Latent Profile Analysis of Excessive Internet Use in Adolescents: Gender-Specific Pathways to Depression Through Loneliness

Abstract

Excessive internet use (EIU) has emerged as a significant concern for adolescent mental health, yet existing research often overlooks the heterogeneity of its psychological impact. This study draws on a nationally representative sample of 71,640 Finnish adolescents to identify distinct EIU profiles using latent profile analysis (LPA) and to examine whether loneliness mediates the EIU–depression relationship differently across profiles and gender groups. Four EIU profiles were identified: Low-risk, Moderate-risk, High-involvement with self-control, and High-risk users. Multi-group structural equation modeling revealed substantial variation in mediation mechanisms. Notably, a novel suppression effect emerged among low-risk users, suggesting potential compensatory benefits of moderate internet engagement, while the indirect pathway via loneliness was absent among high-risk users. Gender differences were most pronounced in the loneliness–depression path, with girls consistently showing stronger associations. Findings underscore the developmental importance of gender-sensitive, person-centered approaches to understanding digital risk.

Introduction

The internet has fundamentally transformed adolescent social and emotional development, serving as a primary platform for information exchange, learning, and interpersonal connection in the digital age (Castells, 2009). However, the pervasive integration of digital technology into adolescents’ daily lives has raised significant concerns about its potential adverse impacts on mental health (Twenge & Campbell, 2019). Excessive internet use (EIU), characterized by compulsive or poorly regulated internet engagement that interferes with daily functioning (Kuss & Lopez-Fernandez, 2016), has emerged as a pressing global public health concern affecting approximately 14.22% of adolescents worldwide (Meng et al., 2022).

Epidemiological evidence reveals that EIU prevalence rates continue to rise across diverse cultural contexts (Pan et al., 2020) with concurrent increases in adolescent depression rates. Approximately 34% of adolescents report elevated depressive symptoms, with 8% meeting criteria for major depressive disorder (Shorey et al., 2022). The intersection of these two public health challenges has generated substantial research interest, yet the underlying mechanisms linking EIU to depressive outcomes remain inadequately understood (Soriano-Molina et al., 2025).

Traditional variable-centered approaches have predominantly treated adolescent internet use behaviors as homogeneous phenomena, assuming uniform relationships between EIU and psychological outcomes across all individuals (Griffiths, 2005; Wang et al., 2024). However, emerging evidence suggests substantial individual differences in both the manifestation and consequences of internet use behaviors (Tóth-Király et al., 2021), challenging the assumption of population homogeneity. To address these limitations and capture the heterogeneous nature of EIU-depression associations, the present study employed a person-centered latent profile analysis (LPA) approach to identify distinct EIU patterns and examine differential mediation mechanisms across subgroups and genders.

Previous Empirical Studies and Theories About EIU and Depressive Symptoms

EIU and Depressive Symptoms Relation

The relation between EIU and depressive symptoms has been extensively documented in cross-sectional research, consistently demonstrating positive associations across diverse adolescent populations (Soriano-Molina et al., 2025). However, longitudinal findings reveal more complex, often bidirectional relationships over time (Yang et al., 2022). Some longitudinal studies support the pathway from EIU to depression, suggesting that problematic internet use disrupts sleep patterns, reduces face-to-face social interactions, and interferes with academic performance, subsequently contributing to depressive symptomatology (Primack et al., 2017; Woods & Scott, 2016). Conversely, other investigations demonstrate that adolescents with pre-existing depressive symptoms may engage in EIU as a maladaptive coping mechanism (Nesi & Prinstein, 2015).

Recent longitudinal research increasingly supports reciprocal influence models, wherein EIU predicts increased depressive symptoms while pre-existing depression exacerbates problematic internet use patterns (Yang et al., 2022). Despite extensive research efforts, effect sizes vary considerably across studies and populations (Cai et al., 2023), with some investigations reporting null or even protective effects under specific conditions (Weinstein, 2018). This variability indicates that the EIU-depression relationship may be more complex than previously assumed, highlighting the importance of individual differences and contextual factors in shaping these associations.

Previous research predominantly relies on a variable-centered perspective, which typically overlooks individual differences by assuming consistent associations between internet use behaviors and psychological outcomes across adolescents (Peng & Liao, 2023; Wang et al., 2024). However, this assumption has been increasingly challenged by evidence suggesting substantial heterogeneity in both EIU manifestation and consequences (Saputra et al., 2024; Tóth-Király et al., 2021). Person-centered approaches offer methodological advantages for capturing this heterogeneity by identifying distinct subgroups with similar behavioral patterns rather than assuming population homogeneity.

The Potential Mediating Role of Loneliness

The mechanisms linking EIU to depressive outcomes can be understood through two complementary theoretical frameworks: the social displacement hypothesis (Kraut et al., 1998) and the social compensation hypothesis (Valkenburg & Peter, 2007). Rather than viewing these as competing theories, recent evidence suggests they may operate simultaneously across different contexts and individual characteristics (Nie, 2001; Smith et al., 2021).

The social displacement hypothesis proposes that EIU reduces opportunities for face-to-face interaction and offline activities, leading to increased feelings of social isolation and subsequent psychological distress (Kraut et al., 1998). Excessive internet engagement may contribute to loneliness by disrupting the formation and maintenance of meaningful social relationships, particularly when online interactions fail to provide adequate social and emotional support (Nie, 2001; Smith et al., 2021).

Conversely, the social compensation hypothesis suggests that internet platforms may provide alternative channels for social connection and support, particularly benefiting adolescents who experience difficulties in offline social contexts (Valkenburg & Peter, 2007). For socially anxious or isolated individuals, internet use may actually reduce loneliness and emotional distress by facilitating social connections that would otherwise be difficult to establish (Kim et al., 2009).

These seemingly opposite ideas can be brought together by understanding that their importance likely depends on individual characteristics, usage patterns, and contextual factors. When socially skilled teens use the internet in reasonable amounts for specific purposes, this may support the compensation idea. However, when teens use the internet too much in a compulsive way that takes away from real-life activities, this may fit with the displacement idea. This combined view suggests that how loneliness affects the relationship between EIU and depression may change greatly depending on different internet use patterns and individual differences (Nowland et al., 2018).

Loneliness, defined as the subjective discrepancy between desired and actual social relationships characterized by feelings of disconnection and unfulfilled social needs (Qualter et al., 2015), represents a critical psychological mechanism in this theoretical framework. Chronic loneliness is consistently associated with increased depressive symptomatology across developmental periods (Cacioppo & Hawkley, 2009). In the context of internet use, EIU may both result from and contribute to loneliness through complex feedback loops (Kim et al., 2009), with the direction and strength of these associations potentially varying across different usage patterns and individual characteristics.

Gender Difference in Adolescent’s Digital Behavior and Emotional Vulnerability

Gender differences in digital behavior and emotional vulnerability represent another critical dimension moderating EIU-depression associations (Twenge & Martin, 2020). Research consistently demonstrates distinct patterns of internet engagement between male and female adolescents, with males more likely to engage in achievement-oriented activities such as online gaming and information seeking, while females prioritize relationship-focused activities including social networking and emotional expression (Nesi & Prinstein, 2015; Svensson et al., 2022; Twenge & Martin, 2020).

Gender role socialization theory suggests that girls and boys develop distinct vulnerability profiles and coping mechanisms that may moderate EIU-mental health relationships (Rose & Rudolph, 2006). Girls typically exhibit higher levels of interpersonal sensitivity, emotion-focused coping, and co-rumination behaviors (Rose & Rudolph, 2006), potentially increasing their susceptibility to social and relational stressors associated with problematic internet use (Germani et al., 2023; Svensson et al., 2022). Boys often demonstrate greater engagement in problem-focused coping strategies but may also exhibit higher levels of externalizing behaviors and risk-taking tendencies (Zhu et al., 2024).

Recent evidence suggests gender-differentiated patterns in internet-related loneliness experiences, with girls showing heightened susceptibility to negative emotional outcomes from online social displacement, while boys may derive compensatory benefits from digital platforms’ emotional expression opportunities (Gioia et al., 2021; Twenge et al., 2021; Valkenburg et al., 2011). These gender-differentiated loneliness experiences may create distinct pathways linking EIU to depressive symptoms, with important implications for both risk identification and intervention development.

Person-centered Approaches and Latent Profile Analysis (LPA)

Person-centered approaches offer several methodological and theoretical advantages over traditional variable-centered methods. Rather than examining relationships between variables across populations, person-centered methods focus on patterns of behavior within individuals, allowing for the identification of naturally occurring subgroups with distinct characteristic profiles populations (Bergman & Magnusson, 1997; Laursen & Hoff, 2006).

LPA, as a model-based clustering technique, provides a statistically rigorous framework for identifying distinct subgroups while maintaining appropriate model comparison and classification accuracy (Muthén & Muthén, 2000; Spurk et al., 2020). In the EIU context, LPA can identify how adolescents cluster based on core addiction symptoms, revealing both similarities and differences between distinct usage profiles (Howard & Hoffman, 2018).

Consistent with Griffiths’ addiction components model (Griffiths, 2005), EIU represents a behavioral pattern involving core addiction features including salience, relapse, tolerance, withdrawal, and conflict (Škařupová et al., 2015). Research indicates that adolescents with problematic internet use report higher levels of salience, tolerance, withdrawal, and conflict compared to non-problematic users, while mood modification shows less consistent differentiation (Jameel et al., 2019).

Accumulating evidence from person-centered approaches further supports the heterogeneity of EIU patterns and their differential associations with mental health outcomes. A Portuguese study identified four distinct addition profiles based on six dimensions, with male gender and peer relationship difficulties emerging as important correlates (Martins et al., 2022). The Chinese study revealed four distinct adolescent profiles with varying levels of aggression, depression, and anxiety, wherein high internet addiction groups demonstrated significantly greater psychological distress (Wang et al., 2024). Longitudinal research using latent transition analysis has revealed dynamic patterns in internet addiction development, with boys showing greater likelihood of transitioning into high-addiction groups while anxiety and depression serve as significant predictors of trajectory changes (Hu et al., 2023).

Recent platform-specific investigations have revealed meaningful usage heterogeneity, with LPA identifying four distinct short video and social media usage patterns among adolescents. High-intensity usage profiles characterized by either extended duration or extensive social connections were associated with significantly poorer mental health outcomes, including elevated depression and reduced life satisfaction (Liu et al., 2024). However, person-centered research examining group differences in EIU-depression mediation mechanisms remains limited (Saputra et al., 2024).

The Presents Study

Based on the theoretical foundations and empirical gaps identified above, the present study aimed to: (1) test loneliness as a potential mediator in the EIU-depression relationship among adolescents; (2) identify distinct latent profiles of EIU using person-centered LPA to capture the heterogeneity in the internet usage patterns; and (3) examine gender differences in the mediation pathways across different EIU profiles. We addressed the following three research questions.

• (1) Does loneliness mediate the association between adolescent EIU and depressive symptoms?

• (2) Do distinct latent profiles exist among adolescents based on their EIU characteristics?

• (3) Do the EIU-loneliness-depressive symptoms mediation pathways show gender differences within each EIU profile, and do these pathways differ in significance and effect size across EIU profiles?

Method

Participants

The current study analyzed data from the School Health Promotion (SHP) Study, a large-scale cross-sectional survey administered by the Finnish Institute for Health and Welfare (THL). This ongoing surveillance program has monitored adolescent health behaviors since 1996, providing comprehensive population-level data on youth well-being indicators (THL, 2024). The study protocol received ethical approval from the institutional review board and was conducted in accordance with the Declaration of Helsinki. Data collection procedures followed strict anonymity protocols with appropriate informed consent requirements.

Data collection occurred during regular school hours using self-administered questionnaires. Students under 15 years of age required parental consent prior to participation. The present analysis utilized data from eighth and ninth-grade students participating in the 2019 survey wave (THL, 2019).

The original dataset included 87,215 respondents. Given the substantial sample size and the need for complete data across all analytical steps including LPA, listwise deletion was employed, resulting in a final analytic sample of 71,640 students. This approach was deemed appropriate as the large remaining sample size (82.1% retention) maintained adequate statistical power while ensuring stable parameter estimation across complex analytical procedures (Little & Rubin, 2019). The final sample showed balanced gender representation (46.8% boys, 53.2% girls) with a mean age of 15.53 years (SD = .58).

Measures

Excessive Internet Use (EIU)

EIU was assessed using a five-item scale based on established addiction criteria (Griffiths, 2005; Škařupová et al., 2015). The items captured five dimensions of problematic internet use, including relapse, salience, conflict, withdrawal and tolerance. Responses were recorded on a 4-point scale ranging from 1 (never) to 4 (very often). Sample items included such as “I have felt anxious when I do not get online” and “I should spend more time with my family, friends or homework, but I spend all my time online.” A composite score was calculated by averaging the five items, with higher scores indicating severe internet use status. The scale demonstrated good internal consistency (α = .824) and has shown validity across diverse European adolescent samples (Škařupová et al., 2015).

Loneliness

Subjective loneliness was assessed using a single-item measure directly asking: “How often do you feel lonely?” Response options ranged from 0 (never) to 4 (always), with higher scores indicating greater lonely feelings. This item captures individuals’ subjective experience of loneliness, whose perceived discrepancy between desired and actual social relationships (Qualter et al., 2015). Single-item loneliness measures have demonstrated adequate validity and reliability in adolescent research, particularly in large-population studies (Allen et al., 2022). Studies in school-aged children have shown that single-item and multi-item measures of loneliness yielded similar results (Eccles et al., 2020; Schütz et al., 2025; Schütz & Bilz, 2023). The single measure exhibits strong correlations with multi-item loneliness scales and predicts relevant mental health outcomes effectively.

Depressive Symptoms

The 6-item Brief Beck Depression Inventory (BDI-6) was used to measure depressive symptoms. This shortened version retains the core domains of the original 21-item BDI, including depressed mood, pessimism, dissatisfaction, guilt, self-dislike, and indecisiveness (Blom et al., 2012). BDI-6 is a number of claims about different features of moods and adolescent students were asked to select one option in each group of sentences that best describes the way they feel at the moment. Sentences in each group were rated on a 4-point scale from 0 (absent) to 3 (severe). For example, the depressed mood item presents four statements ranging from “I do not feel sad” (0) to “I am so sad or unhappy that I can’t stand it” (3). A composite score was calculated by averaging the six items, with higher scores indicating more severe depressive symptoms. The scale showed excellent internal consistency in the current sample (α = .882), supporting its reliability for assessing adolescent depressive experiences in the Finnish context.

Statistical Analysis

Data cleaning and screening were conducted using SPSS 29.0 (IBM Corp., 2022). Descriptive statistics, including means, standard deviations, and correlations among study variables, were calculated. Internal consistency reliability was assessed using Cronbach’s alpha coefficients.

The analytical approach consisted of three sequential steps designed to examine the heterogeneity of EIU patterns and their differential associations with the loneliness-depression pathway across gender groups.

Step 1: Baseline Mediation Model

A structural equation model (SEM) was first established to examine the mediating role of loneliness in the relationship between EIU and depressive symptoms. EIU and depressive symptoms were modeled as latent variables using their observed indicators to control for measurement error, while loneliness (single-item measure) were treated as observed variables. Model fit was assessed using multiple indices: comparative fit index (CFI > 0.95), Tucker-Lewis index (TLI > 0.95), root mean square error of approximation (RMSEA < 0.06), and standardized root mean square residual (SRMR < 0.05) (Hu & Bentler, 1999).

Step 2: Latent Profile Analysis (LPA)

To identify distinct subgroups of adolescents based on EIU patterns, LPA was conducted using the five EIU indicators. LPA represents a person-centered approach identifying distinct subgroups sharing similar response patterns across multiple indicators. Models with 1 to 6 profiles were systematically compared using multiple fit indices: Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), sample-size adjusted BIC (aBIC), entropy values, Vuong-Lo-Mendell-Rubin (K-1 vs. K classes) (VLMR) likelihood ratio test, and bootstrap likelihood ratio test (BLRT). The optimal number of profiles was determined by considering both statistical criteria (lower information criteria, higher entropy > 0.80, significant VLMR and BLRT) and theoretical interpretability. Average posterior probabilities for most likely profile membership were examined to ensure adequate classification accuracy (> 0.80). Each participant was assigned to their most likely profile based on posterior probabilities.

Step 3: Multi-group SEM Analysis with Latent Profiles

The final step examined the loneliness-depression mediation pathway across different EIU profiles and gender groups through multi-group mediation analyses. Using the four EIU profiles identified from the LPA, separate mediation analyses were conducted for each profile, with gender as the grouping variable in Mplus.

Given the good internal consistency of the scales, composite scores were used to reduce model complexity and ensure stable parameter estimation. For each EIU profile, gender-specific direct effects of EIU on depressive symptoms, indirect effects through loneliness, and total effects were estimated using maximum likelihood estimation with robust standard errors (MLR). All profile-gender combinations exceeded recommended minimum sample sizes for SEM analyses (n > 200). Indirect effect significance was assessed using the delta method, with 95% confidence intervals (95% CI) calculated based on asymptotic parameter estimate distributions. Gender difference testing employed Wald chi-square tests comparing parameter estimates across groups.

Data preparation was conducted using SPSS 29.0, while all main analyses (Steps 1–3) were performed using Mplus 8.11. Effect sizes for mediation effects were interpreted using Cohen’s guidelines, with standardized indirect effects of 0.01, 0.09, and 0.25 representing small, medium, and large effects, respectively (Preacher & Kelley, 2011).

Results

Descriptive Statistics and Bivariate Correlations

Table 1 shows the descriptive statistics and correlations of key variables. All correlations were statistically significant and in expected directions, with significant gender differences observed across all variables. Girls reported significantly higher levels of EIU, loneliness, and depressive symptoms than boys.

Table 1.

Correlation Matrix, Descriptive Statistics and Gender Differences.

Variable	1	2	3	Total		Boys		Girls
				M	SD	M	SD	M	SD
1 EIUᵃ	–	0.18***	0.24***	1.86	0.69	1.76	0.71	1.95	0.66
2 Lonelinessᵃ	0.18***	–	0.54***	1.18	1.03	0.85	0.93	1.47	1.03
3 Depressive Symptomsᵃ	0.22***	0.51***	–	0.39	0.57	0.22	0.45	0.53	0.61

Note. Correlation matrix: Lower triangle = Boys; Upper triangle = Girls. ᵃ Variables showing statistically significant gender differences based on ANOVA tests (p < .001). *** p < .001.

Mediation Model of EIU on Depressive Symptoms via Loneliness

The hypothesized mediation model demonstrated excellent fit to the data (RMSEA = 0.053, CFI = 0.971, TLI = 0.963, SRMR = 0.031), supporting its structural adequacy. EIU was significantly associated with higher levels of loneliness (β = .231, p < .001), and loneliness, in turn, was positively related to depressive symptoms (β = .563, p < .001). A significant direct association between EIU and depressive symptoms was also observed (β = .157, p < .001). The indirect effect via loneliness was statistically significant, accounting for approximately 45.3% of the total effect (β = .287, p < .001), indicating a partial mediation effect. These findings answer the first question.

Latent Profiles Analysis of EIU

Table 2 presents the model fit indices for models with one- to six-profile solutions. Although the AIC, BIC, and aBIC values progressively decreased with the inclusion of additional profiles, the rate of improvement diminished beyond the four-profile solution (Nylund et al., 2007). The four-profile model demonstrated excellent classification accuracy (Entropy = 0.908), well-balanced proportions, and interpretable profile distinctions, capturing meaningful variations in EIU patterns. Both VLMR and BLRT supported the four-profile model over the three-profile solution (p < .001), indicating a significant fit improvement. The four-profile model was retained based on superior classification accuracy, theoretical coherence, and model parsimony, supporting question 2.

Table 2.

Model Fit Indices for One- to Six-Profile Solution of EIU.

Profile	LL	AIC	BIC	ABIC	Entropy	BLRT p	VLMR p	Distribution (%)
1	−46,7108.725	934,237.450	934,329.244	934,297.464	–	–	-	–
2	−42,2085.509	844,203.018	844,349.889	844,299.041	0.811	< 0.001	0.3333	71.6/28.4
3	−380,839.485	761,722.970	761,924.917	761,855.001	1.000	< 0.001	0.3333	68.8/9.9/21.4
4	−362,186.930	724,429.861	724,686.884	724,597.899	0.908	< 0.001	< 0.001	9.9/23.5/45.3/21.4
5	−357,691.413	715,450.826	715,762.926	715,654.873	0.900	< 0.001	< 0.001	44.7/15.9/24.1/5.5/9.9
6	−354,871.006	709,822.012	710,189.188	710,062.067	0.872	< 0.001	< 0.001	38.1/25.9/15.6/4.8/9.9/5.7

Note. Numbers in bold indicate “best” fit. LL: Log Likelihood; AIC: Akaike Information Criterion; BIC: Bayesian Information Criterion; ABIC: Adjusted Bayesian Information Criterion; BLRT: Bootstrapped Likelihood Ratio Test (K-1 vs. K classes); VLMR: Vuong-Lo-Mendell-Rubin (K-1 vs. K classes).

Average posterior probabilities for profile membership were Profile 1 = 0.924, Profile 2 = 0.885, Profile 3 = 0.891, Profile 4 = 0.897, all exceeding the 0.80 threshold for adequate classification.

As shown in Figure 1, latent profiles showed distinct characteristics regarding internet use patterns. Profile 1 (9.9%), characterized by high levels across all five EIU indicators, especially strong compulsive use and poor control, was named as “high-risk user”. Profile 2 (23.5%) presents the “high involvement with self-control group,” who used the internet a lot (over average) but with fewer signs of loss of control or withdrawal symptoms. Profile 3 (45.3%) was the “low-risk group,” with consistently low scores across all indicators, exhibiting a healthy pattern of use. Profile 4 (21.4%), as “moderate-risk” users, exhibited relevant moderate to high levels of salience, conflict, tolerance and unsuccessful attempts to control use, indicating engagement strain.

Figure 1.

Mean scores of EIU indicators across profiles.

Multi-Group Analysis Across Latent Profiles

Multi-group mediation analyses stratified by gender within each EIU profile revealed substantial heterogeneity in pathway strengths and significance patterns (see Figure 2).

Figure 2.

Mediation pathways (from EIU to depressive symptoms via loneliness) across four profiles. Values outside path lines = boys; inside triangles = girls. Numbers in parentheses = total effects. Dashed lines = non-significant; double-lined = significant gender differences.

Across profiles, the path from loneliness to depressive symptoms was consistently strong and significant for both boys and girls, with higher effect sizes among girls, especially in Profile 4 (Moderate-risk users: β = .522 for girls vs. .465 for boys, Δβ = –.057, p < .001), indicating a gendered vulnerability to emotional distress.

In contrast, the relation between EIU and loneliness exhibited substantial variability. It was non-significant in Profile 1 (High-risk group), suggesting that in adolescents with severe EIU symptoms, loneliness may not be directly driven by internet overuse. In Profiles 2 through 4, this path was significant in both genders, though effect sizes were modest (e.g., Profile 4: β = .130 for boys; β = .097 for girls), and boys showed significantly stronger sensitivity to the correlation than girls in profile 3.

The direct effect between EIU and depressive was also profile dependent. Notably, with strong suppression effects in Profile 3 (Low-risk users), the direct effect was significantly negative for both boys (β = -.068) and girls (β = -.038), suggesting a potential compensatory or distraction effect of EIU when loneliness was accounted for. In contrast, in Profiles 1, 2, and 4, the direct effects were positive and significant.

Indirect effects were significant in Profiles 2, 3, and 4, but not in Profile 1. The strongest indirect effects were observed in Profile 4 (β = .060 for boys, .050 for girls), aligning with its elevated loneliness and depression levels. However, none of the gender differences in indirect or total effects reached statistical significance.

Discussion

This study employed a person-centered approach to examine the heterogeneous pathways linking EIU to depressive symptoms through loneliness among Finnish adolescents. Our findings make three important contributions to literature. First, we identified four distinct EIU profiles with differential mediation mechanisms. Second, we discovered unexpected suppression effects in the mediation pathway among low-risk users. Third, we found context-dependent gender differences that challenge common assumptions about adolescent digital behavior and mental health.

Profile-Dependent Mediation Mechanisms: Beyond Linear Models

Our most significant theoretical contribution lies in demonstrating that the EIU-loneliness-depression pathway operates differently across distinct usage profiles, fundamentally challenging the assumption of homogeneous relationships prevalent in variable-centered research. This finding extends beyond previous person-centered studies (Cai et al., 2023; Saputra et al., 2024; Soriano-Molina et al., 2025) by revealing not just different usage patterns, but different psychological mechanisms underlying the relationship between internet use and mental health.

Disrupted Mediation in High-risk Users

The absence of significant mediation effects in Profile 1 (High-risk users) represents a theoretically important finding that challenges conventional models of problematic internet use’s impact on depression. Despite these adolescents showing the highest levels across all addiction indicators, the expected pathway through loneliness becomes non-functional. This pattern suggests that at pathological levels of internet use, alternative mechanisms may predominate over social-emotional pathways.

This finding aligns with neurobiological addiction models suggesting that severe problematic internet use may operate through direct reward system dysfunction rather than through traditional psychological mediators (Kuss & Griffiths, 2017). The maintenance of significant direct effects despite absent mediation indicates that severe internet use may bypass typical social-emotional pathways, possibly involving dopaminergic reward systems or executive function deficits (Galvan, 2010; Reed, 2023).

Suppression Mediation: A Novel Protective Mechanism

Perhaps our most theoretically significant finding is the suppression mediation pattern observed in Profile 3 (Low-risk users), where internet use simultaneously operates through protective direct pathways and risk pathways via loneliness. This dual-mechanism pattern represents an important advancement in understanding adaptive technology use. The negative direct effects combined with positive indirect effects through loneliness suggest that low-risk internet use provides direct mood benefits, possibly through social connection, information access, or emotional regulation, while simultaneously creating conditions that could foster loneliness through reduced face-to-face interaction.

This finding supports emerging theories of adaptive technology use (Granic et al., 2014), which argue that digital interactions can foster psychological resilience and development when used purposefully. Importantly, this suppression pattern reaches equilibrium (non-significant total effects), suggesting these opposing forces naturally balance in healthy users. This contrasts sharply with previous research emphasizing primarily negative or null effects of internet use (Primack et al., 2017; Woods & Scott, 2016) and supports recent arguments for more nuanced conceptualizations of digital technology’s role in adolescent development (Odgers & Jensen, 2020). It also offers empirical evidence for the “Goldilocks hypothesis” of optimal internet use levels (Przybylski & Weinstein, 2017), though our data suggest that minimal use of digital media may be optimal for adolescent well-being.

Gender-Specific Vulnerabilities: Context-Dependent Patterns

Our findings demonstrate sophisticated gender differences that operate differentially across usage contexts. Contrary to traditional assumptions about male resilience to social-emotional consequences of technology use, we found that boys show significantly greater EIU and loneliness associations specifically in Profile 3 (Low-risk users). This context-dependent vulnerability may reflect differences in social media engagement patterns and expectations of online social connection.

This finding aligns with recent research suggesting that gender differences in internet use vary across usage contexts (Tarimo et al., 2025), and boys may be more vulnerable to social displacement effects when their usage patterns are moderate rather than pathological (Twenge & Martin, 2020). It may be that at moderate usage levels, boys experience greater discrepancy between online and offline social experiences, leading to enhanced loneliness. Alternatively, this pattern may reflect boys’ greater reliance on digital platforms for emotional expression and social connection, making them more sensitive to disruptions in these channels.

The robust finding of stronger loneliness to depression pathways among girls across Profiles 2, 3, and 4 supports established theories of female vulnerability to internalizing symptoms (Black et al., 2022; Svensson et al., 2022). Effect sizes were consistently medium to large for girls compared to boys, with the largest gender difference observed in Profile 4 (Moderate-risk users.

This pattern aligns with cognitive theories emphasizing girls’ greater tendency toward rumination and co-rumination (Rose & Rudolph, 2006), suggesting that once loneliness is experienced, girls are more likely to translate this emotional state into depressive symptoms. The consistency of this pattern across different usage profiles suggests that this vulnerability is fundamental rather than context-dependent (Dawson et al., 2023), supporting response styles theory and highlighting the importance of targeted interventions for girls experiencing loneliness.

Theoretical Implications and Mechanisms

Integration of Competing Theoretical Frameworks

Our findings provide crucial evidence for integrating rather than choosing between the social displacement and social compensation hypotheses. The suppression mediation pattern in Profile 3 demonstrates that both mechanisms can operate simultaneously within the same individuals, with the relative strength determining overall outcomes. This integration resolves long-standing theoretical tensions in the field (Kim et al., 2009; Valkenburg & Peter, 2007) by demonstrating that these frameworks are complementary rather than competing.

Furthermore, our profile-dependent findings suggest that these theoretical frameworks may operate differentially by risk level: social compensation effects may predominate in low-risk users, while social displacement effects become more prominent as usage patterns become problematic. This hierarchical integration provides a more sophisticated theoretical framework for understanding internet use's complex effects on mental health.

Complexity Theory and Non-Linear Relationships

Our findings strongly support complex theory approaches to understanding technology-mental health relationships, demonstrating that simple linear models are insufficient for capturing the nuanced reality of adolescent digital experiences. The identification of distinct profiles with fundamentally different mechanism patterns supports arguments for person-specific approaches to understanding psychological phenomena (Molenaar, 2004).

The suppression effects observed in low-risk users illustrate how non-linear systems can produce counterintuitive outcomes where moderate levels of potentially problematic behavior may offer protective benefits through alternative pathways (Ellis & Del Giudice, 2019). This finding has broader implications for understanding adolescent risk and resilience, suggesting that optimal developmental outcomes may involve moderate engagement with potentially risky activities (Przybylski & Weinstein, 2017) rather than complete avoidance.

Our study extends previous person-centered research in several important ways. While Martins et al. (2022) identified four addiction profiles based on six dimensions, they did not examine mediation mechanisms or gender differences in pathway functioning. Wang et al. (2024) found four profiles with varying psychological distress levels but did not investigate the specific mechanisms linking internet use patterns to mental health outcomes. Our study advances literature by demonstrating that profile membership not only correlates different outcome levels but operates through fundamentally different psychological mechanisms. This represents a crucial advancement from descriptive to mechanistic understanding of internet use heterogeneity.

Limitations and Future Directions

Several limitations should be acknowledged when interpreting our findings. First, cross-sectional design prevents causal inference about the directionality of relationships. While we conceptualized EIU as predicting mental health outcomes based on established theory, bidirectional relationships are likely. Future research should employ longitudinal designs with latent transition analysis to examine how individuals move between profiles over time and how mediation mechanisms evolve developmentally.

Second, while the single-item loneliness measure has demonstrated validity for large-scale studies, it may not capture the multidimensional nature of loneliness experiences. Future research would benefit from more comprehensive loneliness measurements that distinguish between social and emotional loneliness dimensions. Additionally, incorporating measures of perceived social support, friendship quality, and family relationships could provide a more nuanced understanding of social mechanisms.

Third, our findings are based on Finnish adolescents and may not generalize across cultural contexts where internet use patterns, social norms, or mental health characteristics differ. Cross-cultural replication is necessary to evaluate the stability and generalizability of the identified profiles and psychological mechanisms.

Fourth, the developmental timing of these effects remains unclear. Future research should examine whether these profiles and potential mechanisms remain stable across adolescent development or represent developmentally specific phenomena.

Conclusions

This study demonstrates that the relationship between EIU and depressive symptoms in adolescents is far more complex than previously assumed. By identifying four distinct usage profiles with fundamentally different mediation mechanisms, we challenge the field to move beyond linear, one-size-fits-all models toward more sophisticated, person-centered approaches.

Our discovery of suppression mediation effects in low-risk users provides novel evidence for the protective potential of moderate internet use, while our identification of disrupted mediation in high-risk users suggests that severe problematic use operates through fundamentally different mechanisms. The context-dependent nature of gender differences further underscores the need for nuanced, individualized approaches to understanding and addressing problematic internet use among adolescents.

These findings have immediate implications for both theory and practice, suggesting that effective interventions must be tailored to specific usage profiles and gender contexts. As digital technology continues to evolve and become even more integral to adolescent development, understanding these heterogeneous pathways becomes increasingly critical for promoting optimal mental health outcomes in the digital age.