Associations Between Problematic Internet Use and Mental Health Outcomes of Students: A Meta-analytic Review

Zhihui Cai; Peipei Mao; Zhikeng Wang; Dandan Wang; Jinbo He; Xitao Fan

doi:10.1007/s40894-022-00201-9

. 2023 Jan 28;8(1):45–62. doi: 10.1007/s40894-022-00201-9

Associations Between Problematic Internet Use and Mental Health Outcomes of Students: A Meta-analytic Review

Zhihui Cai ^1,^✉, Peipei Mao ¹, Zhikeng Wang ¹, Dandan Wang ¹, Jinbo He ^2,^✉, Xitao Fan ³

PMCID: PMC9883821 PMID: 36744121

Abstract

The increasing prevalence of problematic internet use has heightened concerns about its adverse impact on internet users’ mental health. Despite reviews investigating the associations between problematic internet use and mental health outcomes, there is a lack of understanding of various aspects of students’ mental health. This study aimed to bridge this gap by providing a more comprehensive overall picture of this issue through a quantitative synthesis based on three-level random-effects meta-analytic models. In total, 223 studies with a cumulative total of 498,167 participants and 512 effect sizes were included in this synthesis. The results showed that problematic internet use was moderately and positively associated with depressive symptoms, anxiety, loneliness, and other mental health outcomes, and negatively related to subjective well-being. Moderator analyses revealed that several study features (i.e., school grade, region, measure of problematic internet use, publication year, and gender) could explain the variations in the findings across individual studies. These research results provide solid evidence for the link between problematic internet use and different mental health outcomes and have implications for future research and interventions on students’ problematic internet use.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40894-022-00201-9.

Keywords: Problematic internet use, Depressive symptoms, Anxiety, Loneliness, Mental health outcomes, Meta-analysis

Introduction

Over the past decades, internet use has grown exponentially among all age groups, with more than 4 billion internet users worldwide as of 2021 (Internet World Stats, 2021). Along with the popularity of internet use is the increasing number of people with problematic internet use. Recent evidence shows that the estimated prevalence of problematic internet use (PIU) ranged from 20.0 to 44.6% in different regions (Chia et al., 2020; Endomba et al., 2022), and this phenomenon was found to be more prominent during the COVID-19 pandemic (Masaeli & Farhadi, 2021). Such findings have heightened concerns about the potential adverse impact of problematic internet use on various aspects of mental health, such as depressive symptoms, anxiety, loneliness, and subjective well-being (Peng et al., 2021; Shan et al., 2021; Yang et al., 2022). Given the accessibility to the internet, the lack of self-control and immaturity of physical and psychological development of students, it is imperative that researchers pay special attention to the issue as related to student population (Dong et al., 2020; Hwang et al., 2020). Thus, this study examines the overall relationship between problematic internet use and students’ mental health outcomes.

Although some meta-analytic reviews have focused on the relationship between problematic internet use and one or more aspects of mental health (e.g., Lei et al., 2020; Saadati et al., 2021, See Table 1), there are obvious deficiencies in existing meta-analytic reviews, such as a small number of primary studies included in these reviews, inconsistent inclusion criteria (e.g., sample, region), and lack of moderator analyses to explore possible factors contributing to the inconsistent findings across studies. As a result, these reviews provided us with a relatively limited and fragmented understanding about the issue. To address such research gaps, this study synthesized empirical studies examining the relationships between problematic internet use and the major mental health outcomes of students. Specifically, the present study conducted five meta-analyses, with each focusing on one of five different mental health outcomes (i.e., depressive symptoms, anxiety, loneliness, subjective well-being, and other mental health outcomes) as discussed in Tang et al. (2021), such that this study could provide a more comprehensive and robust review of this issue among students.

Table 1.

Meta-analyses on the relationships between problematic internet use and mental health

Study	Study count	Years included	Demographics	Outcome	Results
Tokunaga (2017)	247	Before 2015	All age groups	Depressive symptoms, loneliness	r = 0.53 r = 0.29
Lei et al. (2020)	70	2004–2018	Chinese students	Subjective well-being	r = − 0.313 r = − 0.21 r = − 0.183
Lozano-Blasco and Cortes-Pascual (2020)	13	2013–2019	12–18 years	Depressive symptoms	r = 0.3
Saadati et al. (2021)	16	2009–2018	All age groups	Loneliness	r = 0.26

Open in a new tab

Problematic internet use is characterized by excessive use of the internet, which may lead to significant impairment or distress (Laconi et al., 2014; Shaw & Black, 2008). Mental health, on the other hand, involves both negative (e.g., depressive symptoms, anxiety, loneliness, etc.) and positive indicators (e.g., life satisfaction, subjective well-being, etc.). Two theories about the associations between problematic internet use and mental health outcomes were discussed. According to the internet use displacement hypothesis (Nie et al., 2002), the amount of time available to an individual is more or less fixed. Time spent on the internet replaces face-to-face interaction, which could be detrimental to interpersonal relationships with friends and family in real life, thus leading to a sense of alienation. This sense of alienation, in turn, could have a negative impact on the mental health of the individual. On the other hand, however, drawing on the uses and gratification theory of Parker and Plank (2000), some researchers posited that the convenience, entertainment, anonymity, and other characteristics of the internet can satisfy the cognitive and emotional needs of students, which could effectively reduce their sense of loneliness and improve their life satisfaction (Boylu & Günay, 2019; Serin, 2011).

In general, problematic internet use was shown to have positive relationship with negative mental health indicators, and negative relationship with positive mental health indicators. But the empirical findings from many individual studies have been far from being consistent. With regard to problematic internet use’s association with negative mental health indicators (e.g., depressive symptoms, anxiety, loneliness), while relatively strong positive associations were reported in many studies (e.g., Bousono et al., 2017; Przepiorka et al., 2021; Sultana & Uddin, 2019), very weak or even non-significant associations were revealed by some other studies (e.g., Javaeed et al., 2019; Shen et al., 2021; Turan et al., 2020). With respect to problematic internet use’s relationship with positive mental health indicators (e.g., subjective well-being), similar inconsistencies were observed, with some studies reporting strong negative correlations between problematic internet use and subjective well-being (e.g., Agbaria & Bdier, 2021; Dalton & Cassidy, 2020), while there are other studies reporting close to zero or even positive associations between problematic internet use and subjective well-being (e.g., Serin, 2011; Turan et al., 2020; Zeng et al., 2021). Taken together, the empirical findings on the strengths and directions of the relationships between problematic internet use and mental health outcomes are largely inconsistent. Thus, a systematic and comprehensive meta-analytic review is warranted to further enhance our understanding of this issue.

From the individual studies, little insight can be gained about why the findings in one study differed from those in another. The research literature related to quantitative synthesis of individual studies, however, suggests that the characteristics associated with different studies (e.g., methodological and contextual variations across the studies) may often have played a role in contributing to the inconsistencies of findings across the individual studies. As a result, in a meta-analysis study, the study characteristics of the individual studies may often be considered as potential “moderators” that may partially explain the inconsistencies observed across the studies. Based on some previous meta-analytic studies and on the consideration for information availability from the empirical studies in this area, a few variables of study characteristics were selected as potential moderators that could have contributed to the inconsistency in prior findings regarding the associations between problematic internet use and mental health outcomes, as detailed below (i.e., school grade, region, measure of problematic internet use, measure of mental health, publication year, and gender).

School Grade

The associations between problematic internet use and mental health outcomes showed some variations across school grade levels. Some studies reported statistically significant, positive, and strong associations between problematic internet use and depressive symptoms for undergraduate students (e.g., r = 0.64; Dong & Li, 2020). Yet other studies show statistically non-significant associations between problematic internet use and mental health outcomes for secondary students (e.g., Cerruti et al., 2017). Similar phenomenon was observed from other empirical studies (e.g., Awasthi et al., 2020; Ha & Hwang, 2014; Yadav et al., 2013). Thus, “school grade” was proposed as a potential moderator variable in our moderator analyses. More specifically, “school grade” was coded as “secondary” for an included study with its sample consisting of students from middle or high schools, “undergraduate” if the sample consisted of college students, and “others” if the sample consisted of students from primary schools or mixed groups (i.e., both primary and secondary schools, both secondary schools and universities).

Region

Research on the relationship between problematic internet use and mental health has been carried out globally (e.g., U.S.A., China, Turkey, Korea). There has been some evidence that the prevalence rate of problematic internet use varied across different nations. For example, a meta-analysis was conducted to estimate the prevalence of problematic internet use in 31 nations and revealed that problematic internet use rates were the highest in the Middle East (10.9%) and the lowest in Northern and Western Europe (2.6%) (Cheng & Li, 2014). Past empirical studies also implied that there were regional differences in the associations of problematic internet use with mental health outcomes, such as anxiety and loneliness (Akhter et al., 2020; Vigna-Taglianti et al., 2017; Zhang et al., 2013). Thus, it would be meaningful to include region as a potential moderator. In this meta-analysis, depending on where the study was conducted, “region” was coded into five categories: Eastern Asia (e.g., China, Japan, Korea), Western Asia (e.g., Turkey, Iran, Israel), Southern Asia (e.g., India, Bangladesh), Europe (e.g., Spanish, Finland, the UK), and others (e.g., America, Chile).

Measure of Problematic Internet Use

A variety of measures has been used to assess problematic internet use. These measures were based on diverse theoretical models and differed in the dimension or diagnostic criteria (Laconi et al., 2014), which might lead to different findings across studies that used different measurement tools. In Schoenfeld and Zheng (2012), two measures (Young’s Internet Addiction Test and Caplan’s Generalized Problematic Internet Use Scale) were used to estimate the prevalence rate of problematic internet use among undergraduate students, and the study revealed that the proportion of students identified at risk for problematic internet use varied considerably between the two instruments (18.3% vs 50.4%). Similarly, a meta-analysis also showed that the prevalence of problematic internet use differed by assessment tools (Li et al., 2018). As a result, the measure of problematic internet use could be a potential moderator that might have resulted in the heterogeneity of the findings across previous studies. In this synthesis, for the included primary studies, “measure of problematic internet use” was coded as five categories to reflect the most widely used measures for problematic internet use: Internet Addiction Test, Chen Internet Addiction Scale, Young’s Diagnostic Questionnaire, Online Cognition Scale, and others.

Measures of Mental Health

Depending on the mental health outcomes of interest (e.g., depressive symptoms, anxiety), various measures could be used for evaluating the mental health outcomes. There has been some evidence that the measures for mental health outcomes used in a study could affect the associations between problematic internet use and mental health indicators in empirical studies (e.g., Cai et al., 2021; Shen et al., 2021; Zhang et al., 2021). For example, Zafar et al. (2018) indicated that there was a significant and strong positive correlation between mental health assessed by the Mental Health Inventory and problematic internet use (r = 0.73), while Mamun et al. (2020) showed a statistically significant weak relationship between psychological health evaluated by the General Health Questionnaire and problematic internet use (r = 0.14). Therefore, the measure for mental health was proposed and included as a potential moderator in the present meta-analysis study. In this regard, we coded this study feature only in the studies that assessed depressive symptoms or anxiety as the mental health outcomes, due to the limited categories of measures for other mental health outcomes. More specifically, for studies that focused on depressive symptoms and anxiety, “measure of mental health” was coded as these categories: Beck Depression Inventory, Children’s Depression Inventory, Centre for Epidemiologic Studies Depression Scale, Depression Anxiety Stress Scale, Symptom Checklist-90, Self-Rating Depression scale, Patient Health Questionnaire, Beck Anxiety Inventory, Liebowitz Social Anxiety Scale, Self-Rating Anxiety Scale, and others.

Publication Year

With the increasing popularity and rapid development of the internet, the number of internet users has been growing rapidly. In many societies, the internet has become an indispensable part of people’s daily life. The research has suggested that the prevalence rate of problematic internet use is also increasing (Modara et al., 2017; Shao et al., 2018). Moreover, research suggested that, with the society’s more dependence on the internet, the connection between internet habits and mental health, such as depressive symptoms, could be stronger than before (Tokunaga, 2017). Consequently, the research findings on the issue of problematic internet use and mental health could change over time, thus publication year of a study was considered as a potential moderator variable in the present study, and this potential moderator was coded as a continuous variable (i.e., year in which a study was published).

Gender

There is growing amount of evidence that differences can be found in internet use and the prevalence rate of problematic internet use between males and females, which may contribute to distinct psychological problems (Ha & Hwang, 2014; Lozano-Blasco & Cortes-Pascual, 2020; Ostovar et al., 2016). Some studies showed that the relationships of problematic internet use with depressive symptoms, anxiety, and subjective well-being were stronger for females than males (Baloglu et al., 2018; Ha & Hwang, 2014; Lei et al., 2020), but findings to the opposite (Lozano-Blasco & Cortes-Pascual, 2020), as well as findings of no difference between gender groups (Tokunaga, 2017), also existed. With regard to these inconsistent findings, it is prudent to consider the potential moderating effect of gender in the relationship between problematic internet use and mental health outcomes in this study. Operationally, the proportion of females in the study sample was used in the moderator analysis in this meta-analysis.

Current Study

Many studies have been conducted to investigate the relationship between problematic internet use and mental health, and considerable inconsistencies in the findings have emerged across the individual studies. Such a situation calls for a comprehensive synthesis of the relevant individual studies for the purpose of developing a better understanding of the overall relationship between problematic internet use and mental health outcomes, as well as a better understanding of how and why the findings of the individual studies varied. This meta-analysis aimed to estimate the overall relationships between problematic internet use and different mental health outcomes (i.e., depressive symptoms, anxiety, loneliness, subjective well-being, and other mental health outcomes) of students. Then, moderator analyses were conducted to explore whether some study characteristics (i.e., school grade, region, measure of problematic internet use, measure of mental health, publication year, and gender) could have contributed to the observed inconsistent findings across the individual studies.

Methods

Literature Search

This study used five online databases, namely Web of Science, PubMed, PsycINFO, ProQuest and Scopus, and systematically searched for the research literature related to the relationship between PIU and mental health outcomes. The terms used were: (“internet addiction” OR “pathological internet use” OR “problematic internet use” OR “internet dependency” OR “internet disorder” OR “compulsive internet use” OR “excessive internet use” OR “internet overuse”) AND (“mental health” OR “anxiety” OR “loneliness” OR “depression” OR “subjective well-being” OR “life satisfaction” OR “stress” OR “psychopathology” OR “suicide” OR “aggression” OR “unhappiness” OR “positive affect” OR “negative affect”). Furthermore, relevant studies were also identified by using Google Scholar, and those from the reference lists in the included studies, from previous reviews or meta-analyses. The search covered the research literature up to February 2022.

Inclusion and Exclusion Criteria

To be included in this meta-analysis, a study met the following criteria:

The study examined the association between PIU and mental health outcomes (e.g., depressive symptoms, anxiety, loneliness, subjective well-being, suicide, aggression, hostility, stress, etc.).
The PIU measurement used in a primary study assessed generalized PIU rather than specific subtype of PIU (e.g., internet gaming disorder, smartphone addiction, etc.). For measuring generalized PIU, the items focus on the overall internet use (e.g., “How often do you choose to spend your time online rather than going out with others?” and “How often do you spend time online when you’d rather sleep?”), rather than a specific aspect of the network use (e.g., “I have slept less than 4 h more than one time due to using smartphone.” and “I find that I am becoming more and more dependent on smartphone longer and longer.”).
The study sample consisted of students from elementary, middle, high schools, and colleges.
The study reported zero-order correlation r between PIU and mental health outcome, or provided other statistical information (e.g., t, χ², F values), to allow us to derive the correlation between PIU and mental health outcome.
This meta-analysis focused on the studies involving normal population. Studies involving clinical populations were excluded. For example, the study by Alpaslan et al. (2016) involved adolescents with major depressive disorders, and such or similar studies were not included in this meta-analysis.
If a study used a longitudinal design to record the data at different time points, we only used the information of the first measurement. For instance, in the longitudinal study by Ciarrochi et al. (2016), the relationship between compulsive internet use and mental health was measured four times at Grade 8, 9, 10 and 11, respectively. Only the results measured at Grade 8 were used in our meta-analysis.
The study was published in an English academic journal.

Study Selection

The selection procedure was described graphically in Fig. 1. More specifically, in the initial search, 4475 non-duplicate studies were obtained. After screening all titles and abstracts, 468 studies were identified as being relevant for further consideration. Two authors of this study read these articles carefully based on the aforementioned inclusion criteria, and 245 studies were excluded due to various reasons (e.g., no sufficient data, no PIU or mental health outcomes, reviews). Then, the authors of 68 studies that did not report the zero-order correlation between PIU and mental health in their articles were contacted to request this information, and only six responses were received to provide us with the data. Eventually, a total of 223 studies published from 2002 to 2022 were included in the present meta-analysis. The characteristics and citations of all studies were presented in Online Appendix A and Online Appendix C respectively.

Coding Process

Two authors of this study coded the primary studies. They first independently coded 30% of the 223 studies, and verified the coding information together. The inter-rater reliability was assessed by intraclass correlation coefficients (ICC) for continuous variables and Cohen’s Kappa (κ) for categorical variables. Specifically, the internal consistency values of moderator variables were: school grade (κ = 0.974), region (κ = 1.000), measure of PIU (κ = 1.000), measure of mental health (κ = 0.948), publication year (ICC = 0.999), and gender (ICC = 0.998). These results showed high inter-rater reliability among the coders.

Quality Appraisal

Davids et al. (2017) presented the Methodological Quality Appraisal Tool for evaluating the quality of non-randomized, cross-sectional studies in psychology or other social sciences. This measure was adapted in recent meta-analytic reviews for the purpose of assessing the quality of included studies in a systematic review (e.g. He et al., 2020a, 2020b). In this study, the same approach for quality appraisal of the included studies was used. Specifically, this tool includes 6 items, reflecting five domains of study quality (sampling methods, measurement tool, the data sources used, whether PIU or mental health outcomes were examined, and whether the relationship between PIU and mental health was discussed). Each of the six items was rated as either “yes” or “no” and assigned a score of “1” or “0”, respectively. The quality appraisal score was then obtained by dividing the total score by total number of items multiplied by 100 (expressed as a percentage). As discussed previously (e.g., Davids et al., 2017; Roman & Frantz, 2013), appraisal scores of 0–33%, 34–66%, and 67–100% indicate bad, satisfactory, and good methodological quality, respectively. The quality assessment was conducted by two authors independently and discrepancies were resolved through discussion between the two evaluators.

Data Analysis

As mentioned earlier, mental health is multi-faceted, and different studies covered different aspects of mental health. In the current meta-analysis, studies that covered negative (depressive symptoms, anxiety, loneliness, suicide, aggression, hostility, stress, etc.) and positive (subjective well-being) indicators of mental health were included. These five categories of mental health outcomes (i.e., depressive symptoms, anxiety, loneliness, subjective well-being, and other mental health outcomes) were treated as different outcomes, and a meta-analysis was conducted for each of these five mental health outcomes. So essentially, the current synthesis study consists of five meta-analyses.

In our meta-analysis, the zero-order correlation coefficient rs from the included studies between PIU and a specific mental health outcome were treated as effect sizes. When studies reported other types of statistical results, those were transformed into correlation coefficients based on well-documented statistical conversion formulas (Yap & Jorm, 2014). For a negative mental health outcome (e.g., depressive symptoms), a positive r indicated that the higher level of PIU was associated with more serious mental health problems. For the positive mental health outcome (i.e., subjective well-being), a negative r indicated that the higher level of PIU was associated with lower level of subjective well-being, thus more serious mental health issue. Since correlation coefficients were not normally distributed, all effect sizes were initially converted to Fisher’s z-scores for computations. After all the computations, the Fisher’s z-scores were transformed back to correlation coefficients for easier interpretation.

The traditional approach to meta-analysis assumes that the effect sizes are independent of each other. However, some studies included in our meta-analysis reported multiple effect sizes based on the same sample and these effect sizes were related. Therefore, a three-level meta-analysis was performed to deal with interdependency of effect sizes (Assink & Wibbelink, 2016). In the current meta-analysis, three different sources of variances were distinguished in the three-level meta-analytic model: sampling variance at level 1; variance across effect sizes within the same study at level 2; variance across effect sizes between studies al level 3 (Cheung, 2014). The appropriateness and the advantages of using this meta-analytic model over traditional approach for such data have been discussed and supported in the literature (Assink & Wibbelink, 2016). All statistical analyses were conducted by using Viechtbauer’s (2010) metafor package in the R statistical platform.

Assessment for Publication Bias

Analysis for potential publication bias was conducted as preliminary analysis prior to meta-analysis. In meta-analysis research literature, publication bias was assumed that studies reporting statistically significant results were easier to get published than those with statistically non-significant findings, and this is also called “file-drawer problem” (Rosenthal, 1979). This assumption, if correct, may lead to estimation error for the true effect size. With this consideration, several widely accepted methods in meta-analysis research literature were used to assess the likelihood that such a problem existed in our data in the current meta-analysis. First, the funnel plot was applied to estimate the presence or absence of publication bias. If effect sizes of studies were symmetrically distributed on both sides of the mean effect sizes, the absence of publication bias would be supported (Borenstein et al., 2009). Moreover, the trim-and-fill method was also used to assess this problem (Duval & Tweedie, 2000), as proposed by Fernández-Castilla et al. (2021), to detect the potential publication bias.

Results

Characteristics of Included Studies

In total, 223 studies involving 498,167 participants met the inclusion criteria and 512 effect sizes were obtained in the present meta-analysis. More specifically, for depressive symptoms, there were 131 studies with 159 effect sizes that involved a cumulative total of 336,642 participants. For anxiety, there were 67 studies with 91 effect sizes that involved a cumulative total of 92,205 participants. For loneliness, there were 56 studies with 69 effect sizes that involved a cumulative total of 79,921 participants. For subjective well-being, there were 29 studies with 38 effect sizes that involved a cumulative total of 208,968 participants. Finally, for other mental health outcomes, there were 76 studies with 155 effect sizes that involved a cumulative total of 217,028 participants.

Quality Appraisal

The methodological appraisal scores of the included studies ranged from 50 to 100%, indicating overall adequate quality for the included studies. The details of the quality evaluation of each study were presented in Online Appendix B.

Overall Analysis

The overall analysis results for the relationships between PIU and mental health outcomes are shown in Table 2. More concretely, overall, the four negative mental health outcomes (depressive symptoms, anxiety, loneliness, and other mental health outcomes) were all positively associated with PIU (r = 0.313, p < 0.001; r = 0.277, p < 0.001; r = 0.252, p < 0.001; r = 0.289, p < 0.001, respectively). The positive mental health outcome, subjective well-being, was negatively associated to PIU with similar magnitude (r = − 0.213, p < 0.001). These summary findings indicated that higher levels of PIU were associated with more serious mental health issues (depressive symptoms, anxiety, loneliness, and other mental health outcomes), and with lower level of subjective well-being. According to Cohen’s (1992) guidelines that the r value of 0.1, 0.3, and 0.5 refer to small, medium, and large effects respectively, these overall meta-analysis findings indicated moderate relationships between PIU and mental health outcomes across many individual studies.

Table 2.

Results for the overall analysis of the relation between problematic internet use and mental health outcomes

Type of outcomes	No. studies	No. ES	Mean r	95% CI	t value (sig)	% Var. at level 1	Level 2 variance	% Var. at level 2	Level 3 variance	% Var. at level 3
Depressive symptoms	131	159	0.313	(0.299, 0.348)	26.155***	2.449	0.003***	15.088	0.016***	82.463
Anxiety	67	91	0.277	(0.252, 0.316)	17.676***	4.851	0.001*	5.806	0.015***	89.343
Loneliness	56	69	0.252	(0.218, 0.298)	12.899***	4.133	0.004*	16.700	0.017*	78.867
Subjective well-being	29	38	− 0.213	(− 0.270, − 0.163)	− 8.203***	0.833	0.025***	99.167	0.000	0.000
Other mental health outcomes	76	155	0.289	(0.262, 0.331)	16.895***	1.959	0.003***	13.252	0.021***	84.789

Open in a new tab

*p < 0.05; **p < 0.01; ***p < 0.001

The results of the likelihood-ratio tests for heterogeneity across the effect sizes revealed that there was statistically significant within-study effect size heterogeneity (i.e., statistically significant level-2 variances) or statistically significant between-study effect size heterogeneity (i.e., statistically significant level-3 variances) for the five mental health outcomes (i.e., depressive symptoms, anxiety, loneliness, subjective well-being, and other mental health outcomes). Thus, we conducted moderator analyses to explore whether some study characteristics could have moderated the association between PIU and mental health outcomes for these five domains (i.e., depressive symptoms, anxiety, loneliness, subjective well-being, and other mental health outcomes).

Moderator Analyses

Depressive Symptoms

As shown in Table 3, across the studies that examined the relationship between depressive symptoms and PIU, three study characteristics appeared to be moderators for the heterogeneity of the effect sizes: school grade (F_{(2, 156)} = 5.054, p = 0.007), region (F_{(4, 154)} = 2.668, p = 0.034), and publication year (F_{(1, 157)} = 5.482, p = 0.020). More specifically, this association was stronger in the studies involving samples of undergraduate students (r = 0.353) than in those involving samples of secondary school students (r = 0.281). For region as a moderator, this relationship appeared to be weaker in the studies involving samples in East Asia (r = 0.282) than those in other regions. Additionally, the moderation analysis result of publication year showed that the magnitude of the “PIU-depressive symptoms” association became stronger over time. No moderating effects were found for other coded moderator variables (measure of PIU, measure of depressive symptoms, and gender).

Table 3.

Results for the moderator analysis of the relation between problematic internet use and depressive symptoms

Moderator variables	No. studies	No. ES	Mean r	95% CI	t value (sig)	F (df1, df2)	Level 2 variance	Level 3 variance
School grade	131	159				F (2, 156) = 5.054**	0.003	0.015
Secondary	72	89	0.281	(0.257, 0.321)	17.849***
Undergraduate	50	59	0.353	(0.330, 0.408)	18.705***
Others	10	11	0.331	(0.262, 0.426)	8.314***
Region	131	159				F (4, 154) = 2.668*	0.003	0.015
East Asia	66	74	0.282	(0.257, 0.323)	17.476***
West Asia	30	37	0.335	(0.296, 0.399)	13.355***
South Asia	10	10	0.327	(0.248, 0.433)	7.257***
Europe	20	27	0.377	(0.335, 0.457)	12.839***
Others	5	11	0.322	(0.212, 0.456)	5.422***
Measure of PIU	131	159				F (3, 155) = 2.382	0.003	0.016
IAT	64	74	0.330	(0.308, 0.378)	19.380***
CIAS	14	15	0.230	(0.162, 0.307)	6.414***
YDQ	12	13	0.315	(0.247, 0.406)	8.109***
Others	41	57	0.314	(0.283, 0.367)	15.235***
Measure of depressive symptoms	131	159				F (7, 151) = 1.075	0.003	0.016
BDI	27	35	0.316	(0.272, 0.382)	11.739***
CDI	5	6	0.246	(0.135, 0.367)	4.271***
CES-D	28	28	0.342	(0.303, 0.409)	13.270***
DASS-21	16	18	0.335	(0.277, 0.420)	9.622***
SCL-90	6	6	0.333	(0.226, 0.467)	5.677***
SDS	5	6	0.264	(0.150, 0.391)	4.442***
PHQ	9	9	0.359	(0.281, 0.471)	7.841***
Others	35	51	0.283	(0.246, 0.337)	12.719***
Publication year	131	159	0.007	(0.001, 0.012)	2.341*	F (1, 157) = 5.482*	0.003	0.016
Female	123	150	− 0.017	(− 0.167, 0.133)	− 0.230	F (1, 148) = 0.053	0.003	0.017

Open in a new tab

IAT Internet Addiction Test; CIAS Chen Internet Addiction Scale; YDQ Young’s Diagnostic Questionnaire; BDI Beck Depression Inventory; CDI Children’s Depression Inventory; CES-D Centre for Epidemiologic Studies Depression Scale; DASS-21 Depression Anxiety Stress Scale; SCL-90 Symptom Checklist-90; SDS Self-Rating Depression scale; PHQ Patient Health Questionnaire

*p < 0.05; **p < 0.01; ***p < 0.001

Anxiety

Table 4 showed the moderator analysis results for anxiety, and only measure of PIU appeared to be a moderator for the heterogeneity of the effect sizes (F_{(3, 87)} = 2.904, p = 0.039). More precisely, the association between PIU and anxiety was the strongest when PIU was assessed with YDQ (r = 0.377). Other moderators (school grade, region, measure of anxiety, publication year, gender) appeared not to have accounted for any statistically significant amount of between-study effect-size heterogeneity.

Table 4.

Results for the moderator analysis of the relation between problematic internet use and anxiety

Moderator variables	No. Studies	No. ES	Mean r	95% CI	t value (sig)	F (df1, df2)	Level 2 variance	Level 3 variance
School grade	67	91				F (2, 88) = 0.432	0.001	0.016
Secondary	41	56	0.276	(0.241, 0.324)	13.494^***
Undergraduate	23	32	0.272	(0.226, 0.332)	10.405^***
Others	3	3	0.344	(0.194, 0.525)	4.322^***
Region	67	91				F (4, 86) = 0.516	0.001	0.016
East Asia	27	41	0.259	(0.215, 0.314)	10.654^***
West Asia	15	22	0.304	(0.247, 0.381)	9.259^***
South Asia	6	6	0.250	(0.137, 0.372)	4.313^***
Europe	15	18	0.283	(0.219, 0.362)	8.116^***
Others	4	4	0.316	(0.192, 0.462)	4.820^***
Measure of PIU	67	91				F (3, 87) = 2.904^*	0.001	0.014
IAT	31	36	0.282	(0.244, 0.337)	12.375^***
CIAS	10	19	0.205	(0.136, 0.280)	5.739^***
YDQ	6	8	0.377	(0.287, 0.504)	7.232^***
Others	20	28	0.282	(0.235, 0.346)	10.341^***
Measure of anxiety	66	90				F (5, 84) = 0.893	0.001	0.015
BAI	6	6	0.261	(0.166, 0.368)	5.252^***
DASS-21	15	17	0.319	(0.260, 0.401)	9.346^***
LSAS	5	8	0.300	(0.206, 0.415)	5.907^***
SCL-90	6	10	0.305	(0.207, 0.424)	5.778^***
SAS	5	5	0.200	(0.081, 0.326)	3.293^**
Others	32	44	0.266	(0.227, 0.320)	11.724^***
Publication year	67	91	0.001	(-0.007, 0.009)	0.344	F (1, 89) = 0.119	0.001	0.015
Female	66	90	-0.112	(-0.287, 0.063)	-1.277	F (1, 88) = 1.629	0.001	0.016

Open in a new tab

IAT Internet Addiction Test; CIAS Chen Internet Addiction Scale; YDQ Young’s Diagnostic Questionnaire; BAI Beck Anxiety Inventory; DASS-21 Depression Anxiety Stress Scale; LSAS Liebowitz Social Anxiety Scale; SCL-90 Symptom Checklist-90; SAS Self-Rating Anxiety Scale

*p < 0.05; **p < 0.01; ***p < 0.001

Loneliness

The results of the moderator analyses of the connection between PIU and loneliness were shown in Table 5. Two study features turned out to be statistical moderators for the effect size heterogeneity across the studies that examined the relationship between PIU and loneliness: region (F_{(4, 64)} = 2.807, p = 0.033) and gender (F_{(1, 64)} = 5.180, p = 0.026). The averaged effect size (r = 0.174) from the studies involving samples in West Asia, and that (r = 0.409) from studies involving samples in South Asia were the smallest and the largest respectively, with the effect sizes conducted in other regions being somewhere in-between. For gender (operationalized as the proportion of females in a study sample) as a moderator, the studies with larger proportions of females showed weaker association between PIU and loneliness, and this observation is to be discussed later. Other coded moderator variables (school grade, measure of PIU, publication year) showed no statistical moderation effects.

Table 5.

Results for the moderator analysis of the relation between problematic internet use and loneliness

Moderator variables	No. Studies	No. ES	Mean r	95% CI	t value (sig)	F (df1, df2)	Level 2 variance	Level 3 variance
School grade	56	69				F (2, 66) = 0.548	0.004	0.017
Secondary	24	25	0.234	(0.176, 0.300)	7.701***
Undergraduate	28	40	0.262	(0.211, 0.325)	9.383***
Others	4	4	0.304	(0.165, 0.463)	4.219***
Region	56	69				F (4, 64) = 2.807*	0.004	0.015
East Asia	14	14	0.290	(0.224, 0.373)	7.989***
West Asia	15	19	0.174	(0.102, 0.250)	4.748***
South Asia	4	4	0.409	(0.274, 0.594)	5.415***
Europe	13	16	0.243	(0.171, 0.325)	6.415***
Others	10	16	0.273	(0.190, 0.370)	6.228***
Measure of PIU	56	69				F (2, 66) = 0.399	0.004	0.017
IAT	21	22	0.268	(0.209, 0.342)	8.278***
OCS	5	5	0.204	(0.065, 0.349)	2.912**
Others	30	42	0.250	(0.200, 0.309)	9.337***
Publication year	56	69	0.001	(− 0.007, 0.010)	0.258	F (1, 67) = 0.067	0.004	0.017
Female	53	66	− 0.375	(− 0.740, -0.048)	− 2.276	F (1, 64) = 5.180*	0.004	0.015

Open in a new tab

IAT Internet Addiction Test; OCS Online Cognition Scale

*p < 0.05; **p < 0.01; ***p < 0.001

Subjective Well-Being

The results of the moderator analyses for subjective well-being were presented in Table 6, and only publication year was found to be a significant moderator (F_{(1, 36)} = 10.355, p = 0.003). The correlation between PIU and subjective well-being became smaller over time. No moderating effects were found for other coded moderator variables (school grade, region, measure of PIU, gender).

Table 6.

Results for the moderator analysis of the relation between problematic internet use and subjective well-being

Moderator variables	No. studies	No. ES	Mean r	95% CI	t value (sig)	F (df1, df2)	Level 2 variance	Level 3 variance
School grade	29	38				F (2, 35) = 0.201	0.026	0.000
Secondary	16	21	− 0.204	(− 0.279, − 0.134)	− 5.785***
Undergraduate	11	15	− 0.233	(− 0.327, − 0.147)	− 5.357***
Others	2	2	− 0.177	(− 0.411, 0.053)	− 1.568
Region	29	38				F (3, 34) = 0.462	0.025	0.001
East Asia	10	12	− 0.176	(− 0.274, − 0.081)	− 3.751***
West Asia	11	15	− 0.217	(− 0.311, − 0.130)	− 4.953***
Europe	3	3	− 0.280	(− 0.480, − 0.096)	− 3.048**
Others	5	8	− 0.235	(− 0.362, − 0.116)	− 3.936***
Measure of PIU	29	38				F (2, 35) = 2.116	0.023	0.000
IAT	10	13	− 0.266	(− 0.364, − 0.182)	− 6.074***
OCS	3	5	− 0.101	(− 0.246, 0.044)	− 1.409
Others	16	20	− 0.207	(− 0.280, − 0.139)	− 6.033***
Publication year	29	38	− 0.020	(− 0.033, − 0.007)	− 3.218**	F (1, 36) = 10.355**	0.019	0.000
Female	27	36	− 0.081	(− 0.367, 0.205)	− 0.576	F (1, 34) = 0.332	0.026	0.000

Open in a new tab

IAT Internet Addiction Test; OCS Online Cognition Scale

*p < 0.05; **p < 0.01; ***p < 0.001

Other Mental Health Outcomes

As shown in Table 7, region (F_{(4, 150)} = 4.321, p = 0.002) and publication year (F_{(1, 153)} = 5.385, p = 0.022) showed statistically significant moderating effects on the heterogeneity of the effect sizes from the studies that examined the relationship between PIU and other mental health outcomes. More specifically, larger effect sizes were found in the studies involving samples from Europe (r = 0.356), South Asia (r = 0.352), and West Asia (r = 0.344) than in the studies involving samples from East Asia (r = 0.217) and “Others” (r = 0.270). The moderating effect of publication year indicated that the relationship between PIU and other mental health outcomes was stronger in more recent studies. Other coded study features (school grade, measure of PIU, gender) did not show statistical effect on the heterogeneity of the effect sizes across the studies.

Table 7.

Results for the moderator analysis of the relation between problematic internet use and other mental health outcomes

Moderator variables	No. Studies	No. ES	Mean r	95% CI	t value (sig)	F (df1, df2)	Level 2 variance	Level 3 variance
School grade	76	155				F (2, 152) = 0.466	0.003	0.021
Secondary	40	89	0.275	(0.235, 0.329)	11.832***
Undergraduate	30	58	0.303	(0.256, 0.370)	10.898***
Others	6	8	0.316	(0.197, 0.458)	4.951***
Region	76	155				F (4, 150) = 4.321**	0.003	0.017
East Asia	28	71	0.217	(0.170, 0.272)	8.629***
West Asia	21	40	0.344	(0.297, 0.422)	11.343***
South Asia	9	9	0.352	(0.268, 0.468)	7.297***
Europe	9	17	0.356	(0.276, 0.468)	7.670***
Others	9	18	0.270	(0.191, 0.362)	6.387***
Measure of PIU	76	155				F (4, 150) = 2.315	0.003	0.019
IAT	34	57	0.293	(0.251, 0.353)	11.755***
YDQ	3	8	0.449	(0.309, 0.658)	5.470***
CIAS	11	22	0.227	(0.145, 0.316)	5.330***
OCS	3	6	0.404	(0.255, 0.602)	4.875***
Others	25	62	0.277	(0.226, 0.342)	9.698***
Publication year	76	155	0.009	(0.001, 0.017)	2.321*	F (1, 153) = 5.385*	0.003	0.019
Female	70	149	− 0.007	(− 0.221, 0.207)	− 0.063	F (1, 147) = 0.004	0.003	0.018

Open in a new tab

IAT Internet Addiction Test; YDQ Young’s Diagnostic Questionnaire; CIAS Chen Internet Addiction Scale; OCS Online Cognition Scale

*p < 0.05; **p < 0.01; ***p < 0.001

Publication Bias

As can be seen in Figs. 2, 3, 4, 5 and 6, the funnel plots of the five mental health outcomes (i.e., depressive symptoms, anxiety, loneliness, subjective well-being, other mental health outcomes) were basically symmetrically distributed, suggesting little evidence for publication bias. Furthermore, as suggested by Fernández-Castilla et al. (2021) for trim-and-fill method, the optimal values for indicating publication bias would vary based on the magnitude of effect sizes (r = 0.213 to 0.313 in this review) and the number of effect sizes (38 to 159 in this review) included in an analysis. Under our data conditions, for “depressive symptoms” and “other mental health outcomes”, the publication bias would be suggested if $L_{0}^{+}$ is greater than 3. For “anxiety”, “loneliness”, and “subjective well-being”, the publication bias would be suggested if $L_{0}^{+}$ is greater than 2. Our results of the trim-and-fill method for depressive symptoms ( $L_{0}^{+} = 0)$ , anxiety ${(L}_{0}^{+} = 0)$ , loneliness ${(L}_{0}^{+} = 0)$ , subjective well-being ( $L_{0}^{+} = 0.347)$ and other mental health outcomes ${(L}_{0}^{+} = 0)$ indicated a lack of evidence for publication bias.

Fig. 2 — Funnel plot for the association between problematic internet use and depressive symptoms

Fig. 3 — Funnel plot for the association between problematic internet use and anxiety

Fig. 4 — Funnel plot for the association between problematic internet use and loneliness

Fig. 5 — Funnel plot for the association between problematic internet use and subjective well-being

Fig. 6 — Funnel plot for the association between problematic internet use and other mental health outcomes

Discussion

Despite a growing body of research suggesting that problematic internet use is associated with various mental health problems, there has been a lack of understanding about the similarities and differences between problematic internet use and different mental health indicators. Therefore, this study conducted five meta-analyses to estimate the overall associations between problematic internet use and each of five mental health outcomes for students of different grades; and it also sought to examine whether certain study characteristics could have contributed to the heterogeneity of the findings across previous individual studies. The results indicated that higher levels of problematic internet use were related to higher levels of depressive symptoms, anxiety, loneliness, and other negative mental health outcomes, but to lower levels of subjective well-being. The strength of such relationships between problematic internet use and mental health outcomes were moderated by school grade, region, measure of problematic internet use, publication year, and gender.

Overall Associations Between PProblematic Internet Use and Mental Health Outcomes

The results of three-level meta-analysis showed moderate positive correlations between problematic internet use and depressive symptoms (r = 0.313), anxiety (r = 0.277), loneliness (r = 0.252), other mental health outcomes (r = 0.289) and a moderate negative relation with subjective well-being (r = − 0.213). In other words, despite the considerable variations in the findings across the many studies included in this meta-analysis, overall, individuals who reported higher levels of problematic internet use were more likely to have reported higher levels of negative mental health problems, and to have reported lower levels of subjective well-being, a positive indicator of mental health. These findings were generally consistent with previous studies in this area and provided evidence for the internet use displacement hypothesis (e.g., Cikrikci, 2016; Lei et al., 2020; Lozano-Blasco & Cortes-Pascual, 2020; Tokunaga, 2017). At the same time, some studies pointed out that the negative mental health problems might be potential risk factors for problematic internet use (Cikrikci, 2019; Liang et al., 2016; Lu et al., 2017). Individuals suffering from mental health issues are more inclined to use the internet to alleviate their negative emotions. The psychological dependence on the Internet and maladaptive cognition of the virtual world led them to spend more time on the internet. Additionally, similar to the findings in some previous studies (Carli et al., 2013; Cheung et al., 2018; Erceg et al., 2018; Li et al., 2019), the association between problematic internet use and depressive symptoms appeared to be relatively strong and robust. This result has important implications for future practice.