Internet gaming disorder (IGD) is a persistent and maladaptive pattern of online gaming associated with significant functional impairments and is increasingly recognized as a major public health concern [1]. IGD is characterized by impaired control over gaming, prioritization of gaming over other activities despite negative consequences, and persistent gaming behavior despite functional impairment [2]. In population-based research, these manifestations are commonly examined in terms of IGD symptoms, which reflect the severity of problematic gaming behaviors [3-7]. Higher levels of IGD symptoms have been linked to a range of detrimental outcomes, including psychological difficulties, behavioral problems, and social consequences [8-11]. Given the widespread accessibility of online games and the developmental vulnerability of adolescence [12,13], understanding the longitudinal mechanisms underlying IGD symptoms and identifying relevant risk and protective factors during this period is particularly important. However, the dynamic interplay among behavioral and emotional factors related to IGD symptoms remains insufficiently understood.

The displacement hypothesis posits that time is limited, so engagement in one activity reduces the time available for others [14]. In this study, this mainly concerned adolescents’ leisure time outside of structured school activities. In Chinese schools, mobile phone use during class is restricted, and gaming typically occurs after school or during other leisure periods [15]. Therefore, greater engagement in physical activity (PA) may reduce opportunities for extended gaming and, in turn, lower the risk for IGD symptoms [16,17]. In addition, PA may promote healthier lifestyle routines, better emotional regulation, and alternative sources of enjoyment and social interaction, thereby reducing adolescents’ reliance on gaming as a primary leisure activity [18-20]. At the same time, this relationship may be bidirectional. Excessive gaming or more severe IGD symptoms may reduce participation in PA by consuming leisure time, lowering motivation for physical movement, and disrupting daily routines or social patterns [21-23]. Taken together, these mechanisms suggest a potentially bidirectional relationship between PA and IGD symptoms.

Accumulating empirical evidence indicates a negative association between PA and IGD symptoms. Cross-sectional investigations across diverse cultural contexts, such as Australia, Taiwan, Portugal, Germany, and China, have consistently shown that higher PA is associated with decreased IGD symptoms [16,17,24-26]. Systematic reviews and meta-analyses have further indicated that exercise-based interventions can reduce gaming addiction scores and may be among the most effective interventions for adolescent internet addiction [18,27]. Longitudinal findings have provided further, although mixed, evidence. A Swiss cohort study found bidirectional associations between video gaming disorder symptoms and sport or exercise participation over 15 months [22], and another longitudinal study reported that higher PA is associated with decreased IGD symptoms [28]. However, other studies have found that IGD symptoms predict later PA but PA does not predict later IGD symptoms [29] or that gaming time predicts PA trajectories but not vice versa [30]. Notably, most prior longitudinal studies have relied on the traditional cross-lagged panel model (CLPM), which cannot disentangle within-person fluctuations from stable between-person differences [31]. Consequently, the within-person longitudinal association between PA and IGD symptoms remains insufficiently understood.

Loneliness is a distressing emotional state arising from the perceived discrepancy between desired and actual social relationships [32]. Rather than reflecting objective social isolation, it represents a subjective sense of lacking meaningful connections with others [33]. According to the model of compensatory internet use [34], negative life situations may motivate individuals to go online to regulate negative emotions, and problematic online behaviors arise from attempts to compensate for adverse offline experiences [34]. Online gaming may temporarily relieve loneliness by providing social interaction, achievement, and escape from a negative affect [35]. However, when such compensatory use becomes habitual, it may develop into maladaptive patterns characteristic of increased IGD symptoms [36]. Over time, reliance on gaming as a coping strategy may reduce opportunities for meaningful offline interactions and further intensify loneliness [33]. Likewise, increased IGD symptoms may increase loneliness by reducing real-life social engagement, substituting online interactions for offline relationships, and fostering social withdrawal [37]. These mechanisms suggest a dynamic, bidirectional relationship between loneliness and IGD symptoms.

Empirical studies generally support this perspective. Cross-sectional studies across university and adolescent samples have consistently shown a positive association between loneliness and IGD symptoms [38-42]. Meta-analytic evidence has further identified loneliness as a risk factor of IGD symptoms and internet use disorder symptoms more broadly [43,44]. Longitudinal research has also provided support. Baseline loneliness significantly predicted later IGD symptoms among Chinese university students [45], loneliness predicted increases in gaming problems at both within-person and between-person levels [46], and a two-wave longitudinal study of Dutch adolescent gamers identified loneliness to be both an antecedent and a consequence of pathological gaming [47]. However, most longitudinal studies have used traditional structural equation modeling (SEM) or CLPM, both of which are limited in distinguishing within-person fluctuations from stable between-person differences [48]. As a result, the dynamic longitudinal association between loneliness and IGD symptoms remains insufficiently understood. Furthermore, much of the existing evidence focuses on general internet addiction rather than IGD symptoms specifically, and research among adolescent gamers remains relatively scarce.

The bidirectional association between PA and loneliness can be understood within the framework of basic psychological needs theory (BPNT), a subtheory of self-determination theory [49]. BPNT posits that the fulfillment of three innate psychological needs, namely autonomy, competence, and relatedness, is essential for psychological well-being [50,51]. In this context, PA may help satisfy the need for relatedness by providing opportunities for social interaction, belonging, and shared achievement [52]. Regular participation in exercise or group-based sports may thus alleviate loneliness through enhanced social connectedness and emotional support [53]. Conversely, persistent loneliness may undermine these needs by reducing social motivation and self-regulatory capacity, thereby decreasing engagement in PA over time [54]. This reciprocal process suggests that PA and loneliness may influence each other dynamically across time.

Empirical findings generally support this proposition. Cross-sectional research consistently reveals a negative association between PA and loneliness across diverse populations [55-57]. Systematic reviews support this claim [58,59]. Longitudinal evidence further reports the bidirectional association between PA and loneliness among adults and older adults [60,61]. A recent study using a random intercept cross-lagged panel model (RI-CLPM) further demonstrated small but significant within-person bidirectional effects between PA and loneliness [62]. However, most existing longitudinal studies have focused on adult or older populations, with adolescents receiving far less attention. Additionally, many studies have relied on traditional CLPMs. Moreover, relatively few studies have simultaneously examined PA, loneliness, and IGD symptoms within an integrated longitudinal framework. To address these limitations, this study used an RI-CLPM to clarify the bidirectional relationship between PA and loneliness among adolescents.

Emerging evidence suggests that sex may moderate the associations among PA, loneliness, and IGD symptoms. Epidemiological studies consistently show that male adolescents have a higher prevalence of IGD than females [63,64], and males with more severe IGD symptoms appear to be more sensitive to gaming-related rewards, whereas females tend to exhibit stronger affective dysregulation and mood-related symptoms [64]. Sexual differences are also evident in PA levels, with males generally reporting higher engagement in PA than females [65]. Notably, when PA levels are low, males report slightly higher internet addiction than females, but as PA increases, males show a sharper decline in addiction symptoms, suggesting that the protective role of PA may be particularly pronounced among males [65]. At the same time, loneliness and its correlates may operate differently across sexes. Prior studies have shown that the correlation between loneliness and internet addiction tends to be weaker among males [30]. Furthermore, longitudinal evidence shows that moderate-to-vigorous PA negatively predicts gaming time among males but not females [30]. Moreover, sexual differences have been observed in the link between PA and psychological distress [66], where the negative association between PA and loneliness is more pronounced among females [67]. These findings indicate that sex may moderate the associations among PA, loneliness, and IGD symptoms. However, few longitudinal studies have examined whether such sex-specific patterns remain stable over time in youth populations.

To address these gaps, this study adopted a three-wave longitudinal design and an RI-CLPM to examine the temporal associations between PA, loneliness, and IGD symptoms among Chinese adolescent gamers. By distinguishing stable between-person differences from within-person fluctuations, this approach clarified the dynamic reciprocal relationships among these variables over time. Four hypotheses were proposed:

This 1-year longitudinal study was conducted from March 2024 to March 2025, with data collected at three 6-month intervals: wave 1 (baseline [T1], March 2024), wave 2 (T2, September 2024), and wave 3 (T3, March 2025).

Ethical approval for the study was obtained from the Institutional Ethics Committee of Wuhan College (approval #20240306). Prior to each survey administration, students and their parents (or legal guardians) were fully informed about the purpose of the study, data collection procedures, and measures taken to guarantee confidentiality. Parents/guardians and students provided written informed consent. Participation was strictly voluntary, and both students and parents/guardians were assured that refusal or withdrawal would not have any negative consequences on the students’ academic records or school evaluations. To enable longitudinal matching of responses across waves, students were asked to provide their school IDs, which were encrypted, stored securely, and removed from the analytical dataset once the linkage was completed, thereby preserving anonymity. An information sheet describing participants’ rights was distributed prior to survey completion. Participants did not receive any financial compensation for participation. No identifiable images or personal information of participants are included in the manuscript or Multimedia Appendix 1.

A detailed overview of participant enrollment, retention, and attrition across each wave is provided in Figure 1.

As presented in Table 1, the two groups did not differ significantly with respect to self-reported academic performance, parental education, perceived family financial status, single-parent household status, weekly gaming time, PA, or loneliness. However, significant group differences were observed for sex (P=.03), age (P=.001), and IGD symptoms (P=.02).

Table 2 demonstrates the bivariate correlations. IGD symptoms were consistently and negatively associated with PA across all three waves, with correlation coefficients ranging from –0.26 to –0.19. Similarly, loneliness exhibited significant negative correlations with PA at each wave, with r values between –0.44 and –0.25. In contrast, IGD symptoms and loneliness were positively correlated over time, with coefficients ranging from 0.17 to 0.29.

Table 3 presents the results of longitudinal measurement invariance testing for PA, loneliness, and IGD symptoms. The configural invariance models showed acceptable fit, with the CFI and TLI exceeding 0.90 and the RMSEA and SRMR below 0.08, indicating that the basic factor structures of these constructs were stable across the three waves. Metric invariance was supported, as changes in fit indices (ΔCFI≤0.01; ΔRMSEA≤0.015) fell within recommended thresholds, suggesting that the factor loadings remained consistent over time. Scalar invariance was also established, with minimal declines in model fit (ΔCFI≤0.01; ΔRMSEA≤0.015), indicating equivalence of item intercepts across measurement occasions. In addition, measurement invariance across sexes was examined using the same sequential procedure (configural, metric, and scalar invariance). The results similarly supported measurement invariance between males and females, indicating that the constructs were comparable across sexes. Detailed results of measurement invariance tests across sexes are presented in Table S1 in Multimedia Appendix 1.

A fully unconstrained model showed acceptable fit to the data (χ²₈=35.12, CFI=0.985, TLI=0.974, RMSEA=0.043, SRMR=0.030). Subsequently, a model constraining autoregressive and cross-lagged paths to be equal across adjacent waves was estimated. The chi-square difference test indicated that imposing these equality constraints does not significantly worsen model fit (Δχ²₉=15.35, P=.08). Therefore, the more parsimonious constrained model was retained as the final model (χ²₁₇=50.47, CFI=0.989, TLI=0.977, RMSEA=0.042, SRMR=0.030). The results of model fit comparison are presented in Table S2 in Multimedia Appendix 1. The results of the RI-CLPM examining longitudinal associations among PA, loneliness, and IGD symptoms are summarized in Figure 2.

Autoregressive paths for each variable were significant across waves, indicating substantial temporal stability for PA, loneliness, and IGD symptoms. Cross-lagged effects revealed significant reciprocal associations among PA, loneliness, and IGD symptoms across the three measurement waves. Specifically, higher PA predicted decreased subsequent loneliness (T1→T2: β=–0.12, 95% CI −0.16 to −0.08, P<.001; T2→T3: β=–0.13, 95% CI −0.17 to −0.08, P<.001), whereas increased loneliness predicted lower PA at the following waves (T1→T2: β=–0.22, 95% CI −0.26 to −0.18, P<.001; T2→T3: β=–0.22, 95% CI −0.26 to −0.18, P<.001). Similarly, higher PA was associated with decreased subsequent IGD symptoms (T1→T2: β=–0.10, 95% CI −0.16 to −0.04, P=.009; T2→T3: β=–0.10, 95% CI −0.16 to −0.04, P=.009), while increased IGD symptoms predicted decreases in PA over time (T1→T2: β=–0.08, 95% CI −0.12 to −0.04, P=.02; T2→T3: β=–0.08, 95% CI −0.12 to −0.04, P=.02). In addition, increased loneliness predicted increases in subsequent IGD symptoms (T1→T2: β=0.14, 95% CI 0.11-0.17, P<.001; T2→T3: β=0.14, 95% CI 0.11-0.17, P<.001), and increased IGD symptoms, in turn, predicted increased loneliness at the next wave (T1→T2: β=0.18, 95% CI 0.14-0.22, P<.001; T2→T3: β=0.18, 95% CI 0.14-0.22, P<.001).

Within-person residual correlations revealed that PA was negatively associated with both loneliness and IGD symptoms at all three waves, with standardized coefficients ranging from –0.33 to –0.28 (all P<.001). In contrast, loneliness and IGD symptoms were positively correlated at the within-person level across waves, with coefficients ranging from 0.21 to 0.38 (all P<.001).

At the between-person level, latent trait correlations further supported these associations. The latent factor of PA was negatively associated with the latent traits of loneliness (r=–0.31, 95% CI –0.45 to –0.17, P<.001) and IGD symptoms (r=–0.33, 95% CI –0.49 to –0.17, P<.001), indicating that adolescents with higher overall levels of PA across the study period tended to report consistently decreased loneliness and IGD symptoms. Moreover, loneliness and IGD symptoms were strongly and positively associated at the between-person level (r=0.47, 95% CI 0.35-0.59, P<.001), suggesting a stable covariation between these two constructs over time.

The effects of time-invariant covariates on the RIs are reported in Table S3 in Multimedia Appendix 1. Sex was significantly associated with the RI of PA (β=−0.08, 95% CI −0.13 to −0.02, P=.005) and IGD symptoms (β=−0.08, 95% CI −0.15 to −0.01, P=.001), indicating that females (0=male, 1=female) reported slightly lower overall levels of PA and IGD symptoms across the study period. In contrast, age, self-reported academic performance, perceived family financial situation, and single-parent family status were not significantly associated with the RIs (all P>.05).

The time-varying effects of gaming time on PA, loneliness, and IGD symptoms are presented in Table S4 in Multimedia Appendix 1. Gaming time was negatively associated with PA at each wave (T1: β=−0.11, 95% CI −0.19 to −0.03, P<.001; T2: β=−0.09, 95% CI −0.15 to −0.03, P=.001; T3: β=−0.06, 95% CI −0.10 to −0.02, P=.04), suggesting that higher gaming engagement is concurrently associated with lower PA. In addition, gaming time was positively associated with loneliness at each wave (T1: β=0.07, 95% CI 0.01-0.13, P=.009; T2: β=0.09, 95% CI 0.03-0.15, P=.001; T3: β=0.06, 95% CI 0.02-0.10, P=.02), suggesting that higher gaming engagement is concurrently associated with increased loneliness. Moreover, gaming time was positively associated with IGD symptoms at each wave (T1: β=0.10, 95% CI 0.04-0.16, P<.001; T2: β=0.06, 95% CI 0.01-0.12, P=.01; T3: β=0.08, 95% CI 0.02-0.14, P=.004), suggesting that higher gaming engagement is concurrently associated with increased IGD symptom severity.

Multigroup RI-CLPM was conducted to examine whether the longitudinal associations among PA, loneliness, and IGD symptoms varied by sex. In the first step, an unconstrained model allowing all cross-lagged and autoregressive paths to vary freely across males and females was estimated and demonstrated acceptable model fit: χ²₈=21.46, CFI=0.971, TLI=0.954, RMSEA=0.038, SRMR=0.036. In the second step, a constrained model was tested in which all cross-lagged paths were set equal across groups. This model also exhibited good fit: χ²₁₈=60.24, CFI=0.942, TLI=0.915, RMSEA=0.077, SRMR=0.021. However, the chi-square difference test revealed a significant decline in model fit when the cross-lagged paths were constrained (Δχ²₁₀=38.78, P<.001), suggesting that the model results significantly differed by sex. Figures 3 and 4 present the results.

Specifically, sexual differences emerged in the path from PA to loneliness. A Wald test of parameter constraints indicated a significant difference between females and males (Wald test estimate=6.27, P=.01). For males, PA significantly predicted loneliness (T1→T2: β=–0.08, 95% CI −0.15 to −0.01, P=.03; T2→T3: β=–0.09, 95% CI −0.16 to −0.02, P=.01). For females, the cross-lagged effects from PA to loneliness were significant but stronger (T1→T2: β=–0.17, 95% CI −0.23 to −0.11, P<.001; T2→T3: β=–0.17, 95% CI −0.23 to −0.11, P<.001). Sexual differences were also found in the pathway from loneliness predicting IGD symptoms (Wald test estimate=5.87, P=.02). Among males, loneliness significantly predicted IGD symptoms (T1→T2: β=0.09, 95% CI 0.03-0.15, P=.02; T2→T3: β=0.09, 95% CI 0.03-0.15, P=.02). Among females, the paths from loneliness to later IGD symptoms were also significant (T1→T2: β=0.16, 95% CI 0.11-0.21, P<.001; T2→T3: β=0.16, 95% CI 0.11-0.21, P<.001).

For the remaining cross-lagged paths, no significant sexual differences were observed. Specifically, PA negatively predicted subsequent IGD symptoms for both males (T1→T2: β=−0.08, 95% CI −0.15 to −0.01, P=.02; T2→T3: β=−0.09, 95% CI −0.16 to −0.02, P=.01) and females (T1→T2: β=−0.11, 95% CI −0.17 to −0.05, P=.004; T2→T3: β=−0.11, 95% CI −0.17 to −0.05, P=.004). IGD symptoms also predicted decreases in PA across time among males (T1→T2: β=−0.08, 95% CI −0.14 to −0.02, P=.03; T2→T3: β=−0.09, 95% CI −0.15 to −0.03, P=.02) and females (T1→T2: β=−0.09, 95% CI −0.15 to −0.03, P=.02; T2→T3: β=−0.08, 95% CI −0.14 to −0.02, P=.03). In addition, IGD symptoms predicted subsequent increases in loneliness among both males (T1→T2: β=0.16, 95% CI 0.11-0.21, P<.001; T2→T3: β=0.17, 95% CI 0.12-0.22, P<.001) and females (T1→T2: β=0.18, 95% CI 0.13-0.23, P<.001; T2→T3: β=0.19, 95% CI 0.14-0.24, P<.001). Likewise, loneliness predicted later decreases in PA for males (T1→T2: β=−0.20, 95% CI −0.25 to −0.15, P<.001; T2→T3: β=−0.21, 95% CI −0.26 to −0.16, P<.001) and females (T1→T2: β=−0.22, 95% CI −0.27 to −0.17, P<.001; T2→T3: β=−0.23, 95% CI −0.28 to −0.18, P<.001).

To ensure that the main findings were not influenced by missing data or analytic choices, additional analyses were carried out. A complete case analysis (listwise deletion approach) was applied, including only participants who participated in all three assessment waves (n=1138, 85.4%). Although this reduced the sample size, the pattern and size of the cross-lagged paths among PA, loneliness, IGD symptoms were almost identical to those from the FIML model. The similar pattern also appeared in the sex-specific models, indicating that the main conclusions were not sensitive to data exclusion. Figures S1-S3 in Multimedia Appendix 2 present these results. This procedure follows current recommendations for longitudinal research and aligns with previous applications of RI-CLPMs in developmental studies [75,76].

This study used a three-wave RI-CLPM to investigate the within-person associations between PA, loneliness, and IGD symptoms among Chinese adolescent gamers. The results revealed bidirectional and longitudinal relationships between PA, loneliness, and IGD symptoms. Multigroup RI-CLPM analyses indicated that the effect of PA on loneliness is stronger among females and that the association between loneliness and subsequent IGD symptoms is also stronger for females. These findings highlight the dynamic nature of these associations, emphasizing the importance of considering temporal and sex-specific patterns in understanding fluctuations in these behaviors over time.

This study has several limitations. First, variables, especially PA, were assessed using self-report questionnaires, which may be subject to response biases, such as social desirability or recall bias. Future studies could incorporate multi-informant reports or objective measures (eg, accelerometers) to improve accuracy. Second, although the three-wave design enabled examination of temporal dynamics, the short intervals and limited number of waves restrict the ability to capture longer-term developmental processes. More frequent assessments across longer periods are needed in future studies. Third, although cross-lagged effects were statistically significant and fell within the medium-to-large range [78], their magnitudes should still be interpreted with caution in the light of the large sample size and the reliance on self-reported measures. Fourth, although missing data analyses suggested that the missing-at-random assumption was reasonable and FIML estimation was applied, differential attrition related to sex, age, and IGD symptoms may still introduce potential bias. Fifth, the sample consisted of Chinese adolescent gamers from urban schools, which may limit generalizability to other cultural or contextual settings, particularly rural populations. Future research should consider including more diverse samples and examining potential urban-rural differences. Finally, the study did not collect data on the types of games played (eg, competitive, social, or cooperative), which could have provided additional insights into how different gaming experiences influence the observed outcomes.

This study used a three-wave longitudinal design to examine the within-person dynamics between PA, loneliness, and IGD symptoms among Chinese adolescent gamers. The results revealed bidirectional relationships between these variables, reflecting a mutually reinforcing cycle. Importantly, sexual differences were observed in the strength of these associations, with the effects of PA on loneliness and loneliness on IGD symptoms being stronger for females. By using an RI-CLPM to distinguish within-person changes from stable between-person differences, this study extends prior research based mainly on cross-sectional designs or traditional CLPMs and provides a clearer understanding of how behavioral and emotional factors jointly shape adolescent problematic gaming. The results suggest that interventions should be sex sensitive, addressing both PA and loneliness simultaneously to effectively mitigate the risk of IGD symptoms, particularly among females.