full sample, the temporal network was sparse yet clear (Density=0.071), with edge weights ranging from -0.015 to 0.098 (M=0.048) after model fitting. Both the contemporaneous (Density=0.439) and between-subjects networks exhibited more edges (Density=0.133), with contemporaneous edges showing weaker associations (Range: -0.090-0.483, M=0.152) compared to between-subjects edges (Range: 0.189-0.644, M=0.394). Network visualization is shown in Figure 2. The original temporal network was fully saturated without pruning (Density=1.000, Range: -0.079-0.126, M=0.011). Contemporaneous and between network structures were equally compact (Density=0.929), and contemporaneous edge weights ranged from -0.111 to 0.539 (M=0.072). Between-subjects edges ranged from -0.104 to 0.322 (M= 0.066). The original network visualization is shown in Figure 3. Comparing the baseline model to the step-up model in the full sample in terms of goodness of model fitting and network visualization, we ultimately selected the step-up model’s network for subsequent analysis. Longitudinal network in full sample (step-up model). Note. Before network estimation, models were fitted and selected, eliminating most non-significant edges. Purple nodes represent mental health symptoms, while blue nodes indicate mobile phone addiction. Blue edges indicate positive associations, red edges denote negative associations. The thicker, darker the edge, the stronger the relationship. (a) Edges are directed partial correlations in temporal network, removed the autoregressive paths; (b, c) Edges are partial correlations in contemporaneous, between networks. Longitudinal network in full sample (baseline model). Note. The original network structure, with edges not pruned. Purple nodes represent mental health symptoms, while blue nodes indicate mobile phone addiction. Blue edges indicate positive associations, red edges denote negative associations. The thicker, darker the edge, the stronger the relationship. (a) Edges are directed partial correlations in temporal network; (b, c) Edges are partial correlations in contemporaneous, between networks.
We also estimated the network based on gender groupings in Figure 4. In the temporal network, both sets exhibited saturated structures (Density=1.000), but females’ edges connected more strongly (Female: Range: -0.186∼0.141, M=0.011; Male: Range: -0.081∼0.132, M=0.009). The density of both contemporaneous and between networks was comparable under gender grouping (Density=0.929). The slight differences between the two were reflected in the edge weights in contemporaneous (Female: Range: -0.226∼0.628, M=0.072; Male: Range: -0.159∼0.536, M=0.072), between networks (Female: Range: -0.182∼0.408, M=0.061; Male: Range: -0.117∼0.318, M=0.068). The edge-weight matrices for all networks are shown in Supplemental Table S1. Longitudinal network in gender subgroup. Note. The original network structure, with edges not pruned. Purple nodes represent mental health symptoms, while blue nodes indicate mobile phone addiction. Blue edges indicate positive associations, red edges denote negative associations. The thicker, darker the edge, the stronger the relationship. (a, d) Edges are directed partial correlations in temporal network; (b, c, e, f) Edges are partial correlations in contemporaneous, between networks.
3.2.2. Temporal associations
In the full sample, the most prominent cross-domain temporal effects originated from mobile phone addiction toward mental health symptoms in the temporal network. Specifically, inability to control cravings positively predicted hostility (β=0.030) and paranoid (β=0.022). Interpersonal sensitivity also predicted increased withdrawal and escape (β=0.028). Gender-specific analysis further revealed distinct patterns. For females, productivity loss became a relatively important predictor variable, negatively predicting anxiety (β=-0.182) and hostility (β=-0.146). Inability to control cravings also showed positive temporal association with anxiety (β=0.141). For males, the cross-domain temporal effects were characterized by an inability to control cravings, which positively predicted academic stress (β=0.132), paranoid (β=0.117), and depression (β=0.107).
3.2.3. Contemporaneous associations
At the contemporaneous level, inability to control cravings and emotional disturbance emerged as a significant bridge link in the full sample (r=0.128). Both productivity loss and academic stress, as well as withdrawal and escape and interpersonal sensitivity, showed the same degree of association (r=0.108). In the female subgroup, inability to control cravings subsequently developed strong contemporaneous associations with interpersonal sensitivity (r=0.234), obsessive compulsive tendencies (r=-0.202), and hostility (r=0.183). Productivity loss and depression showed a negative correlation (r=-0.159), and feeling anxious and lost, hostility (r=0.118), and inability to control cravings and emotional disturbance (r=0.105) were positively correlated among males.
3.2.4. Between-subject associations
Between-subjects analysis, reflecting stable association between individuals, highlighted productivity loss and academic stress as the strongest cross-domain link in the full sample (r=0.344). The interactions between inability to control cravings and emotional disturbance (r=0.325), as well as productivity loss and obsessive-compulsive tendencies (r=0.308), were also considerable. In the female group, these were reflected in the following three node pairs: inability to control cravings and emotional disturbance (r=0.249), productivity loss and academic stress (r=0.204), and feeling anxious and lost and academic stress (r=0.193). For males, the top three between level associations were productivity loss and obsessive-compulsive tendencies (r=0.235), feeling anxious and lost and anxiety (r=0.195), productivity loss and psychological imbalance (r=0.143).
3.3. Node centrality estimation
3.3.1. Out- and In-EI in temporal networks
Estimation and visualization of node centrality for the full sample, females, and males are shown in Figures 5 and 6. In the temporal network, OEI and IEI were defined as indicators of node centrality. OEI indicated a stronger predictive impact on other symptoms over time. In contrast, the variables with the highest IEI influence, reflecting greater susceptibility to influence from other nodes. Node centrality in contemporaneous and between networks were undirected, demonstrated using EI. All networks’ EI values are shown in Supplemental Table S2. Expected Influences in full sample. Note. Purple nodes represent mental health symptoms, while blue nodes indicate mobile phone addiction. The zero line was shown as a dashed line. (a, b) Out- and In- Expected Influence in temporal network. Out Expected Influence indicated a stronger predictive impact on other symptoms over time. In Expected Influence reflected a greater susceptibility to influence from other nodes; (c) Expected Influence in contemporaneous network; (d) Expected Influence in between network. Expected Influences in gender subgroups. Note. Purple nodes represent mental health symptoms, while blue nodes indicate mobile phone addiction. Squares represent the female group, triangles represent the male group. The zero line was shown as a dashed line. (a, b) Out- and In- Expected Influence in temporal network. Out Expected Influence indicated a stronger predictive impact on other symptoms over time. In Expected Influence reflected a greater susceptibility to influence from other nodes; (c) Expected Influence in contemporaneous network; (d) Expected Influence in between network.
For the full sample, inability to control cravings emerged as the node with the highest OEI (OEI=0.115). This was followed by depression (OEI=0.102) and obsessive-compulsive tendencies (OEI=0.098). Temporal IEI indicated the most significant impact on paranoid (IEI=0.084), followed by anxiety (IEI=0.082) and hostility (IEI=0.069). Similarly, females also had the highest OEI in inability to control cravings (OEI=1.129). Among females, three variables exhibited the strongest temporal outgoing EI. Inability to control cravings showed the largest OEI (OEI=1.129), followed by withdrawal and escape (OEI=1.116) and feeling anxious and lost (OEI=1.015). In IEI, they were interpersonal sensitivity (IEI=0.424), maladaptation (IEI=0.367), and depression (IEI=0.298). For males, inability to control cravings also emerged as the node with the highest temporal outgoing influence (OEI=1.010). This was followed by psychological imbalance (OEI=0.615) and withdrawal and escape (OEI=0.287), although their magnitudes were notably smaller than those observed in females. Regarding temporal ingoing influence, academic stress had the highest IEI (IEI=0.198), closely followed by anxiety (IEI=0.197) and interpersonal sensitivity (IEI=0.195).
3.3.2. EI in contemporaneous networks
At the contemporaneous level, similar results were observed across the full sample, as well as among both females and males. The first two nodes of the EI for all three were anxiety (full sample vs. females vs. males: EI=1.193 vs. 1.229 vs. 1.164) and depression (full sample vs. females vs. males: EI=1.129 vs. 1.141 vs. 1.105). The slight difference was reflected in the third-highest node. The third-highest node for males was inability to control cravings (EI=1.080), but for the full sample and females, it was interpersonal sensitivity (full sample: EI=1.078; females: EI=1.129).
3.3.3. EI in between networks
In the full sample, productivity loss emerged as the most central node at the between-subject level (EI=1.542), followed by anxiety (EI=1.422) and interpersonal sensitivity (EI=1.253), indicating that these variables were most strongly connected to other nodes across individuals. When stratified by sex, distinct patterns were observed. Among females, emotional disturbance showed the highest value (EI=1.508), with anxiety (EI=1.423) and interpersonal sensitivity (EI=1.250) ranking second and third, respectively. In contrast, for males, interpersonal sensitivity was the most central between-subject node (EI=1.506), followed by anxiety (EI=1.269) and emotional disturbance (EI=1.111).
3.4. Network comparisons between genders
The results of the Network Comparison Test (NCT) indicated that the symptom networks of males and females at T1 were statistically invariant. No significant differences were observed in overall network structure (M=0.142, p=0.331>0.05), global network strength (S=0.019, p=0.918>0.05), individual edges, or EI centrality (p>0.05). The same result was also obtained in the comparison at T2 although the differences were slightly more pronounced (network structure: M=0.168, p=0.083>0.05; global network strength: S=0.038, p=0.815>0.05; edges or EIs: p>0.05). These comparison results are given in Supplemental Figures S1-S6.
4. Discussion
This study first adopted a longitudinal network approach to disentangle temporal, contemporaneous, and between-subject associations between mobile phone addiction and mental health symptoms. Reviewing the study’s core questions, we found that the strong associations of temporal networks, contemporaneous networks, and between-subjects networks were all involved by inability to control cravings and productivity loss of mobile phone addiction. These symptoms were able to predict the occurrence of subsequent mental health symptoms, such as hostility, anxiety, and academic stress. In particular, inability to control cravings showed the highest symptom impact. Emotional disturbance, interpersonal sensitivity, and depression were the most contemporaneous relevant mental health problems in each sample groups. Similarly, anxiety was the most prominent symptom. Between-subjects networks also revealed different core symptoms and network structures. Mobile phone addiction was associated with academic stress, emotional disturbance, and obsession-compulsive tendencies.
4.1. Mobile phone addiction inspired subsequent mental health problems
Temporal network captures the sequential, directional lagged effects within individuals over time. It reveals the directional pathways through which one state triggers or maintains another, unveiling the underlying chronological mechanisms of the syndrome. By identifying which node temporally drives subsequent symptoms, clinicians can design targeted, preventative interventions to disrupt the propagation of the symptom cascade before it escalates. At the temporal level, our findings consistently indicated that mobile phone addiction functioned as a primary driver of subsequent symptoms in mental health. 63 Across the full sample, difficulty controlling cravings for mobile phone use emerged as a central predictor of later hostility. This association reflected the fact that repeated failures to regulate phone use led to irritability, lowered tolerance for external demands, and heightened proneness to anger. Previous research has shown that impaired behavioral control is closely linked to aggressive and hostile responses, particularly when individuals experience perceived interference with goal directed behavior.64,65 Over time, such dysregulation may manifest as increased hostility rather than internalizing symptoms alone.
Gender-specific temporal patterns further illuminated different psychological processes. Interestingly, productivity loss negatively predicted subsequent anxiety among females. It may reflect a short-term avoidance or disengagement process. Females tend to report higher levels of perfectionistic concerns and role overload, making them particularly vulnerable to stress associated with academic and social responsibilities.66,67 In this context, reduced task engagement resulting from smartphone-related productivity loss may temporarily alleviate performance pressure and provide psychological relief, thereby lowering subsequent anxiety levels. Productivity loss became their self-protective withdrawal strategy. 68 This interpretation is also consistent with theories of avoidance coping, 69 which suggest that disengagement from stressors can reduce anxiety in the short term despite potentially leading to maladaptive outcomes over time. In contrast, for males, inability to control cravings predominantly predicted increases in academic stress over time. This pattern suggests that uncontrolled phone use may directly interfere with academic functioning and performance expectations, which are often more strongly tied to stress appraisal among males. Empirical studies have indicated that males are more likely to externalize the consequences of behavioral dysregulation into performance-related stress rather than affective distress. 70 As a result, persistent difficulties in regulating phone use may accumulate into heightened academic strain rather than immediate emotional symptoms.
4.2. Strong contemporaneous associations between behavioral and emotional states
Contemporaneous network analysis revealed within-individual correlation structures between mobile phone addiction and mental health symptoms after partialing out both between-person traits and temporal lagged effects. Highly central nodes within this network serve as optimal targets for diagnostic screening, as these core indicators exhibit strong synchronous associations with the immediate escalation of the surrounding symptom system. In the full sample, difficulty controlling cravings in using phone showed a strong contemporaneous association with emotional disturbance, pointing to the bridge between behavioral dysregulation and affective instability. This conclusion aligns with the compensatory internet use theoretical model, whereby mobile phone users who experience negative real-world events or negative emotions engage in compensatory online addiction as a form of self-healing. 71 Previous studies have found that mobile phone addiction is positively associated with negative emotions. 72
Among females, a strong contemporaneous association was found between difficulty controlling cravings and interpersonal sensitivity. Mobile phone addicts are more likely to ignore real social interaction and turn to virtual social interaction. It also leads to alienation and sensitivity of interpersonal relationships in reality. 72 Girls’ identity and interpersonal relationships are closely related, and they pay more attention to maintaining peer relationships. 73 Prior research has shown that females, on average, report higher levels of interpersonal concern and relational self-focus, particularly in contexts involving perceived self-regulatory failure.67,74 Mobile phone addiction damages this peer relationship and makes them prone to bad social behavior. For girls, relational aggressive behaviors are more likely to be adopted, 75 such as rejection and exclusion. 76 In this light, difficulty controlling phone use may not only represent a behavioral issue but also activate immediate concerns about social judgment or relational expectations. Empirical studies on problematic smartphone use have similarly found that social-related distress and sensitivity are closely intertwined with perceived loss of control, especially among females. 77 Therefore, interpersonal sensitivity has become a significant mental health problem for female mobile phone addicts.
Other contemporaneous couplings, such as the negative correlation between productivity loss and depression, were observed in males. The productivity loss of middle school students was mainly reflected in academic procrastination, and males reported more of it. 78 Research on self-regulation and procrastination suggests that disengaging from effortful tasks can momentarily reduce negative mood. 79 Mobile phone addiction as a manifestation of self-regulation failure further reflects the lack of motivation of adolescents. Short-term fixes to the irritability and depression that come with being faced with tasks take precedence over completing academic tasks.79,80 Procrastination gives a temporary emotional boost, although it’s not beneficial in the long run.
4.3. Stable cross-domain associations at the between-subject level
Between networks reflects the long-term covariation between an individual’s average level of mobile phone addiction and mental health symptoms by identifying stable associations. 58 Rather than short-term fluctuations in symptoms in contemporaneous networks, between subjects’ associations are based on common characteristics of the adolescent participants. We concluded from the between network that students with high inefficiency tend to have high academic pressure. Interestingly, this also revealed a dilemma faced by Chinese students. Clinically, these findings provide macro-level insights for long-term risk profiling, allowing clinicians to identify specific sub-populations or enduring symptom clusters that consistently co-occur over extended periods. Their academic achievement is not only personal, but also tied to family and class responsibilities. 81 Therefore, they had a heavy academic load. However, the higher the academic pressure, the lower the academic achievement. 82 As an academic barrier, stress affects working memory and reduces academic efficiency.83,84 If studies can’t be finished or academic performance was poor, the pressure was even higher, and mobile phone addiction could therefore even increase the perception of stress. 85
In females’ between networks, inability to control cravings and emotional disturbance showed a stable association. Unlike the short-term activation observed in the contemporaneous network, females with long-term uncontrolled phone using also continuously experience emotional vulnerability. This revealed stable trait-like individual differences in the co-occurrence of mobile phone use difficulties and emotion regulation problems among girls. Although previous studies have found that girls show greater self-control, 86 this did not apply to mobile phone use. Adolescent girls use mobile phones as a communication tool for social networks, 87 a manifestation of their tendency to be social and independent. 88 They are eager to chat and build friendships on social networks. Girls are better able to express and interpret emotions based on their interdependent survival mode and gender roles that value interpersonal relationships. 89 Friendships of adolescent girls are characterized by high intimacy, support, and loyalty, which can also bring interpersonal stress and easily lead to emotional conflict. 90 Therefore, emotional disturbance is also a common psychological feature of this group.
For males, our results indicated an association between productivity loss and obsessive-compulsive tendencies. In the non-clinical adolescent sample, the dimension of obsessive-compulsive tendencies was generally higher in females, but the stable connection between these tendencies and productivity loss mainly occurred in the male between-network. The high score of obsessive-compulsive tendencies in females reflected their level of expression of this trait, which due to concerns, self-monitoring, or perfectionism,91–93 they are more likely to report as being high. Gender subgroup network analysis revealed the characteristics of obsessive-compulsive tendencies in males, specifically manifested as repeated checks leading to the slowdown of the task and interruptions in learning due to incorrect completion or intrusive thoughts. Similar previous studies have found that in the non-clinical population, males exhibit more doubts and intrusive thoughts, and that these were positively correlated with low perseverance and low self-control. 94 At the same time, the productivity loss among boys manifests as reduced effort in academic studies. Low academic effort brings more benefits to boys, such as becoming popular due to conforming to typical gender temperament and preferring easy achievement realization to reduce learning investment. 95 Poor self-actualization and poor school life adaptation also reflected a stable characteristic of both low efficiency and obsessive-compulsive tendencies in adolescent boys. 96
4.4. Core symptoms shown in the longitudinal network
EI reflects the overall connection strength and influence potential of a node in network analysis, with high EI values indicating nodes that may play critical roles in activating and maintaining overall psychological symptoms. Symptoms of high EI can be considered as the focus of intervention or screening for adolescent mental health problems. 60
Our temporal network results showed that the core symptom of the full sample or the gender sub-sample was inability to control cravings, which had the highest OEI. Uncontrolled mobile phone use was the node with the highest impact on other mental health symptoms. Similar to previous work, excessive use on mobile phone was the most severe symptom. 97 Middle school students experience busy and high-pressure studies, and their schools and families strictly control mobile phones. 98 To relieve emotions, promote relaxation, and meet other needs, it is difficult for them to refrain from using mobile phones.99,100
At the contemporaneous level, anxiety was the symptom with the highest EI value, which also confirmed results of previous studies.101,102 Although anxiety did not become a constitutive symptom of the strongest edge in the contemporaneous network, as a core symptom, it helped to activate multiple other symptoms. In middle school, adolescent anxiety is particularly manifested in test anxiety due to intensive testing and immediate visualization of results. 103 In addition, these children also face changes in family relationships, peer relationships, and self-cognition, so anxiety has become a common emotional problem among Chinese middle school students. 104
The core symptoms in between networks were low productivity for all, emotional disturbance for girls, and interpersonal sensitivity for boys. The core symptom of academic inefficiency, such as excessive use of mobile phones to the extent that they negatively affect school work, also appeared in Yang’ s study. 105 Emotional disturbance in females reflects rumination and is characterized by emotion. This is further associated with depression. 106 Zhang et al. noted that, 102 due to social expectations, an independent and strong male image, it is often difficult for males to disclose interpersonal problems. However, peer and family relationship conflict is more direct in males, and at the between level, stable interpersonal sensitivities were observed in our study. To sum up, inability to control cravings could be designated as the primary target for clinical intervention, it is capable of predicting temporally and actively triggering a series of other symptoms that occur at the next time point. By clinically targeting and neutralizing this symptom, clinicians may effectively disrupt the temporal maintenance mechanism of the network and prevent the escalation of the broader symptom cluster. Anxiety or productivity loss demonstrated the highest EI within the contemporaneous or between-person network, meaning they maintain the strongest synchronous ties with other symptoms in the system. Clinicians can utilize these highly central symptoms as key screening indicators to identify individuals at high risk for the overall syndrome.
4.5. Network comparisons of gender at the same time point
To provide additional evidence regarding gender differences, we conducted NCTs of the male and female networks at T1 and T2. No significant gender differences were observed in global strength, edge weights, or node centrality indices at either time point. However, the absence of significant differences should not be interpreted as evidence that the two networks were identical or equivalent 107 ; rather, it indicates that no statistically significant differences were detected in the cross-sectional network structures. Furthermore, cross-sectional and longitudinal network analyses capture different aspects of symptom organization. Cross-sectional networks characterize contemporaneous associations among symptoms at a given time point, whereas longitudinal network analyses focus on temporal relationships. Therefore, gender differences in longitudinal processes may exist even when cross-sectional network structures appear similar. 108
4.6. Limitations and prospects
Our study developed longitudinal network models of mobile phone addiction and mental health symptoms in a nonclinical sample of adolescents. This study theoretically advances the previous research by simultaneously identifying stable, trait-level associations between addictive behaviors and psychological distress across the population, while capturing rapid, state-level changes within individual adolescents over time. Our findings provide a more precise framework for assessment and intervention. Practitioners can utilize central variables from the contemporaneous network as core indicators for early screening, while leveraging macro-level between-person associations for long-term risk profiling in school mental health programs. However, some limitations must be considered. Participants with a current or past psychiatric diagnosis were excluded, which may have introduced selection bias and limited the generalizability of the findings to adolescents with clinically significant psychopathology. Although this approach is commonly used to obtain relatively homogeneous community samples and reduce confounding effects of psychiatric comorbidity, treatment, and medication use, the resulting symptom network may not fully capture the structure of mental health symptoms in clinical populations. Future studies should include both clinical and non-clinical adolescents to examine whether symptom networks differ across populations. Only two time points were followed longitudinally in our study, with three or more waves may provide a more fully dynamic VAR process and adequately capture temporal dynamics. Although we controlled for exogenous variables, such as only-child status, family economic status, and family structure, the residuals of the variables were used. However, the one-year interval between T1 and T2 may have been too long to adequately capture the temporal changes in mental health symptoms and mobile phone addiction. Associations between symptoms may fluctuate over shorter periods, and important intermediate processes may have been missed due to the relatively long follow-up interval. Future studies may benefit from adopting shorter assessment intervals such as three or six months to better characterize developmental changes. In addition, we did not assess major life events or environmental changes occurring during the follow-up period. Adolescents may have experienced significant events, such as academic stress, family changes, interpersonal difficulties, or health-related problems, which could have influenced both mental health symptoms and mobile phone addiction. The absence of these measures limits our ability to account for potential confounding factors and may partly explain changes observed in the networks. Future longitudinal studies should incorporate repeated assessments of life events and contextual factors to better understand the mechanisms underlying symptom network changes.
5. Conclusion
The results of this study indicate that mobile phone addiction and mental health are linked through multiple pathways that operate at several different levels of analysis. The loss of control and inefficiency caused by mobile phone use are the main manifestations. Adolescent hostility, emotional disturbance, and academic stress are issues that require equal attention. The gender-specific usage of mobile phones also indicates that different psychological issues require personalized intervention. For instance, these include anxiety, interpersonal sensitivity, and emotional disturbance among females, as well as academic pressure, depression, and obsessive-compulsive tendencies among males. Interpreting these networks results is essential for advancing both theory and applied research in digital behavior and mental health.
Supplemental Material
Supplemental Material - Temporal relationships between mobile phone addiction and mental health: A longitudinal network analysis study from non-clinical adolescents
Supplemental Material for Temporal relationships between mobile phone addiction and mental health: A longitudinal network analysis study from non-clinical adolescents by Jiaying Li, Haiyan Wu, Jing Zhang, Huizhen Fu, Chang Liu, Yun Li in Digital Health
Footnotes
Ethical considerations
Author declarations Ethics approval, and consent This study was approved by the medical ethics review committee of Guangzhou Medical University (approval number: 202505006). The written informed consent of the participants’ parents or teachers was obtained before the formal investigation began. Consent for publication Written informed consent was obtained from all authors, participants, and their parents, and permission was given for the publication of the article.
Author contributions
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Guangzhou Basic and Applied Basic Research Program (Young Doctoral “Qihang” Project, Grant No. 2025A04J4058), the Young Scientists Fund of the National Natural Science Foundation of China (Grant No. 32400894), the Basic and Applied Basic Research Foundation of Guangdong Province (Grant No. 2023A1515110350), the Youth Project of the Guangdong Office of Philosophy and Social Science (Grant No. GD24YXL06), the Guangdong-Hong Kong-Macao Greater Bay Area Medical and Health Industry High Quality Development Rule of Law Guarantee Research Center (Grant No. 2024TSZK016).
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data Availability Statement
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.
Data guarantor
The corresponding author acted as the guarantor of the study and takes responsibility for the integrity of the work.
Supplemental material
Supplemental material for this article is available online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
