Digital stress and problematic social media use among college students: exploring dual emotional pathways and the moderating role of cognitive reappraisal Authors: Yangyang Zhan, Lan Luo, Xinna Hu Published: 2025 ABSTRACT Digital stress has emerged as a significant form of psychological strain in online environments and is closely associated with problematic social media use (PSMU). However, the emotional and cognitive processes underlying this relationship remain unclear. Guided by the stress-coping framework and the Interaction of Person–Affect–Cognition–Execution (I-PACE) model, the present study examined whether anxiety and depression mediate the association between digital stress and PSMU, and whether cognitive reappraisal moderates these indirect effects. A total of 2,103 Chinese university students (1,035 males; mean age = 20.37 years, SD = 1.26) completed self-report measures of digital stress, anxiety, depression, cognitive reappraisal, and PSMU. Structural equation modeling and latent moderated structural equation analyses were conducted to test the proposed model. Results indicated that higher levels of digital stress were associated with greater PSMU, both directly and indirectly through anxiety and depression. Moreover, cognitive reappraisal attenuated the strength of these associations, suggesting a moderating role in the relationship between digital stress and emotional responses. These findings support a dual emotional pathway framework, indicating that digital stress may be linked to problematic social media engagement through high-arousal (anxiety-related) and low-arousal (depression-related) emotional processes. The results provide empirical support for extending emotion regulation models to digital contexts and underscore the importance of strengthening cognitive reappraisal skills to promote healthier social media engagement among university students. Introduction With social media becoming deeply integrated into everyday life, individuals have grown increasingly dependent on these platforms. As of June 2024, more than 4.59 billion people were active users of social networking platforms worldwide [1]. Compared to face-to-face social interaction, social media—characterized by its unique spatiotemporal flexibility and anonymity—has attracted a substantial proportion of the youth population [2]. Although social media offers various conveniences, such as facilitating information access and interpersonal communication, it has also been associated with several adverse consequences. Problematic social media use (PSMU) is defined as a behavioral pattern marked by diminished control over social media engagement, leading to excessive and prolonged use that results in adverse physiological and psychological outcomes [3]. A meta-analytic review has demonstrated a significant negative association between PSMU and adolescents’ physical and mental well-being [4]. In recent years, the proportion of young users on social media platforms has continued to increase, making PSMU a critical issue threatening the mental health of college students [5]. Therefore, an in-depth investigation into the underlying mechanisms of PSMU among college students is warranted. Existing studies have predominantly examined the antecedents of PSMU from a real-world perspective. For instance, Leijse et al. identified parental supervision, emotional instability, poor social competence, and peer pressure as significant risk factors for PSMU among college students [6]. Similarly, Kennerd et al. reported that students experiencing suicidal ideation tend to exhibit higher levels of PSMU [7]. However, the mechanisms through which unique stressors inherent in the digital ecology—such as information overload and digital stress—drive PSMU remain theoretically underexplored. The relationship between digital stress and problematic social media use Digital stress, a novel form of psychological strain within the digital environment, is defined as a sustained sense of psychological tension experienced by individuals during social media use. This tension arises from factors such as information overload, social comparison, fear of missing out (FoMO), and social-normative pressures [8]. Unlike traditional stressors, digital stress is characterized by its continuity, contextual specificity, and complexity; various social stimuli often occur simultaneously and jointly influence individuals’ emotional and cognitive responses [9]. Previous studies have shown that although these subdimensions exert distinct psychological effects, they tend to exhibit substantial convergence and shared variance in their statistical structure [10]. Consequently, recent research has conceptualized digital stress as an integrated latent construct reflecting individuals’ overall perceived stress within social media environments [11, 12]. This integrative approach not only facilitates the identification of its global psychological effects but also helps overcome the limitations associated with single-dimensional explanations. Among college students, a high level of online engagement in both social and academic activities increases their susceptibility to this multifaceted form of stress arising from complex online interactions. Empirical evidence suggests that under intense digital stress, individuals often seek immediate emotional relief or social support through social media, thereby developing a stress-coping-oriented usage pattern [13]. However, when such compensatory use becomes frequent and persistent, it can gradually deplete self-regulatory resources and impair cognitive control, increasing the risk of PSMU [14]. From a theoretical perspective, the Interaction of Person–Affect–Cognition–Execution (I-PACE) model [15, 16] proposes that external stressors act as situational inputs that activate individuals’ emotional and cognitive systems, thereby influencing media-use behaviors through motivational and reinforcement mechanisms. Among college students, digital stress often triggers immediate anxiety and tension, prompting continuous browsing of social media feeds to alleviate discomfort and seek reassurance—forming a typical “stress–response–reinforcement” cycle [17]. Furthermore, individuals who lack effective cognitive reappraisal skills tend to accumulate negative emotions, which subsequently impair self-regulatory functioning. Furthermore, a growing body of research has identified negative affect as a critical psychological pathway linking digital stress to PSMU [18]. Digital stress not only imposes cognitive load but is also frequently accompanied by emotional responses such as anxiety and depression, which drive individuals to seek temporary psychological relief through social media [19–22]. Accordingly, the relationship between digital stress and PSMU can be conceptualized as a continuous psychological process involving “stress–emotion–use.” Based on this, the present study proposes Hypothesis 1: Digital stress is significantly and positively associated with PSMU among college students. The mediating role of negative emotions As a chronic psychosocial stressor within the digital environment, digital stress not only directly influences individuals’ behavioral patterns but also indirectly contributes to PSMU by eliciting specific emotional responses. According to stress and coping theory [17], individuals experiencing sustained stress tend to adopt emotion-focused coping strategies to restore psychological equilibrium. Given its instantaneous interaction and affective feedback mechanisms, social media is often perceived as a convenient medium for emotional regulation [2]. However, excessive reliance on social media for emotional regulation may foster an “externalized regulation” pattern, which undermines self-control and reinforces a maladaptive cycle of “stress–emotional dysregulation–overuse” [19]. This process suggests that the impact of digital stress on PSMU is largely mediated by the emotional mechanisms it activates. Previous research has demonstrated that attentional biases toward negative emotional content and social anxiety significantly contribute to PSMU [23]. Among various negative emotions, PSMU shows strong positive correlations with anxiety and depressive symptoms, whereas its association with other affective states, such as loneliness, is less pronounced [24]. These findings indicate that anxiety and depression are the most salient psychological responses to digital stress, underscoring their central role in understanding how stress leads to maladaptive social media behaviors. According to the Tripartite Model [25], although both share the common component of negative affectivity, they differ significantly in physiological arousal and motivational orientation: anxiety is characterized by heightened arousal and vigilance, whereas depression manifests as low arousal and motivational deficits. This distinction is particularly significant within the context of the digital environment. Information overload, social comparison, and fear of missing out (FoMO) tend to increase individuals’ sensitivity to external threats and social evaluation, thereby eliciting anxiety [26]. Conversely, persistent online exclusion, feedback deprivation, and social fatigue are more likely to contribute to emotional exhaustion and depressive experiences [27]. Thus, the influence of digital stress on the emotional system may operate through two prototypical response patterns: a high-arousal tension response typified by anxiety and a low-arousal exhaustion response typified by depression. From the perspective of the high-arousal pathway, anxiety is typically accompanied by heightened vigilance and anticipatory worry toward perceived threat cues [28]. When individuals perceive potential negative evaluations or information omissions in social media environments, their anxiety levels increase, and attentional resources become disproportionately allocated to social cues [29]. To alleviate this persistent psychological tension, individuals frequently refresh social media feeds or check likes and comments to obtain immediate reassurance. Although such behavior may provide temporary relief, from a reinforcement-learning perspective, it reinforces excessive use through immediate rewards, ultimately fostering maladaptive dependence [16]. Accordingly, it can be inferred that anxiety mediates the relationship between digital stress and PSMU (Hypothesis 2). The low-arousal pathway, in contrast, is primarily manifested through emotional exhaustion and avoidance behaviors commonly associated with depression. Persistent digital stress can gradually deplete psychological resources, leading to feelings of helplessness and loss of control. These feelings, in turn, encourage immersive media use as a means of escaping real-world stressors [30]. This process aligns with the Compensatory Internet Use Theory [31], which proposes an “emotional compensation” mechanism whereby individuals engage in online activities as a short-term strategy for affective repair. However, the reduced motivation and diminished self-efficacy characteristic of depression [32] further impair self-monitoring abilities, resulting in uncontrolled social media use and worsening depressive symptoms [21]. Therefore, depression may serve as an indirect mediating pathway linking digital stress to PSMU (Hypothesis 3). Although anxiety and depression differ in their emotional characteristics, they are not mutually exclusive. The Emotional Cascade Model [33] proposes that anxiety tends to elicit frequent emotional activation and cognitive rumination, thereby provoking impulsive behaviors, whereas depression manifests as emotional blunting and behavioral inhibition, resulting in avoidant media use and social withdrawal. The former represents an over-engagement mechanism driven by emotional hyperactivation, while the latter reflects an avoidance–dependence mechanism stemming from emotional exhaustion. Within digital contexts, these two psychological processes may coexist and intertwine [34, 35]. Given that both anxiety and depression theoretically represent core emotional mechanisms underlying digital stress, the present study employs a parallel mediation model to illustrate two affective pathways—the “high-arousal over-engagement” and the “low-arousal avoidance–dependence” routes—in the association between digital stress and PSMU. This design reflects an integrative understanding of the dual-pathway model of emotional processing and provides a comprehensive theoretical framework for elucidating the psychological mechanisms through which digital stress shapes individuals’ media-use behaviors. The moderating role of cognitive reappraisal Although digital stress may indirectly contribute to PSMU through the elicitation of negative emotions, the strength of this mediating pathway varies among individuals. Prior research has demonstrated that individual differences in emotion regulation strategies significantly influence the transformation of stress into emotional responses [36]. Among various emotion regulation strategies, cognitive reappraisal is considered one of the most adaptive. It involves reinterpreting the meaning of an emotional event or adjusting cognitive appraisals to modify the quality or intensity of emotional experiences [37, 38]. This process typically occurs at the early stage of emotion generation. Within digital environments, the prevalence of information overload, social comparison, and evaluative threat makes individuals particularly susceptible to “situationally amplified” stress reactions [39]. Individuals with stronger cognitive reappraisal capacity can redefine the meaning of digital stimuli during the early stage of information processing—for example, perceiving others’ posts as references rather than competition, or understanding information flow as selective input rather than an uncontrollable burden. Through this process, cognitive reappraisal effectively mitigates threat perception and reduces the likelihood of experiencing negative emotions such as anxiety and depression [40]. From the Process Model of Emotion Regulation perspective [38], cognitive reappraisal is an antecedent-focused regulation strategy that reshapes the interpretation of situational meaning before emotional activation occurs, thereby functioning as a “cognitive gate.” This mechanism not only inhibits the transformation of stress into negative emotions but also indirectly reduces the likelihood of emotion-driven compensatory social media use. For example, a daily diary study demonstrated that individuals with higher cognitive reappraisal ability exhibited significantly lower daily emotional exhaustion and psychological maladjustment when experiencing FoMO-related negative affect [10]. Other empirical studies have similarly shown that cognitive reappraisal directly buffers the detrimental impact of digital stress on mental health and, by attenuating negative emotional reactions, decreases the likelihood of compensatory or impulsive social media engagement [41, 42]. Accordingly, this study proposes that cognitive reappraisal may exert a “gating effect” within the process through which digital stress influences PSMU. Individuals with higher cognitive reappraisal ability can reconstruct cognitive representations of digital stressors, thereby reducing the intensity and persistence of negative emotional responses and disrupting the vicious cycle of emotion-driven compensatory social media use. This moderating function is expected to operate primarily in the first half of the mediating pathway, given that the core function of cognitive reappraisal occurs during the early phase of emotion generation in stress appraisal [43]. Therefore, Hypothesis 4 (H4) posits that cognitive reappraisal moderates the first half of the mediating pathway through which digital stress influences PSMU via negative emotions. The current study Drawing on the preceding theoretical and empirical evidence, this study aims to construct and empirically test a psychological model centered on digital stress, examining its association with college students’ PSMU, as well as the indirect roles of anxiety and depression and the moderating effect of cognitive reappraisal. Grounded in stress-coping theory and the I-PACE model, digital stress is conceptualized as an external stressor closely linked to individuals’ emotional experiences and media use behaviors. Anxiety and depression represent high-arousal and low-arousal emotional response types, respectively, and may form distinct indirect pathways between digital stress and PSMU, thereby reflecting the potential psychological processes underlying the stress–emotion–use mechanism. According to the Process Model of Emotion Regulation, cognitive reappraisal regulates the intensity of emotional responses to digital stress by reinterpreting the meaning of stressful experiences. Individuals with higher cognitive reappraisal ability are likely to experience reduced levels of anxiety and depression when confronted with digital stress, thereby attenuating the association between digital stress and PSMU. Accordingly, an integrated moderated mediation model is proposed to examine the statistical relationships among digital stress, negative emotions, and PSMU, as well as potential individual differences in these associations. This study is designed to provide a comprehensive understanding of the psychological mechanisms connecting college students’ digital stress and PSMU, thereby offering empirical evidence for future longitudinal or intervention research and practical implications for emotion regulation and healthy media engagement. Participants and methods Procedure and participants The present study was conducted in accordance with the principle of voluntary participation, with participants recruited from five universities located in Jiangxi, Guizhou, and Guangdong provinces. To ensure balanced representation across academic backgrounds and study stages, a convenience sampling approach was employed, with efforts made to include students from various university types and academic years. Specifically, universities were categorized according to their primary disciplinary orientation (medical, comprehensive, teacher education, and science and engineering) to ensure representation of students from diverse academic fields. Although some institutions have become increasingly comprehensive in recent years, classification was based on each university’s main disciplinary focus to preserve the representativeness of academic differences. Within each university, participants were further recruited from different academic years (from freshman to fifth year) to maintain proportionality across study stages. Data collection was conducted by trained postgraduate researchers who administered surveys in randomly selected classes in offline settings. Survey responses were submitted anonymously via the online platform Wenjuanxing (www.wjx.cn), with strict confidentiality of personal information maintained. The procedure included obtaining informed consent, distributing unique QR codes for participation, and limiting each session to approximately 30 min. The study protocol was reviewed and approved by the Ethics Committee of the authors’ affiliated institution and was conducted in accordance with the ethical principles of the Declaration of Helsinki. A total of 2,318 questionnaires were collected. To ensure data validity, two lie-detection items (e.g., “Please select ‘often’ for this item”) were embedded within the questionnaire. Based on exclusion criteria—including failure on lie-detection items, abnormal response times, and duplicate submissions—215 questionnaires were discarded, resulting in 2,103 valid responses, primarily obtained online. The effective response rate was 90.72%, providing a robust dataset that enhanced the reliability and validity of the findings. Participants ranged in age from 18 to 26 years (M = 20.37, SD = 1.26) and included 1,035 males and 1,068 females across the first to fifth academic years. Given that the sample size (N = 2,103) far exceeded the minimum recommended threshold for structural equation modeling under medium effect conditions (approximately 200–500 participants), a separate power analysis was deemed unnecessary. At this sample size, the precision of parameter estimation and statistical power were considered sufficient [44, 45]. Method Digital stress The Digital Stress Scale (DSS), adapted by Liu Qiuqi et al. [46], was used to assess individuals’ levels of digital stress. The scale consists of 31 items rated on a five-point Likert scale ranging from 1 (never) to 5 (always). It encompasses six dimensions: accessibility stress (AS), approval anxiety (AA), fear of missing out (FoMO), communication overload (CO), online vigilance (OV), and social comparison (SC). Higher total scores indicate greater perceived digital stress. In the present study, the Cronbach’s α coefficient for the overall scale was 0.952, while those for the six subdimensions were 0.873, 0.879, 0.834, 0.882, 0.891, and 0.875, respectively. Results from the confirmatory factor analysis (CFA) indicated good structural validity: χ²/df = 4.163, CFI = 0.938, TLI = 0.932, SRMR = 0.039, and RMSEA = 0.046 (90% CI = 0.044–0.047). Previous studies have also demonstrated satisfactory reliability and validity of the DSS among Chinese college students [46]. Cognitive reappraisal The Emotion Regulation Questionnaire (ERQ), translated and adapted by Wang Li et al. [47], was used to assess individuals’ levels of cognitive reappraisal. The cognitive reappraisal subscale, consisting of six items, was selected for the present study. Responses were rated on a seven-point Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree). Higher scores indicate more frequent use of cognitive reappraisal strategies. In the present study, the Cronbach’s α coefficient for this subscale was 0.934. Results of CFA indicated good structural validity: χ²/df = 4.525, CFI = 0.988, TLI = 0.974, SRMR = 0.012, and RMSEA = 0.048 (90% CI = 0.035–0.063). Previous studies have also demonstrated satisfactory reliability and validity of the ERQ among Chinese college students [48]. Negative emotions Participants’ levels of anxiety and depression were assessed using the Self-Rating Anxiety Scale (SAS) and the Self-Rating Depression Scale (SDS) [49, 50]. Both scales required participants to retrospectively assess their actual experiences and emotional states over the past two weeks. Each questionnaire consisted of 20 items rated on a four-point Likert scale, ranging from 1 (rarely or none of the time) to 4 (most or all of the time). Several items were reverse-coded before summing the total score, with higher scores indicating greater levels of the respective emotional state. In the present study, the Cronbach’s α coefficients were 0.866 for the SAS and 0.797 for the SDS, both indicating good internal consistency. To evaluate the structural and discriminant validity of the two scales within the current sample, CFA were conducted. Results indicated that the two-factor model, with anxiety and depression as distinct latent variables, provided a significantly better fit than the one-factor model, in which all items loaded onto a single latent variable: Δχ²(1) = 92.41, p < 0.001. The two-factor model demonstrated satisfactory fit indices: χ²/df = 4.382, CFI = 0.939, TLI = 0.928, SRMR = 0.046, and RMSEA = 0.049 (90% CI = 0.045–0.054). These results suggest that, although the two constructs were highly correlated (r = 0.723), they can still be regarded as distinct emotional dimensions for analysis. Previous studies have similarly demonstrated satisfactory reliability and discriminant validity of the SAS and SDS among Chinese college students [51, 52]. Problematic social media use The PSMU Scale developed by Li et al. was used to assess participants’ levels of PSMU [53]. The scale consists of seven items rated on a five-point Likert scale (1 = never, 5 = always), with higher scores indicating greater levels of PSMU. In the present study, the scale demonstrated satisfactory internal consistency, with a Cronbach’s α coefficient of 0.820. CFA showed a good model fit: χ²/df = 4.381, CFI = 0.980, TLI = 0.966, SRMR = 0.021, and RMSEA = 0.051 (90% CI = 0.040–0.062). Previous research by Dou et al. [54] also confirmed that this scale has good reliability and validity among Chinese college students. Statistical processing All statistical analyses were performed using SPSS version 26.0 and Mplus version 8.3. Initially, SPSS was employed to conduct normality testing (Kolmogorov–Smirnov test), assess multicollinearity, perform Harman’s single-factor test, and carry out descriptive statistics and correlation analyses. Subsequently, structural equation modeling (SEM) was conducted in Mplus to examine mediating effects and moderating mechanisms. All models were estimated using the Maximum Likelihood Robust (MLR) estimator to account for the effects of non-normality on standard errors and significance testing [55]. Missing data were addressed using the Full Information Maximum Likelihood (FIML) approach, which maximizes the utilization of available sample information. In the mediation analyses, bias-corrected nonparametric bootstrapping with 5,000 resamples was employed to estimate confidence intervals for indirect effects [56]. An indirect effect was considered statistically significant if the 95% confidence interval did not include zero. For moderated mediation models, the Latent Moderated Structural Equations (LMS) approach was employed to test moderating effects [57]. This method estimates interaction parameters directly at the latent variable level, thereby avoiding measurement errors associated with manually constructed product terms. Because the LMS approach does not provide traditional fit indices (e.g., CFI, TLI, RMSEA), the likelihood ratio test (− 2ΔLL) was used to compare the goodness of fit between the model containing interaction terms (full model) and the model without them (restricted model) [58]. A significant − 2ΔLL, combined with an adequate fit of the restricted model, indicates that the inclusion of latent interaction terms significantly improves model fit. Across all analyses, model fit was evaluated using the Comparative Fit Index (CFI), Tucker–Lewis Index (TLI), Standardized Root Mean Square Residual (SRMR), and Root Mean Square Error of Approximation (RMSEA). Model fit was considered acceptable when CFI and TLI values exceeded 0.90, and SRMR and RMSEA values were below 0.08 [59]. In LMS analyses, the means of latent variables were fixed at 0, and their standard deviations were fixed at 1 to ensure model identification. Results Test of normality Normality was assessed for digital stress, negative emotion, cognitive reappraisal, and PSMU using the Kolmogorov–Smirnov (K–S) test. The results indicated that the absolute kurtosis values for all variables ranged from 0.099 to 2.193, and the absolute skewness values ranged from 0.226 to 0.876. According to Drezner [60], absolute normality is rarely achieved in practice; however, data can be considered approximately normally distributed when the absolute kurtosis is below 10 and the absolute skewness is below 3. Therefore, the data in this study were deemed to approximate a normal distribution. Common method bias and multicollinearity testing Two complementary approaches were employed to examine common method bias. (1) Harman’s single-factor test identified fourteen factors with eigenvalues greater than 1. The first factor accounted for 23.15% of the total variance, which is below the 40% threshold. (2) Single common-method factor approach: A common latent factor was extracted from all observed variables. The results indicated that the data did not fit the model adequately, with χ²/df = 59.635, CFI = 0.399, TLI = 0.332, SRMR = 0.194, and RMSEA = 0.167. These findings suggest that common method bias was within an acceptable range [61]. In addition, variance inflation factor (VIF) testing was conducted to assess potential multicollinearity among the variables. The results showed that VIF values ranged from 1.17 to 2.31, all below the conventional cutoff value of 5 [62], indicating that multicollinearity was not a significant concern and that the data were suitable for further analysis. Descriptive statistics and correlational analysis of variables After gender was dummy-coded (male = 1, female = 2), correlation analyses and subsequent tests were conducted. A significant positive correlation was found between digital stress and anxiety, depression, and PSMU (r = 0.372–0.630, p < 0.001). Cognitive reappraisal was not correlated with digital stress or PSMU but showed a significant negative correlation with anxiety and depression (r = -0.303 to -0.336, p < 0.001). The means, standard deviations, and correlation coefficients for each variable are presented in Table 1. Since gender and age were correlated with several core variables, they were included as control variables in subsequent analyses. Table 1Descriptive statistics and correlations of Variables. (N = 2103)1234567Gender1Age-1DS0.092**-0.155***1CR0.0240.042-0.0061Anxiety0.025-0.144***0.372***-0.303***1Depression0.081**-0.141***0.416***-0.336***0.723***1PSMU0.138***-0.060*0.630***-0.0480.368***0.383***1 M 1.51020.3702.6645.1081.4851.6582.284 SD 0.5001.2600.6501.1400.3170.4000.658DS Digital Stress, CR Cognitive Reappraisal, PSMU Problematic Social Media Use. *p < 0.05, **p < 0.01, ***p < 0.001 Table 1Descriptive statistics and correlations of Variables. (N = 2103)1234567Gender1Age-1DS0.092**-0.155***1CR0.0240.042-0.0061Anxiety0.025-0.144***0.372***-0.303***1Depression0.081**-0.141***0.416***-0.336***0.723***1PSMU0.138***-0.060*0.630***-0.0480.368***0.383***1 M 1.51020.3702.6645.1081.4851.6582.284 SD 0.5001.2600.6501.1400.3170.4000.658DS Digital Stress, CR Cognitive Reappraisal, PSMU Problematic Social Media Use. *p < 0.05, **p < 0.01, ***p < 0.001 Descriptive statistics and correlations of Variables. (N = 2103) DS Digital Stress, CR Cognitive Reappraisal, PSMU Problematic Social Media Use. *p < 0.05, **p < 0.01, ***p < 0.001 Additionally, independent sample t-tests were conducted to examine gender differences across various variables. The results indicated that females scored significantly higher than males in digital stress (Mmale = 2.603, Mfemale = 2.723, t(2101) = -4.245, p < 0.001, d = -0.185), depression (Mmale = 1.625, Mfemale = 1.690, t(2100) = -3.733, p < 0.001, d = -0.163), and PSMU (Mmale = 2.192, Mfemale = 2.373, t(2101) = -6.363, p < 0.001, d = -0.277). Test of mediation effects Given the relative complexity of the proposed model, item parceling was performed prior to model estimation to simplify the structure, reduce computational demands, and minimize random error in the LMS procedure [63, 64]. Following the factor balance method, parcels were created for the Digital Stress Scale, the Self-Anxiety Scale, and the Self-Depression Scale. After standardizing all variables, a structural equation model (SEM) was constructed with digital stress as the independent variable, PSMU as the dependent variable, and gender and age as control variables. The model demonstrated a satisfactory fit: χ²/df = 3.589, CFI = 0.921, TLI = 0.905, SRMR = 0.054, RMSEA = 0.064 (90% CI = 0.060, 0.068). Digital stress significantly and positively predicted PSMU (β = 0.716, t = 44.330, p < 0.001). Subsequently, anxiety and depression were incorporated as mediating variables based on the initial model. The modified model demonstrated a good fit: χ²/df = 3.727, CFI = 0.923, TLI = 0.904, SRMR = 0.053, RMSEA = 0.063 (90% CI = 0.059, 0.066). Digital stress significantly and positively predicted PSMU (β = 0.643, t = 30.838, p < 0.001). Additionally, digital stress positively predicted anxiety (β = 0.380, t = 17.726, p < 0.001), which in turn positively predicted PSMU (β = 0.123, t = 4.389, p < 0.001). Similarly, digital stress significantly predicted depression (β = 0.425, t = 18.330, p < 0.001), which further predicted PSMU (β = 0.086, t = 4.259, p < 0.001). These findings indicate that the predictive effect of digital stress on PSMU among college students was partially mediated by anxiety and depression, which functioned as parallel mediators. The significance of the mediation effects in the full model was assessed using the bias-corrected bootstrap method. Following the recommendations of Efron and Tibshirani [65], 5,000 bootstrap resamples were performed. The 95% confidence intervals for both indirect paths did not include zero, indicating significant indirect effects. The total indirect effect of anxiety and depression was 0.084, accounting for 11.55% of the total effect (6.46% via the anxiety path and 5.09% via the depression path). Detailed results are presented in Table 2. A Wald test was conducted to compare the two mediation paths, revealing no significant difference between them (Wald = 0.974, p = 0.323). Although the proportion of the indirect effect was relatively small, the consistent direction and statistical significance suggest that emotional variables play a stable partial mediating role in the association between digital stress and PSMU. Table 2The breakdown table of total indirect Effect, indirect Effect. (N = 2103)PathEstimateS.E.t95%CIRelative effect valuesDS→ PSMU0.6430.02130.838[0.602, 0.684]88.45%DS→ Anxiety→ PSMU0.0470.0114.213[0.022, 0.060]6.46%DS→ Depression→ PSMU0.0370.0094.094[0.018, 0.054]5.09%Total Indirect Path0.0840.0126.972[0.045, 0.082]11.55%DS Digital Stress, PSMU Problematic Social Media Use Table 2The breakdown table of total indirect Effect, indirect Effect. (N = 2103)PathEstimateS.E.t95%CIRelative effect valuesDS→ PSMU0.6430.02130.838[0.602, 0.684]88.45%DS→ Anxiety→ PSMU0.0470.0114.213[0.022, 0.060]6.46%DS→ Depression→ PSMU0.0370.0094.094[0.018, 0.054]5.09%Total Indirect Path0.0840.0126.972[0.045, 0.082]11.55%DS Digital Stress, PSMU Problematic Social Media Use The breakdown table of total indirect Effect, indirect Effect. (N = 2103) DS Digital Stress, PSMU Problematic Social Media Use Test of moderation effects Following the recommendations of Fang and Wen [58], the present study employed a latent moderated structural equation (LMS) model to examine the moderating role of cognitive reappraisal. First, a baseline model without latent interaction terms was estimated, showing a satisfactory fit (χ²/df = 4.277, CFI = 0.944, TLI = 0.935, SRMR = 0.056, RMSEA = 0.059, 90% CI = 0.056–0.061). Next, a model including the latent interaction term (digital stress × cognitive reappraisal) was tested. Results showed that the baseline model yielded an AIC of 99,579.668, whereas the inclusion of the interaction term reduced the AIC to 99,572.533 (ΔAIC = 7.135), indicating improved model parsimony and fit. Moreover, the log-likelihood of the baseline model was − 49,711.834 and improved to − 49,706.266 after adding the interaction term. The difference test yielded Δ(− 2LL) = 11.136, Δdf = 2, χ²(2) = 11.136, p = 0.004, indicating that inclusion of the latent interaction significantly improved model fit. Overall, the moderated mediation model demonstrated superior fit relative to the baseline model. Finally, as shown in Fig. 1; Table 3, the latent interaction between digital stress and cognitive reappraisal significantly predicted both anxiety (β = −0.075, t = − 4.030, p < 0.001) and depression (β = −0.070, t = − 3.741, p < 0.001). These results suggest that cognitive reappraisal significantly moderates the effects of digital stress on negative emotions, such that individuals with higher cognitive reappraisal experience lower levels of anxiety and depression under stress. A Wald test comparing the two moderating paths indicated no significant difference between the effects on anxiety and depression (Wald = 0.084, p = 0.771). Further analysis of conditional indirect effects across levels of cognitive reappraisal revealed a significant moderating effect in the anxiety pathway (β = −0.005, SE = 0.002, p = 0.012). The indirect effects at low, medium, and high levels of cognitive reappraisal were 0.042, 0.038, and 0.033, respectively (all p < 0.001), indicating that higher cognitive reappraisal attenuated the indirect effect of digital stress on PSMU via anxiety. A similar pattern was observed for the depression pathway (β = −0.003, SE = 0.002, p = 0.045), with indirect effects showing a corresponding decreasing trend (0.022 to 0.017, p < 0.01). Overall, cognitive reappraisal not only weakened the direct effects of digital stress on anxiety and depression but also reduced its indirect influence on PSMU through negative emotions, demonstrating a clear emotional buffering effect. Fig. 1Path factor. Note: AS = Accessibility Stress; AA = Approval Anxiety; FoMO = Fear of Missing Out; CO = Communication Overload; OV = Online Vigilance; SC = Social Comparison; DS = Digital Stress; CR = Cognitive Reappraisal; PSMU = Problematic Social Media Use. *p < 0.05, **p < 0.01, ***p < 0.001 Path factor. Note: AS = Accessibility Stress; AA = Approval Anxiety; FoMO = Fear of Missing Out; CO = Communication Overload; OV = Online Vigilance; SC = Social Comparison; DS = Digital Stress; CR = Cognitive Reappraisal; PSMU = Problematic Social Media Use. *p < 0.05, **p < 0.01, ***p < 0.001 Table 3The regression analysis of the moderating effect of cognitive reappraisal. (N = 2103)Outcome VariablePredictor Variable R 2 EstimateSEt95% CIAnxietyGender0.2550.0030.0190.171[-0.034, 0.041]Age-0.0470.020-2.366[-0.086, -0.008]DS0.3600.02017.648[0.320, 0.400]CR-0.3240.025-12.961[-0.373, -0.275]DS×CR-0.0750.019-4.030[-0.112, -0.039]DepressionGender0.3210.0550.0182.975[0.019, 0.091]Age0.0590.0291.991[0.001, 0.116]DS0.4030.02218.369[0.360, 0.446]CR-0.3650.024-15.084[-0.413, -0.318]DS×CR-0.0700.019-3.741[-0.106, -0.033]PSMUGender0.5470.0530.0232.362[0.009, 0.097]Age0.0550.0182.974[0.019, 0.091]DS0.6470.02130.738[0.605, 0.688]Anxiety0.1240.0284.392[0.069, 0.180]Depression0.0680.0183.758[0.033, 0.104]DS Digital Stress, CR Cognitive Reappraisal, PSMU Problematic Social Media Use Table 3The regression analysis of the moderating effect of cognitive reappraisal. (N = 2103)Outcome VariablePredictor Variable R 2 EstimateSEt95% CIAnxietyGender0.2550.0030.0190.171[-0.034, 0.041]Age-0.0470.020-2.366[-0.086, -0.008]DS0.3600.02017.648[0.320, 0.400]CR-0.3240.025-12.961[-0.373, -0.275]DS×CR-0.0750.019-4.030[-0.112, -0.039]DepressionGender0.3210.0550.0182.975[0.019, 0.091]Age0.0590.0291.991[0.001, 0.116]DS0.4030.02218.369[0.360, 0.446]CR-0.3650.024-15.084[-0.413, -0.318]DS×CR-0.0700.019-3.741[-0.106, -0.033]PSMUGender0.5470.0530.0232.362[0.009, 0.097]Age0.0550.0182.974[0.019, 0.091]DS0.6470.02130.738[0.605, 0.688]Anxiety0.1240.0284.392[0.069, 0.180]Depression0.0680.0183.758[0.033, 0.104]DS Digital Stress, CR Cognitive Reappraisal, PSMU Problematic Social Media Use The regression analysis of the moderating effect of cognitive reappraisal. (N = 2103) DS Digital Stress, CR Cognitive Reappraisal, PSMU Problematic Social Media Use The moderating effect of cognitive reappraisal was further examined using simple slope analysis. Participants were categorized into high and low cognitive reappraisal groups based on scores that were one standard deviation above and below the mean, respectively, for the moderating effect of cognitive reappraisal on the first half of the mediation path (digital stress → anxiety), the simple slope analysis indicated that digital stress significantly predicted anxiety among college students at low levels of cognitive reappraisal (b = 0.638, p < 0.001). At high levels of cognitive reappraisal, the positive predictive effect of digital stress on anxiety was attenuated (b = 0.490, p < 0.001; see Fig. 2). Second, in terms of the moderating effect of cognitive reappraisal on the first half of the mediation path (digital stress → depression), the simple slope analysis showed that digital stress significantly predicted depressive symptoms among college students at low levels of cognitive reappraisal (b = 0.714, p < 0.001). At high levels of cognitive reappraisal, the positive association between digital stress and depressive symptoms was reduced (b = 0.547, p < 0.001; see Fig. 3). Fig. 2Cognitive reappraisal moderates the relationship between digital stress and anxiety Cognitive reappraisal moderates the relationship between digital stress and anxiety Fig. 3Cognitive reappraisal moderates the relationship between digital stress and depression Cognitive reappraisal moderates the relationship between digital stress and depression Discussion Digital stress and PSMU The present study identified a significant positive correlation between digital stress and PSMU among college students, with the direct effect accounting for 89.80% of the total effect. This finding supports Hypothesis 1 and aligns with the central proposition of the I-PACE model [15], which suggests that external stressors can contribute to maladaptive media use by affecting emotional regulation and cognitive control processes. These results provide new empirical evidence for understanding the psychological role of digital stress in social media use behavior. Previous studies have suggested that the social media environment has a dual nature, functioning both as a source of stress and as a coping resource [66]. When individuals are immersed in a digital environment characterized by information overload, social comparison, and constant connectivity, they are more likely to experience psychological strain and tension [9], which may lead to more frequent or even excessive social media use [67]. The findings of this study align with these results, further emphasizing the critical role of digital stress in the risk of PSMU. From the perspective of cognitive mechanisms, the association between digital stress and PSMU can be interpreted in two ways: (1) stress experiences may induce an attentional bias toward social cues, increasing sensitivity to social feedback [68]; and (2) prolonged exposure to stress may deplete self-control resources, making it more difficult for individuals to regulate their online behavior [69]. These interpretations provide plausible psychological explanations for the direct effect, although further longitudinal or experimental studies are necessary for validation. Overall, the findings provide partial support for the applicability of the I-PACE model in the digital context, suggesting that digital stress may be a key environmental antecedent of PSMU. Future research should further differentiate among distinct types of digital stress to clarify their specific psychological pathways influencing social media use. At a practical level, these findings underscore the need to develop targeted mental health education programs focused on managing digital stress. For example, interventions such as “digital disconnection training” and “social boundary management” could be integrated into university mental health curricula to help students recognize and mitigate the cumulative effects of digital stress, thereby promoting healthier and more self-regulated patterns of social media use. The mediating role of negative emotions The present study found that the relationship between digital stress and PSMU was partially mediated by two emotional variables: anxiety and depression. This finding aligns with the core assumptions of the I-PACE model [15], which posits that external stressors are linked to individuals’ emotional responses, subsequently affecting patterns of media use. Although the current cross-sectional design does not permit verification of causal direction, the results suggest that emotional experiences may serve as a crucial psychological link between digital stress and social media use. Consistent with prior literature, the present study found that anxiety and depression exhibit distinct correlational patterns under conditions of digital stress. Anxiety is typically associated with heightened environmental sensitivity and vigilance [70]; in digital contexts, factors such as information updates, social comparison, and constant online availability may covary with anxious responses. In contrast, depression often reflects inefficient coping or emotional exhaustion resulting from prolonged stress exposure [71, 72]. The high correlation between anxiety and depression (r = 0.723) indicates substantial empirical overlap, potentially representing a broader negative affective response. However, these constructs remain theoretically distinguishable, reflecting different pathways of emotional processing [73]. Therefore, the present results should be interpreted as part of a broader pattern of covariation between digital stress and negative affect, rather than as evidence of separate causal pathways involving anxiety and depression. Although the indirect effects of the emotional variables accounted for only a modest proportion of the total effect (approximately 11%), they nonetheless highlight a meaningful psychological process: in digital contexts, negative emotional experiences may partially mediate the relationship between stress and behavioral dysregulation. This finding aligns with the extended model of emotion regulation [43], which suggests that individuals under sustained external stress are more likely to engage in immediate relief strategies to alleviate negative affect. Given the low effort and high immediacy of social media use [2], such behaviors often accompany emotion-regulation attempts in stressful digital environments. Importantly, this coexistence does not necessarily imply that emotional dysregulation causes problematic use; rather, it may reflect a mutually reinforcing or co-varying psychological process [74, 75]. Taken together, these findings suggest that emotional processes should be considered an important—but not the sole—dimension when interpreting the relationship between digital stress and social media use. Future research could employ longitudinal or experimental designs to examine the temporal sequence of various emotional responses and explore additional emotional variables beyond anxiety and depression (e.g., loneliness or tension) to develop a more comprehensive affective mediation framework. At the applied level, these findings suggest that university-based mental health interventions should focus on helping students recognize and label their emotional responses in digital contexts and strengthen adaptive emotion regulation strategies during high-stress conditions. For example, integrating emotion-awareness training with self-reflective practices tailored to digital environments may help reduce the correlation between accumulated stress and excessive social media use [76]. The moderating role of cognitive reappraisal The present study identified a significant interaction between the level of cognitive reappraisal and the relationship between digital stress and negative emotions. Specifically, higher levels of cognitive reappraisal were associated with weaker correlations between digital stress and both anxiety and depression. This finding aligns with emotion regulation theory [38], which posits that an individual’s cognitive processing style during stress can influence the intensity of emotional experiences. In other words, although digital stress is generally positively correlated with higher levels of negative emotions, the strength of this association varies among individuals with different tendencies toward cognitive regulation. According to the model proposed by Gross and Jazaieri [77], cognitive reappraisal typically operates at the early stage of emotion generation by altering one’s subjective interpretation of stress-inducing events. In digital contexts, this may involve reinterpreting information overload, social comparison, or negative feedback, which are associated with lower levels of anxiety and depression. The present findings align with this theoretical framework, suggesting that cognitive reappraisal may serve as a broadly protective factor in digital environments [78]. Nevertheless, this conclusion should be interpreted cautiously, as the current research design does not allow for the identification of the dynamic sequence of underlying psychological mechanisms. It is noteworthy that the moderating effects of cognitive reappraisal were consistent across both anxiety and depression pathways, suggesting that its influence may arise from a general emotional processing mechanism rather than from the selective regulation of specific emotions [79]. This finding supports the cross-situational stability of emotion regulation strategies and indicates that individual differences in emotional–cognitive resources may be a key factor underlying variations in emotional responses to digital stress. However, the current results should be interpreted as statistical interactions rather than direct evidence of psychological mechanisms. Future research could incorporate experimental or longitudinal designs to examine the temporal sequence and potential mediating processes of cognitive reappraisal—such as threat appraisal, attentional allocation, or metacognitive awareness—in order to more comprehensively elucidate its regulatory mechanisms. From a practical standpoint, the present findings suggest that emotion regulation training may represent a promising approach for interventions targeting digital stress. Rather than focusing solely on reducing social media usage, helping students learn to reinterpret stressful digital situations and identify irrational beliefs or patterns of excessive comparison may more effectively alleviate negative emotional responses and reduce the association between stress and PSMU [80]. Research contributions and limitations Guided by the I-PACE model, the present study developed an integrative framework incorporating emotional mediation and cognitive regulation to examine the relationship between digital stress and PSMU. The results largely aligned with the theoretical assumptions of the I-PACE model, providing preliminary empirical evidence for the psychological connections between stress experiences and media use behaviors in digital environments. This study extends the applicability of the I-PACE model by demonstrating that digital stress—a novel form of stressor—can be consistently associated with emotional responses and behavioral tendencies among university students. Furthermore, the use of latent moderated structural equation modeling enhanced estimation precision and offers a methodological reference for future research. Nevertheless, these conclusions should be interpreted in light of several limitations. First, the cross-sectional design of this study allows for the identification of statistical associations among variables but does not permit causal inference. Previous research has indicated that emotional states and social media use may exhibit bidirectional influences [20]. Therefore, the present model should be regarded as an empirical test of a theoretical framework rather than a verification of causal mechanisms. Future studies employing longitudinal tracking or experimental manipulation are necessary to clarify the temporal sequence and dynamic interplay among digital stress, emotional processes, and behavioral outcomes. Second, the sample was drawn exclusively from Chinese university students, and the cultural as well as educational context may have influenced the results. The collectivist orientation and competitive academic environment characteristic of Chinese culture may increase individuals’ sensitivity to social comparison and external evaluation [81], thereby intensifying the psychological burden of digital stress during social interactions. In contrast, in cultures that emphasize personal autonomy and self-realization, individuals may rely more heavily on independent emotion regulation strategies when coping with similar stressors [82]. Cross-cultural research is therefore essential for assessing the generalizability of the current model and for elucidating how variations in cultural orientation shape stress perception, emotional responses, and coping mechanisms, thereby enriching the understanding of sociocultural processes within digital ecosystems. Third, because the data were obtained through self-report questionnaires, potential influences of social desirability and subjective reporting errors cannot be entirely ruled out, despite controlling for common method bias. Future studies could incorporate objective indicators—such as digital behavior logs or physiological measures (e.g., heart rate variability, electrodermal activity)—to enhance the ecological validity of psychological assessments. Finally, regarding the measurement of cognitive reappraisal, the present study focused solely on its overall level and did not differentiate its roles across distinct stages of stress perception, emotion generation, and response regulation. Future research could employ dynamic tracking or experimental paradigms to delineate the temporal dynamics of cognitive reappraisal within the digital stress process. Furthermore, individual difference factors—such as generational characteristics, online identity, or social motivation—may influence the threshold for stress-to-emotion transformation, offering promising directions for future investigation. Conclusion The study revealed that digital stress was positively associated with PSMU, with anxiety and depression partially mediating this relationship. Moreover, a significant interaction was observed between cognitive reappraisal and digital stress in predicting negative emotions, indicating that higher levels of cognitive reappraisal weakened the link between stress and emotional responses. These results offer preliminary empirical support for the linkage in digital contexts and, to some extent, validate the assumptions of the I-PACE model. However, it is important to note that the study employed a cross-sectional design; therefore, the findings indicate statistical associations rather than causal relationships. Future longitudinal or experimental studies are necessary to verify the temporal and causal mechanisms underlying these relationships. From a practical perspective, the findings suggest that enhancing university students’ cognitive reappraisal abilities and emotional regulation skills may help mitigate the behavioral risks associated with digital stress. Future research should incorporate multimethod and cross-cultural approaches to develop a more comprehensive understanding of the psychological mechanisms underlying digital stress.