The study data were obtained from the Chengdu Positive Child Development project, which conducted a three-wave study among Chinese children and adolescents during the COVID-19 lockdown period (37). Chengdu was one of the worst-affected cities in China during COVID-19. Moreover, as the capital of Sichuan province in western central China, Chengdu was strongly impacted by earthquakes of different magnitude levels that occurred in China. All participants in this three-wave study experienced the most challenging and severe phase of the COVID-19 pandemic in Chengdu. While the pandemic was the primary factor, the participants may have also been secondarily affected by earthquakes of different intensities.

A random cluster sampling method was used to select five schools in Chengdu to recruit the project participants, including three high schools and two schools that provide integrated primary and middle education under unified management. A self-administered questionnaire was used to collect data. Three waves of data were collected from 2020 to 2022 with a one-year interval, including baseline (Time 1, T1: June–July 2020), first follow-up (Time 2, T2: June 2021), and second follow-up (Time 3, T3: June 2022). A total of 7–393 participants were enrolled at T1. Of these, 5–913 remained at T2 and 2–354 completed the study at T3, reflecting attrition rates of 20% from T1 to T2 and 60% from T2 to T3. Inconsistent regional lockdowns during the COVID-19 outbreak posed logistical challenges for data collection and contributed to increased participant attrition across time points.

Attrition analyses revealed that participants who dropped out were significantly more likely to be older, exhibited higher baseline IA levels, and reported lower LS (ps <.001). No significant differences were observed in sex (p = .076) or PTSS (p = .828) between those who completed all three waves and those lost to follow-up. The final sample for the current study comprised 2–354 participants who completed all three waves (50.3% females, SD = 1.77). Participants’ ages at T1 ranged from 6 years to 14 years (Mage = 9.62).

Both university and administrative departments at the data collection schools were approved by the Research Ethics Committee to conduct the research (approval no. K2020025 and KS2022913). The study’s purpose, process, participants’ rights, privacy protection, and data-retention policy were introduced to all voluntary participants. Consent was provided by participants, and the legal guardians permitted their participation in the study.

PTSS were measured using the Chinese version of the 13-item Child Revised Impact of Events Scale (38). Participants rated each item on a six-point frequency scale (0 = not at all; 5 = often). The total scores ranged from 0 to 65, with a higher score representing a higher level of PTSD symptoms. The scale has been shown to have good validity and reliability in previous studies (39). The scale showed good internal consistency in the current study. Cronbach’s alpha coefficients for PTSS at T1, T2, and T3 were 0.89, 0.87, and 0.87, respectively.

IA was measured using the Chinese version of the 20-item Young’s Internet Addiction Test (40). A five-point Likert scale was used to rate each item (1 = does not apply; 5 = always), with total scores ranging from 20 to 100. A higher total score indicates a more severe IA level. A prior study demonstrated the test’s great validity and reliability (41). The scale showed excellent internal consistency in the current study. Cronbach’s alpha coefficients for IA at T1, T2, and T3 were 0.93, 0.94, and 0.95, respectively.

LS was measured using the Chinese version of the five-item Satisfaction with Life Scale (20). Each item is rated on a six-point scale from 1 (strongly disagree) to 6 (strongly agree), with the overall score lying between 5 and 30. A higher total score represents a higher level of LS. The robust validity and reliability of this scale was demonstrated by a previous study (42). In this study, the scale showed an acceptable internal consistency. Cronbach’s alpha coefficients for LS at T1, T2, and T3 were 0.77, 0.83, and 0.86, respectively.

First, a descriptive analysis of all the study variables was conducted and a correlation matrix constructed using SPSS (Version 24). Second, a longitudinal mediation approach was applied using structural equation modeling (SEM) to examine the relationships among PTSS, IA, and LS across the three waves. This method allowed all three variables to be included in the model at each time point and enabled the analysis of all possible mediation pathways (43). It provides a robust framework for assessing whether the effect of a predictor on an outcome is mediated by variables across different time points while accounting for potential confounding factors (43). The study variables (i.e., PTSS, IA, and LS across the three waves) were modeled as latent variables, with subscales for each item as indicators. Age and sex were used as covariates that might be linked with the study variables. Furthermore, tests for common-method bias, temporal differences, measurement equivalence, and univariate normality were conducted across the three waves. AMOS (Version 23) was employed to assess the significance of mediation effects using 5–000 bootstrapped replications. Maximum likelihood estimation was applied to assess the parameters. Some suggested indexes and their standards were used to estimate the goodness-of-fit and quality-of-mediation model (44). An acceptable model required that the ratio of χ² to degree of freedom should be smaller than 5.0. The Tucker-Lewis (TLI), normed fit (NFI), and comparative fit (CFI) indices should be higher than 0.90. The standard root mean square residual (SRMR) and root mean square error of approximation (RMSEA) should be lower than 0.08 and 0.05, respectively (44). As all participants consented to completing all the items, and no data were missing across all three waves.

The descriptive-analysis results, including means, standard deviations, and intercorrelations among the study variables, are presented in Table 1 . Significant correlations were observed among IA, LS, and PTSS across the three study waves; however, T3 PTSS did not significantly correlate with T1 IA (r = .035, p = .086), T1 LS (r = –.023, p = .275), and T2 LS (r = –.023, p = .262). All study variables were significantly intercorrelated after controlling for sex and age, excluding a nonsignificant intercorrelation between T1 LS and T3 PTSS (r = –.038, p = .066).

Univariate normality was assessed by examining skewness and kurtosis values for all the study variables across the three waves (44). For LS, skewness ranged from –0.930 to –0.822 and kurtosis from 0.117 to 0.423. For IA, skewness ranged from 1.384 to 1.668 and kurtosis from 1.574 to 2.943. For PTSS, skewness ranged from 1.239 to 1.576 and kurtosis from 1.373 to 2.884. All values fell within the commonly accepted thresholds for approximate normality (i.e., |skewness| < 2 and |kurtosis| < 7), indicating that the variables exhibited acceptable levels of univariate normality (44).

To assess measurement equivalence across the three waves, a series of increasingly restrictive confirmatory-factor-analysis (CFA) models were conducted using AMOS (44). Criteria for establishing measurement invariance included the ΔCFI ≤ 0.01 and ΔRMSEA ≤ 0.015 (44). The configural invariance model (Model 1), which allowed all parameters to vary across time, demonstrated an acceptable fit (χ²/df = 2.935, CFI = 0.977, RMSEA = 0.029), indicating that the factor structure was consistent across waves. The metric invariance model (Model 2), which constrained factor loadings to be equal over time, also fit the data well (ΔCFI = 0.004, ΔRMSEA = 0.002), suggesting loading invariance. The scalar invariance model (Model 3), which added equality constraints on item intercepts, resulted in minimal change in fit relative to the metric model (ΔCFI = 0.004, ΔRMSEA = 0.001), supporting scalar invariance. Finally, the residual invariance model (Model 4), which further constrained item residual variances, showed a small but acceptable decline in fit (ΔCFI = 0.012, ΔRMSEA = 0.005), indicating partial support for strict invariance. Overall, these findings provide evidence for longitudinal measurement invariance, supporting meaningful comparisons of the latent constructs across waves.

Harman’s one-factor CFA model showed poor fit to the data (CFI = 0.322, NFI = 0.319, TLI = 0.285, SRMR = 0.141, RMSEA = 0.150), indicating that common-method bias was unlikely to be a serious concern in this study. Moreover, repeated-measures analysis of variance was used to test of temporal differences in the study variables. PTSS showed a highly significant change over time (F = 45.90, p <.001), indicating substantial variation across the three waves. IA also demonstrated a significant time effect (F = 18.72, p <.001), although the magnitude of change was smaller than that of PTSS. LS showed a modest but still statistically significant change over time (F = 3.63, p <.05). Overall, these results support the use of a longitudinal mediation model for further analysis using SEM.

A complete longitudinal mediation model was estimated to evaluate the links among PTSS, IA, and LS across the three waves by controlling for sex and age as covariates (see Figure 1 ). The model demonstrated good fit (χ²/df = 4.629, CFI = 0.952, NFI = 0.940, TLI = 0.947, SRMR = 0.058, RMSEA (90% CI) = 0.039 (0.038–0.041)).

Results from the complete longitudinal mediation model revealed several significant pathways among the study variables across the three waves. Specifically, T1 IA positively predicted T2 IA (β = 0.492, SE = 0.031, p <.001) and negatively predicted T2 LS (β = –0.103, SE = 0.011, p <.001). T1 LS negatively predicted T2 IA (β = –0.142, SE = 0.146, p <.001) and positively predicted both T2 LS (β = 0.450, SE = 0.091, p <.001) and T2 PTSS (β = –0.081, SE = 0.214, p <.005). T1 PTSS positively predicted T2 PTSS (β = 0.285, SE = 0.022, p <.001). At the second wave, T2 IA positively predicted T3 IA (β = 0.205, SE = 0.024, p <.001), negatively predicted T3 LS (β = –0.087, SE = 0.009, p <.001), and positively predicted T3 PTSS (β = 0.083, SE = 0.016, p <.001). T2 LS negatively predicted T3 IA (β = –0.064, SE = 0.065, p <.001) and both positively predicted T3 LS (β = 0.213, SE = 0.026, p <.001) and negatively predicted T3 PTSS (β = –0.081, SE = 0.043, p <.001). T2 PTSS positively predicted T3 PTSS (β = 0.058, SE = 0.011, p <.005). Furthermore, results indicated some nonsignificant paths, including T1 IA to T2 PTSS (β = 0.043, SE = 0.046, p >.05), T1 PTSS to T2 IA (β = -0.016, SE = 0.013, p >.05) and T2 LS (β = -0.005, SE = 0.005, p >.05), and T2 PTSS to T3 IA (β = 0.020, SE = 0.017, p >.05) and T3 LS (β = -0.039, SE = 0.007, p >.05).

To enhance model conciseness, all nonsignificant paths were removed from the complete longitudinal mediation model, resulting in the clipped model (see Figure 2 ). The clipped model demonstrated a good goodness-of-fit (χ²/df = 4.587, CFI = 0.952, NFI = 0.939, TLI = 0.947, SRMR = 0.058, RMSEA (90% CI) = 0.039 (0.038–0.040)). The results revealed several significant relationships among IA, LS, and PTSS at adjacent time points. T1 IA significantly predicted T2 IA (β = 0.483, SE = 0.030, p <.001) and negatively predicted T2 LS (β = –0.10, SE = .010, p <.001). T1 LS negatively predicted T2 IA (β = –0.143, SE = 0.144, p <.001), positively predicted T2 LS (β = 0.453, SE = 0.092, p <.001), and negatively predicted T2 PTSS (β = -0.098, SE = 0.196, p <.001). T1 PTSS significantly predicted T2 PTSS (β = 0.296, SE = .021, p <.001). At the next time point, T2 IA significantly predicted T3 IA (β = 0.211, SE = .023, p <.001), negatively predicted T3 LLS (β = –0.094, SE = 0.009, p <.001), and positively predicted T3 PTSS (β = 0.081, SE = 0.015, p <.001). T2 LS negatively predicted T3 IA (β = –0.061, SE = 0.065, p = .013), positively predicted T3 LS (β = 0.218, SE = 0.026, p <.001), and negatively predicted T3 PTSS (β = –0.08, SE = 0.043, p <.001). T2 PTSS significantly predicted T3 PTSS (β = 0.045, SE = 0.009, p = .011). Overall, the results of the complete longitudinal mediation ( Figure 1 ) and clipped ( Figure 2 ) models revealed largely consistent findings regarding the relationships among the study variables across the three waves. Both models supported the same overall pattern of associations and underlying structure.

The results showed that T2 LS mediated the link between T1 IA and T3 PTSS [indirect effect (95% CI) = 0.047 (0.027–0.069)] and T2 IA mediated the relationship between T1 LS and T3 PTSS [indirect effect (95% CI) = –0.052 (–0.074 to –0.035)]. T2 LS mediated the association between T1 IA and T3 IA [indirect effect (95% CI) = 0.108 (0.080–0.141)]. T2 IA mediated the relationship between T1 LS and T3 LS [indirect effect (95% CI) = 0.112 (0.085–0.139)].

The results revealed consistent and significant bidirectional relationships among IA, LS, and PTSS across the three waves in the longitudinal mediation model. Specifically, at T1, LS was negatively correlated with both IA (r = –.445, p <.001) and PTSS (r = –.228, p <.001), while IA was positively correlated with PTSS (r = .272, p <.001). At T2, LS remained negatively correlated with IA (r = –.326, p <.001) and PTSS (r = –.133, p <.001), and IA was positively correlated with PTSS (r = .224, p <.001). Similarly, at T3, LS was negatively correlated with IA (r = –.349, p <.001) and PTSS (r = –.329, p <.001) while IA was positively correlated with PTSS (r = .321, p <.001). These findings demonstrate stable and reciprocal associations among the three study variables over time.

The results revealed that LS at the one-year follow-up significantly mediated the relationship between baseline IA and PTSS at the two-year follow-up. This finding supports the hypothesis and aligns with previous applications of self-determination theory (29, 30) and CBM (32, 33) in the context of IA. Specifically, higher IA levels in the early stages of a disaster may lead to reduced LS, which in turn increases the long-term risk of PTSS among Chinese children and early adolescents exposed to the disaster. This is consistent with prior research indicating an upward trend in IA among youth following traumatic events (4). Individuals with IA often experience discomfort and negative life changes, which may reduce their LS after a disaster (18). Lower LS can contribute to dissatisfaction with real life, deficits in quality of life, impaired social relationships, and disrupted life habits, ultimately heightening vulnerability to psychological problems (17). This result may also be explained by previous research showing that a higher degree of IA can impair individuals’ perceptions of LS derived from real-life social networks (45–47), thereby increasing the likelihood of long-term mental-health issues such as PTSS (48–50).

Interestingly, the findings revealed that the IA level at the one-year follow-up significantly mediated the relationship between baseline LS and PTSS at the two-year follow-up. This suggests that lower LS at the onset of a disaster might contribute to increased IA over time, which in turn heightens the risk of long-term psychological problems such as PTSS. Previous research has shown that disaster events such as the COVID-19 pandemic often disrupt daily life through factors such as social isolation, repeated testing, and regional lockdowns, all of which contribute to reduced LS (51–53). Furthermore, in the aftermath of a disaster, diminished LS may impair emotional regulation and increase vulnerability to stress, making individuals more likely to turn to the Internet as a form of avoidant coping (54). People with growing IA often rely on excessive online engagement to escape or numb distressing emotions in response to mounting real-life stressors (55–57). However, such avoidance hinders the emotional processing of trauma. When trauma remains unprocessed, it can become more deeply entrenched, leading to more severe PTSS (e.g., emotional numbing, intrusive thoughts, and hyperarousal) (58) following disasters. Additionally, although IA is frequently conceptualized as a mental-health outcome, our findings indicate that it could also function as a risk factor and mediator in the development of PTSS. This dual role aligns with prior evidence supporting the bidirectional relationships between IA and other mental-health problems (59, 60). Future research is suggested to further examine the underlying pathways through which IA longitudinally shapes psychological well-being in this dual role.

Beyond the identified mediation pathways, the results revealed significant bidirectional associations between IA and PTSS across the three waves. These findings underscore the persistent, reciprocal relationship between IA and PTSS over time and are consistent with existing evidence highlighting strong intersections and potential comorbidity between the two [e.g (4, 14, 15).,]. Notably, no significant directed longitudinal effects were found from baseline PTSS to later LS or IA. One plausible explanation is that, within the framework of a two-year follow-up period, early PTSS may not exert a lasting impact sufficiently strong to influence long-term trajectories of subjective well-being or behavioral coping strategies, such as patterns of Internet use. Furthermore, children and early adolescents may possess greater psychological adaptability and benefit from external regulatory support (e.g., parental guidance), particularly during extended periods of home confinement and increased family time during the COVID-19 pandemic. Such support may buffer the enduring effects of trauma on daily functioning and LS (52). Additionally, both LS and Internet-use behaviors are shaped by a constellation of contextual factors, including peer relationships, academic pressures, and school dynamics, which may weaken the influence of PTSS over time (61–63). Collectively, these findings suggest that the longitudinal influence of PTSS on LS and IA may be governed by more complex and indirect mechanisms. Future research would benefit from examining potential mediators and moderators, such as personality traits, coping strategies, and family-relationship quality to further clarify these pathways (52, 64).

Based on the large sample of disaster-exposed children and young adolescents and established measures, this three-wave study explored the associations among IA, LS, and PTSS from the perspective of a longitudinal mediation model. The different roles of LS in the mediating relationships indicated that IA and PTSS were closely linked but had independent patterns. These findings provide further evidence for the longitudinal mediated association mechanism between IA, LS, and PTSS. Notwithstanding these strengths, several limitations in the current study should be noted. First, the self-report questionnaire and assessment scales were used to evaluate the variables; this may lead to biased estimation because of social desirability, recall partiality, misinterpretation, or lack of insight (65). Future research may employ clinical diagnostic interviews (e.g., CAPS-5) or multiple-informant reports to improve measurement accuracy (66). Second, this study used a three-wave longitudinal design, which clarified temporal order and allowed tests of longitudinal mediation among IA, LS, and PTSS. The findings remain longitudinally correlational and predictive. However, since the longitudinal observational study was limited in the experimental conditions, such as manipulation and randomization, future experimental or intervention studies are needed to provide more conclusive causal evidence (67). Finally, these findings may not be generalizable to other disaster-exposed populations or to individuals who have experienced different types of trauma. Future research is recommended to investigate the causal relationships among the study variables across diverse populations and trauma exposures.

The present findings emphasize the protective role of LS in buffering the long-term negative influence of IA and PTSS, the potentially adverse effect of inchoate IA symptoms on LS, and their impact on long-term psychological health among children and young adolescents exposed to disasters. These results highlight the need for developing psychological intervention programs that strengthen positive life forces, enhance social support, and promote adaptive coping strategies in disaster-affected youth (68, 69). Customized interventions focusing on LS improvement may increase resilience and reduce excessive reliance on the virtual world following traumatic experiences. For example, mindfulness-based interventions (70), solution-focused brief therapy (71), and positive psychology interventions (72) have shown promise in boosting positive emotions, meaning, and LS in the aftermath of disasters. Policymakers and practitioners should consider incorporating these strategies into pre- or post-disaster support systems for vulnerable children and young people.

This exploratory study revealed that life satisfaction plays a vitally protective role by mediating the influence of early Internet addiction on long-term psychological outcomes. Moreover, Internet addiction mediated the links between earlier life satisfaction and later PTSS, indicating that both enhancing life satisfaction and reducing problematic Internet use are critical for mitigating mental symptoms among disaster-exposed youth. Future research should investigate these pathways by using longer-term dynamic tracking and multi-method approaches to elucidate causal mechanisms and aid the development of targeted interventions.