The S-O-R model, originally rooted in environmental psychology, provides a structured approach for analyzing how external or internal cues shape human behavior. It posits that environmental inputs (S) impact inner cognitive and psychological conditions (O), which eventually dictates the observable response (R) [16]. Within this framework, stimuli refer to various cues or conditions that trigger individuals’ cognitive and emotional reactions, while the organism represents internal processes operating between environmental inputs and later actions. These processes involve cognitive appraisals and affective experiences that shape how stimuli are interpreted. The response reflects the observable behaviors that result from these internal processes [17].

Although the S-O-R model was initially developed to explain human responses in environmental and consumer contexts, its utility has significantly expanded into the realms of technology use and digital behavior [18]. Its core logic lies in linking stimuli to behavioral responses through mediating organismic states, thereby establishing an integrative analytical framework. While the S-O-R framework traditionally focuses on environmental stimuli, recent research has extended it to incorporate individual psychological and dispositional characteristics [19]. Owing to its theoretical flexibility, the S-O-R framework has been widely employed to investigate various addictive behaviors, including live-streaming addiction [20], artificial intelligence addiction [21], gaming disorder [22], and alcohol and other drug addiction [23]. However, many existing studies treat the organism component as a unified internal state, with limited attention to the differentiation of underlying psychological processes. As a result, the specific mechanisms through which addictive behaviors develop remain insufficiently specified. Distinguishing between these processes is crucial for formulating more precise and effective response strategies.

In response to this deficiency, the research broadens the S-O-R framework by differentiating between two types of organismic processes: cognitive strain and habitual behavior. Specifically, information overload is conceptualized as a form of cognitive strain, whereas social media stalking reflects a pattern of habitual monitoring. By incorporating this dual-path structure, the present research seeks to offer deeper insights into how psychological predispositions relate to SMA.

Social media stalking (SMS) involves the repeated and impulsive monitoring of other users’ profiles to track their recent online activities. Such behaviors may occur either intentionally or unintentionally and are often driven by curiosity and a desire to remain informed about others’ lives [24].

Because individuals engaged in SMS often devote excessive time to social media, such engagement can interfere with daily routines, including nocturnal sleep [25]. This pattern creates a cycle of habitual checking and increased time investment, which may facilitate the emergence of compulsive tendencies. Research indicates that SMS strongly predicts excessive social media use [24], as individuals who monitor others’ information tend to spend more time on platforms, a behavior associated with addictive tendencies [26].

Although SMS involves frequent engagement, it should not be equated with SMA itself. Instead, it is conceptualized as an intermediate behavioral process characterized by habitual monitoring, which may increase individuals’ exposure to social media content and repeated interaction with the platform, without necessarily implying a loss of control. By distinguishing this habit-based process from addiction, this perspective helps clarify how routine monitoring behaviors may contribute to the development of problematic engagement. Therefore, the study postulates that:

Information overload (IO) is triggered when people receive more data than they can effectively process [27]. While prior research often views IO as a consequence of addiction, this perspective overlooks the antecedent impact of the social media environment on cognitive processes. In contemporary digital contexts, where users are continuously exposed to multi-source and rapidly updating streams, IO functions as a pervasive environmental condition rather than a mere symptomatic outcome. Consequently, it is essential to reframe IO as a cognitive mechanism active during the usage process rather than a secondary result variable.

According to Cognitive Load Theory, people have finite mental resources. When exposed to large volumes of complex information, processing demands increase substantially, and IO can hinder the efficient and effective use of information. This may lead to a diminished sense of control over one’s situation and trigger a range of negative responses, including uncertainty, stress, and reduced motivation [28]. As information volume keeps expanding, distinguishing useful, accurate, and high-quality content becomes progressively more challenging. Consequently, individuals may perform below expected levels, abandon tasks, procrastinate, or rely on superficial and low-quality information processing strategies [28]. These patterns of engagement may increase susceptibility to problematic use.

Thus, beyond being a potential consequence of excessive use, IO can also function as a cognitively mediated source of pressure arising from information environments. By weakening self-regulatory capacity, it may indirectly shape patterns of social media engagement and increase susceptibility to addictive behavior.

The S-O-R logic dictates that intrinsic or extrinsic triggers evoke specific internal conditions, which in turn influence behavioral responses [45]. A critical distinction within this framework lies in its emphasis on indirect relationships, where the impact of a stimulus is transmitted through mediating mechanisms [46]. This study identifies social media stalking and information overload as the primary cognitive and behavioral pathways through which psychological predispositions are transformed into addictive outcomes.

Social media stalking represents a behavioral pathway characterized by habitual monitoring and repeated checking of others’ activities. Although such behaviors involve frequent engagement, they do not necessarily imply a loss of control and are therefore conceptually distinct from social media addiction. Instead, they can be understood as intermediate behavioral patterns that increase exposure to social media and reinforce routine interaction with the platform. Prior research has shown that cyberstalking as a mediator, linking dark personality traits to dysfunctional social media engagement [47]. Moreover, stalking behaviors have been shown to correlate positively with Internet addiction [48] and compulsive social media use [24]. By establishing a cycle of constant profile checking and heightened platform engagement, SMS serves as the behavioral bridge that reinforces the transition from initial psychological needs to compulsive usage patterns.

Parallel to behavioral pathways, information overload functions as a cognitive mediator. This study conceptualizes information overload as a state of cognitive constraint where excessive informational input exhausts an individual’s processing capacity [49]. Rather than being solely a consequence of heavy use, information overload can function as a cognitive condition that shapes how individuals engage with social media. When cognitive resources are excessively consumed, users may rely on superficial and low-quality information processing strategies [28], which can increase susceptibility to problematic use. Existing studies have linked information overload to fatigue [50], stress [51], and maladaptive usage patterns [52], suggesting its role as a mechanism underlying behavioral outcomes.

By distinguishing these behavioral and cognitive processes, the study conceptualizes social media stalking and information overload as intermediate mechanisms that precede addictive use. This distinction helps clarify how psychological predispositions are translated into problematic engagement through different pathways.

Accordingly, we propose:

Aligning with the S-O-R logic, the current study examines how specific psychological predispositions are associated with SMA through the dual pathways of cognitive strain and habitual monitoring. Fig. 1 presents the conceptual framework and the hypothesized paths between these variables.

Using convenience sampling, this study recruited Chinese adults aged 18 years and above who had prior experience with social media and reported using at least one social media platform daily. Data were collected in January 2026, primarily through Credamo, a commonly used online survey platform in China, with a small proportion of responses obtained through offline channels. All individuals participated voluntarily after providing informed consent. This study was reviewed and approved by the Institutional Review Board of the School of Journalism and Communication, Beijing Normal University (Approval No.: BNUJ&C20250408005), where the corresponding author was affiliated at the time of ethics application. After excluding responses that exhibited patterned answering, contained missing values, or were completed in an unusually short time, 258 valid responses were kept for subsequent analyses, satisfying the minimum threshold of 200 necessary for structural equation modeling (SEM) models with moderate complexity [53].

The final sample included 155 females (60.1%) and 103 males (39.9%), averaging 31.45 years of age (Mean = 31.45, SD = 8.65). Regarding age distribution, participants fell into the following brackets: 18–27 (40.3%), 28–37 (40.7%), 38–47 (12.0%), and 48 years or older (7.0%). Regarding employment status, the majority of participants were employed full-time (71.7%), followed by students (13.6%), freelancers (8.5%), and entrepreneurs (3.5%), with a small proportion identifying as workers, unemployed, or retired. With respect to educational attainment, most participants held an undergraduate degree (69.8%), while 16.7% reported a postgraduate degree. The remaining participants reported education levels ranging from junior college to middle school. Participants’ average daily engagement with social media was also recorded. Approximately 29.1% indicated using social media for 2–3 h per day, followed by 21.7% for 3–4 h, 15.1% for 4–5 h, and 10.1% for 5–6 h. Smaller proportions reported either relatively light use (less than 2 h per day) or heavy use (more than 6 h per day). The distribution of demographic variables across the final sample is illustrated in Table 1.

To ensure cultural and linguistic equivalence, scales adopted from previous literature underwent a Chinese translation process. All items were rated on a five-point Likert scale (1 = strongly disagree; 5 = strongly agree).

Statistical analysis was conducted using AMOS 26.0 and SPSS 27.0 (IBM Corp., Armonk, NY, USA). Consistent with the two-step SEM procedure, the measurement model was first evaluated for reliability and validity, after which the structural paths were tested.

The measurement model exhibited a satisfactory fit, indicating that the proposed conceptual structure aligns well with the empirical data: χ²/df = 1.45, CFI (comparative fit index) = 0.97, TLI (Tucker-Lewis index) = 0.97, RMSEA (root mean square error of approximation) = 0.04. Three items (SMS3, IO5, and SMA1) displayed factor loadings under the suggested 0.70 cut-off and were consequently omitted from later analyses [56]. The choice to discard these items was informed by both empirical data and theoretical grounds. Specifically, retaining items with low loadings may weaken construct validity and introduce measurement error. After removal, all constructs maintained adequate content coverage, suggesting that the exclusion did not compromise the conceptual integrity of the measures. Furthermore, both Cronbach’s alpha and composite reliability (CR) for all constructs exceeded 0.70, indicating robust reliability and internal consistency [57]. Although several constructs showed relatively high Cronbach’s alpha values, which may indicate potential item redundancy, this also reflects a high degree of internal consistency among items capturing similar underlying dimensions. Given that the retained items represent theoretically coherent aspects of each construct, the reliability levels were considered acceptable for the purposes of this study. Detailed results for factor loadings, Cronbach’s alpha, composite reliability, and average variance extracted (AVE) are reported in Table 2. Discriminant validity was assessed using the Fornell–Larcker criterion. As shown in Table 3, the square root of the AVE for each construct was greater than its correlations with all other constructs, thereby supporting the adequacy of discriminant validity [58].

We conducted a descriptive statistical analysis on the dataset. As illustrated in Table 3, the average scores of the six primary variables ranged from 2.634 to 3.478. Specifically, SMS exhibited the highest mean (Mean = 3.478, SD = 0.943), while BP showed the lowest (Mean = 2.634, SD = 1.023). These results suggest that participants experienced moderate levels of psychological traits and social media-related behaviors. The standard deviations, which varied between 0.943 and 1.136, indicate a reasonable degree of dispersion in the responses, providing a suitable basis for further SEM analysis.

Pearson correlation analysis examined links among the variables. Table 3 shows that all constructs were positively related at the 0.01 significance level. In particular, boredom proneness, anxiety, and social comparison orientation were significantly associated with both information overload and social media stalking. Moreover, information overload and social media stalking were strongly correlated with social media addiction. Overall, the correlations were of moderate to strong magnitude, providing preliminary support for the proposed hypotheses and satisfying the prerequisite for subsequent structural equation modeling. In addition, multicollinearity among the constructs was examined using variance inflation factors (VIFs). The VIF values ranged from 1.998 to 2.691, all of which were below the recommended threshold of 5 [59], indicating that multicollinearity was not a serious concern in this study. Moreover, we used the common latent factor technique to test common method bias (CMB). The one-factor constrained model exhibited poor indices, with χ² = 2985.54, df = 495, CFI = 0.66, TLI = 0.64, RMSEA = 0.14, while the standardized regression weights in this model displayed a difference of less than 0.2 [60]. More importantly, the one-factor constrained model demonstrated inferior fit in comparison to the six-factor unconstrained model (Δχ² (df) = 2288.57 (15), p < 0.001). This affirms that CMB does not pose a significant concern in our study.

The structural model demonstrated a good overall fit to the data (χ²/df = 1.46, CFI = 0.97, TLI = 0.97, RMSEA = 0.04). The results of the hypothesis testing are shown in Table 4. Regarding the mediating variables, both social media stalking (β = 0.245, p < 0.01) and information overload (β = 0.553, p < 0.01) were significantly and positively associated with social media addiction, providing support for H1 and H2. Boredom proneness showed significant positive associations with social media stalking (β = 0.112, p < 0.05) and information overload (β = 0.577, p < 0.01), supporting H3a and H3b. However, its direct relationship with social media addiction was not significant (β = 0.137, p > 0.05), leading to the rejection of H3c. Anxiety did not exhibit significant associations with social media stalking (β = 0.084, p > 0.05), information overload (β = 0.097, p > 0.05), or social media addiction (β = 0.116, p > 0.05); therefore, H4a, H4b, and H4c were not supported. Social comparison was strongly associated with social media stalking (β = 0.651, p < 0.01) and was also positively associated with information overload (β = 0.165, p < 0.05), supporting H5a and H5b. However, its direct link to social media addiction was not significant (β = 0.041, p > 0.05), resulting in the rejection of H5c.

The structural model demonstrated strong explanatory power, accounting for 56.7% of the variance in information overload (R² = 0.567), 61.7% of the variance in social media stalking (R² = 0.617), and 85.9% of the variance in social media addiction (R² = 0.859). The notably high R² for social media addiction should be interpreted with caution, as it may partly reflect the conceptual proximity between the mediators and the outcome, as further discussed in the limitations section.

As shown in Table 5, social media stalking significantly mediated the relationship between boredom proneness and social media addiction (β = 0.027, 95% CI [0.003, 0.071]), supporting the proposed mediating effect. In addition, social media stalking also mediated the relationship between social comparison and social media addiction (β = 0.160, 95% CI [0.071, 0.278]). However, the indirect link from anxiety to social media addiction through social media stalking was not significant (β = 0.021, 95% CI [−0.020, 0.079]).

With respect to information overload, the results indicated a significant indirect association between boredom proneness and social media addiction (β = 0.319, 95% CI [0.209, 0.455]), suggesting that information overload played a substantial mediating role in this relationship. Similarly, information overload significantly mediated the relationship between social comparison and social media addiction (β = 0.091, 95% CI [0.005, 0.199]). In contrast, the indirect pathway from anxiety to social media addiction via information overload was not statistically significant (β = 0.054, 95% CI [−0.033, 0.153]).

This study refines the theoretical understanding of SMA by distinguishing between cognitive and behavioral mediating pathways. While previous S-O-R research often overlooks specific organism-level processes, our findings identify information overload and social media stalking as two functionally distinct routes linking internal stimuli to addictive outcomes. By conceptualizing these variables as mechanistic precursors rather than mere symptoms, this research offers a granular account of how psychological predispositions translate into problematic engagement, thereby broadening the interpretive scope of the S-O-R model in digital contexts.

Regarding practical implications, these findings provide a strategic blueprint for mitigating social media addiction. Preventive measures might be more impactful when focusing not only on reducing screen time but also on addressing the underlying psychological drivers that initiate excessive engagement. For example, strategies that promote alternative offline activities during periods of boredom may lessen dependence on social media as a substitute coping outlet. Additionally, the significant role of information overload suggests that platform design should consider cognitive load management. Features such as usage reminders, adaptive content filtering, and information-density regulation may help reduce perceived overload, thereby interrupting the progression from frequent engagement to addictive use patterns.

Notwithstanding the insights provided by this work, certain constraints persist that offer valuable avenues for subsequent investigations. First, a notable limitation is the relatively high R² values observed in the structural model, particularly regarding social media addiction. While diagnostic tests (e.g., VIF) ruled out severe multicollinearity, this high degree of explained variance may partly stem from the inherent behavioral overlap between habit-based monitoring and addiction symptoms. Furthermore, the use of a cross-sectional, self-reported approach during a single collection period might have introduced common method bias and potentially inflated the reported correlations. Prospective studies ought to utilize longitudinal frameworks or experimental methods to more robustly validate these effect sizes and capture the underlying dynamic mechanisms.

Second, the current model assumes a linear and unidirectional pathway; however, the relationships between information overload, social media stalking, and addiction may be reciprocal or self-reinforcing. Future studies are encouraged to explore these potential bidirectional dynamics to offer a more sophisticated insight into the developing digital landscape.

Finally, the employment of data from a solitary source may constrain the universal relevance of these insights. The non-significant results observed for certain variables, such as anxiety, further suggest that specific cultural or contextual mechanisms may not have been fully captured within the current scope. Future research should incorporate extensive contextual factors and more varied, broad-based datasets to improve the stability and cross-contextual generalizability of the results.