aInternalizing and externalizing categories were not mutually exclusive.
Markers indicate estimates, with the size of the marker indicating weight; horizontal lines represent 95% CIs; diamonds represent pooled estimates, with the outer points indicating 95% CIs.
eTable 1. Search Strategy
eTable 2. Inclusion and Exclusion Criteria
eTable 3. Study Quality Evaluation Coding Criteria
eTable 4. Study Quality Scoring for Each Study Included in the Meta-analysis
eTable 5. Coding System of Study Variables
eTable 6. Study and Sample Characteristics
eFigure 1. Funnel Plot Depicting Standard Error by Fisher Z of Included Studies in Meta-analysis of Screen Time and Externalizing Problems
eFigure 2. Funnel Plot Depicting Standard Error by Fisher Z of Included Studies in Meta-analysis of Screen Time and Internalizing Problems
Customize your JAMA Network experience by selecting one or more topics from the list below.
Eirich R, McArthur BA, Anhorn C, McGuinness C, Christakis DA, Madigan S. Association of Screen Time With Internalizing and Externalizing Behavior Problems in Children 12 Years or Younger: A Systematic Review and Meta-analysis. JAMA Psychiatry. 2022;79(5):393–405. doi:10.1001/jamapsychiatry.2022.0155
Is there an association between screen time and children’s internalizing and externalizing behavior problems in the extant screen time literature?
In this systematic review and meta-analysis of 87 studies (98 independent samples) including 159 425 children 12 years or younger, greater duration of screen time was weakly but significantly correlated with externalizing (eg, aggression, inattention) and internalizing (eg, anxiety, depression) behavior problems. Results varied as a function of demographic (eg, sex) and methodological factors (eg, informant and measurement method).
The findings showed weak but significant correlations between screen time and children’s behavior problems and suggest that methodological variability may have contributed to mixed findings in the literature.
Currently, there is a lack of consensus in the literature on the association between screen time (eg, television, video games) and children’s behavior problems.
To assess the association between the duration of screen time and externalizing and internalizing behavior problems among children 12 years or younger.
For this systematic review and meta-analysis, MEDLINE, Embase, and PsycINFO databases were searched for articles published from January 1960 to May 2021. Reference lists were manually searched for additional studies.
Included studies measured screen time (ie, duration) and externalizing or internalizing behavior problems in children 12 years or younger, were observational or experimental (with baseline data), were available in English, and had data that could be transformed into an effect size. Studies conducted during the COVID-19 pandemic were excluded. Of 25 196 nonduplicate articles identified and screened for inclusion, 595 met the selection criteria.
Data Extraction and Synthesis
The study followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline. Extracted variables were child age, sex, and socioeconomic status; informants and measurement type for screen time and behavior problems; study publication year; and study design and quality. Data were extracted by 2 independent coders and were pooled using a random-effects model.
Main Outcomes and Measures
The primary outcome was the association of screen time duration with externalizing (eg, aggression, attention deficit/hyperactivity disorder symptoms) and internalizing (eg, depression, anxiety) behaviors or diagnoses.
Of the 595 full-text articles assessed for eligibility, 87 studies met all inclusion criteria, comprising 98 independent samples and 159 425 participants (mean [SD] age, 6.07 [2.89] years; 83 246 [51.30%] male). Increased duration of screen time had a small but significant correlation with more externalizing problems (90 samples; r, 0.11; 95% CI, 0.10-0.12) and internalizing problems (43 samples; r, 0.07; 95% CI, 0.05-0.08) in children. Several methodological moderators explained between-study heterogeneity. There was evidence of significant between study heterogeneity (I2 = 87.80).
Conclusions and Relevance
This systematic review and meta-analysis found small but significant correlations between screen time and children’s behavior problems. Methodological differences across studies likely contributed to the mixed findings in the literature.
The effects of screen time on children’s mental health has been rigorously debated.1,2 Some literature suggests that screen time may be associated with risk of poor mental health outcomes by displacing sleep and physical activities as well as social exchanges and learning opportunities known to foster well-being.3,4 Screen media may also impede self-regulation strategies and increase arousal levels owing to fast-paced and intense audiovisual effects, which may be associated with inattention and aggressive behavior.5,6 However, it has been argued that concern with regard to screen time and its effect on child mental health is not empirically justified owing to conflicting research results and methodological shortcomings.7-10
Meta-analyses can address discrepancies in the literature by providing greater statistical precision through pooled results from multiple individual studies11 and can detect whether variations in study findings are explained by moderators. In the screen time literature, methodological differences have likely contributed to the conflicting findings: effect sizes may differ based on the statistical analyses conducted,12 how variables are measured,13 and whether data are cross-sectional or longtudinal.7 Effect sizes may also vary as a function of child sex (eg, sex differences in the associations of screen time14,15 with behavior problems),16-18 child age (eg, association of increased screen use with increasing age among children),15,19 and socioeconomic risk (eg, association of increased screen use with behavior problems in the context of low socioeconomic status).20,21 Identifying when and for whom correlations are stronger or weaker may inform the ongoing screen time debate and help develop targeted interventions for children most at risk for potential behavior problems associated with screen time.
Owing to substantial research to date on duration of screen time and behavior problems, it appeared timely to meta-analytically summarize this body of research. Given that the context and nature of screen use changes from childhood to adolescence from mostly parent-mediated and television-centric to less parent-mediated and weighted toward social media,19,22,23 this systematic review and meta-analysis focused on children 12 years or younger. Because of their different mechanisms,24,25 we conducted 2 separate meta-analyses for externalizing (eg, aggression, attention deficit/hyperactivity disorder symptoms) and internalizing (eg, anxiety, depression) behavior problems.25 In addition, we sought to identify sources of between-study heterogeneity.
In this systematic review and meta-analysis, searches were conducted in the MEDLINE, Embase, and PsycINFO databases by a science librarian for articles published from January 1960 to May 2021. The concepts of screen time, internalizing and externalizing behavior, and children were captured by searching database-specific subject headings and text word fields (eTable 1 in the Supplement). Synonymous terms were combined with the Boolean “OR” and then combined with the Boolean “AND.” The concept of children (12 years or younger) was searched using the “age limits” database functions and via text word search. In all databases, truncation symbols were used in text word searches to capture variations in phrasing and spelling. No language limits were applied. Reference lists in included studies and review articles were manually searched for additional studies. This study followed the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guideline.
Studies were included if (1) the age range of included children was 12 years or less, (2) screen time duration was measured (hours and/or minutes), (3) behavior problems (ie, internalizing, externalizing) were measured, (4) the studies were observational or experimental (with baseline data), (5) statistical data were available, and (6) the article was available in English. Because child screen time and mental distress have increased during the COVID-19 pandemic,26,27 we excluded studies conducted during the pandemic. Inclusion and exclusion criteria are detailed in eTable 2 in the Supplement. Studies were assessed for inclusion by 2 coders (R.E. and C.A.). Any study deemed to meet inclusion criteria by either coder was considered for full-text review.
Each study was evaluated by 2 independent coders (C.A. and C.M.) for quality based on an adapted 16-item quality assessment tool28 (eTable 3 in the Supplement), with each item coded as 0 (no) or 1 (yes) (eTable 4 in the Supplement). Intercoder agreement for the overall quality score was good (intraclass correlation coefficient, 0.75). Discrepancies were resolved via consensus.
Moderators extracted were (1) child sex (percentage of males in the study); (2) child age at outcome measurement (in months); (3) screen time informant (child or parent); (4) screen time measurement method (ie, activity log, questionnaire, or interview); (5) type of internalizing (anxiety, depression, or somatization) or externalizing (aggression or attention deficit/hyperactivity disorder symptoms) behavior problems; (6) informant (child, parent, clinician or coder, peers, teacher, or combination); (7) measurement method (diagnostic and structural interview, observer report, or questionnaire); (8) whether informants differed for screen time and behavior problems; (9) clinical sample (diagnosis of a preexisting behavior problem [yes or no]); (10) publication year; (11) study design (cross-sectional, longitudinal with baseline outcome adjustment, or longitudinal with no baseline adjustment); (12) sociodemographic risk (less than 80% vs 80% or more of the sample had at least 1 of the following: low income, low caregiver educational level, or an adolescent parent); and (13) geographic location. eTable 5 in the Supplement shows the data extraction document used. All included studies were independently coded by 2 trained coders (C.A. and C.M.), with excellent reliability29 for continuous moderators (intraclass correlation coefficient, 0.80-1.00). The mean percentage of agreement for categorical moderators was 92.5% (range, 85%-100%). Discrepancies were resolved via consensus.
When multiple studies conducted analyses on the same sample, the study with the largest sample size and most comprehensive information was selected. Adjusted statistics were selected over unadjusted statistics whenever available. Global measures of internalizing or externalizing symptoms and of screen time were selected over discrete symptoms (eg, depression, anxiety) or screen types (eg, tablet, video games). Similar to the methods of other meta-analyses,30-32 when a single study measured screen time and/or behavior problems at multiple time points, effect sizes with the largest temporal distance between measures were selected. When studies reported nonsignificant findings without any corresponding statistic or P value, a P value of .50 was entered.33 Studies reporting β coefficients between –0.50 and 0.50 were imputed to correlations (r).34 Effect sizes were pooled when a single effect size among many in a study could not be selected based on the aforementioned criteria. In addition, if a study provided effect sizes from multiple discrete samples with different population parameters (eg, different cohorts), these samples were entered into the meta-analysis separately.
Pooled effect size estimates and moderator analyses were conducted using Comprehensive Meta-Analysis, version 3.0 (Biostat).35 Effect sizes were transformed into correlations (r) with 95% CIs using random-effects modeling.36 Correlations were interpreted as small (0.1), moderate (0.2), or large (0.3) based on conventional standards.37
Q and I2 statistics were used to assess between-study heterogeneity. Random-effect meta-regressions and subgroup comparisons were conducted to assess continuous and categorical moderators, respectively. Subgroup comparisons were only conducted when there were at least 3 samples (ie, ≥3 cells) available for each categorical comparison.35 Inspection of funnel plots and the Egger test were used to estimate publication bias.38 A 2-tailed P value of .05 was used for the primary analyses, and owing to the number of moderator analyses conducted, a conservative 2-sided α value of .01 was used to assess the significance of moderators.39
The search strategy revealed 25 196 nonduplicate abstracts to be reviewed for determination of meeting inclusion criteria. A total of 595 full-text articles were assessed for eligibility, and 87 studies (with 98 unique samples) were included in the meta-analysis (Figure 1).
The 98 samples consisted of 159 425 participants. Sample sizes in the included studies ranged from 15 to 15 291 participants (median, 532.5 participants). The mean (SD) age of participants when screen time was assessed was 6.07 (2.89) years (range, 0.5-11.0 years), and the mean (SD) age when behavior problems were assessed was 7.16 (2.70) years (range, 1.3-12.0 years). A total of 83 246 participants (51.25%) were male (median percentage of males per study, 51.31% [range, 36.60% to 78.70%]). Of the 98 total samples, 44 (44.9%) were in North America, 24 (24.5%) in Europe, 1 (1.0%) in Africa, 14 (14.3%) in Asia, 7 (7.1%) in Australia or New Zealand, 5 (5.1%) in the Middle East, and 2 (2.0%) in South America; 1 sample (1.0%) was multisite (detailed study characteristics are shown in eTable 6 in the Supplement).
In the 90 samples from 80 studies (124 027 children), the correlation was small but significant (r, 0.11; 95% CI, 0.10-0.12) (Figure 2).16,40-118 The Egger test result provided evidence that studies with smaller sample sizes had more extreme effect sizes, and the funnel plot showed asymmetry (eFigure 1 in the Supplement), indicating possible publication bias and/or small-study effects. There was evidence of significant between-study heterogeneity (Q, 729.78; P < .001; I2 = 87.80); therefore, moderators were explored (Table 1).
Meta-regression analyses of the 90 samples showed that the effect size for the association between screen time and externalizing problems increased as the percentage of males in studies increased (b, 0.007 [SE, 0.002]; z, 3.23; P = .001). Effect sizes also decreased as the study publication year increased (b, −0.003 [SE, 0.001]; z, –3.98; P < .001). The effect size for the association between screen time and externalizing problems decreased as study quality increased (b, −0.017 [SE, 0.004]; z, –4.30; P < .001) (Table 2).
Correlations between screen time and externalizing problems were stronger in studies examining aggression (21 samples; r, 0.17; 95% CI, 0.13-0.20) compared with those examining attention deficit/hyperactivity disorder symptoms (25 samples; r, 0.09; 95% CI, 0.06-0.11). Correlations were weaker in longitudinal studies that controlled for baseline externalizing problems (19 samples; r, 0.06; 95% CI, 0.04-0.08) compared with cross-sectional studies (50 samples; r, 0.13; 95% CI, 0.11-0.15) and longitudinal studies without baseline control (21 samples; r, 0.11; 95% CI, 0.08-0.14). Studies conducted in the Middle East had stronger associations between screen time and externalizing problems (5 samples; r, 0.23; 95% CI, 0.17-0.29) compared with those conducted in all other geographic locations. Stronger correlations were also found in studies using observer reports (8 samples; r, 0.20; 95% CI, 0.14-0.26) compared with those using questionnaires (74 samples; r, 0.10; 95% CI, 0.08-0.12). In addition, studies using peers to assess externalizing problems had stronger correlations (8 samples; r, 0.20; 95% CI, 0.15-0.26) compared with those that used child report (4 samples; r, 0.08; 95% CI, 0.00-0.15), parent report (57 samples; r, 0.10; 95% CI, 0.08-0.12), teacher report (7 samples; r, 0.07; 95% CI, 0.02-0.13), and a combination of informants (8 samples; r, 0.09; 95% CI, 0.04-0.14).
In 43 samples from 40 studies (99 603 children), the effect size between child screen time and internalizing problems was weak but significant (r, 0.07; 95% CI, 0.05-0.08) (Figure 3).16,42,43,45,50-52,56,60,66,69,73,74,76,78,81,83,84,87,88,92,94,95,98,101,105-107,110,113-116,119-125 The funnel plot showed some asymmetry (eFigure 2 in the Supplement), indicating possible publication bias; however, the Egger test result did not suggest that smaller sample sizes had more extreme effect sizes. The Q statistic was significant (Q, 285.10; P < .001; I2, 85.27), and moderator analyses were conducted to explain between-study heterogeneity (Table 1). Only 1 moderator analysis had significant results: studies that used different informants to measure internalizing problems and screen time had a stronger correlation (9 samples; r, 0.08; 95% CI, 0.05-0.10) compared with those with the same informant across measures (34 samples; r, 0.01; 95% CI, –0.02 to 0.03).
The increasing rate of screen time in early childhood has engendered concern among clinicians, policy makers, and parents regarding its possible effects on children’s mental health. This study found small but significant correlations between screen time and children’s internalizing and externalizing behavior problems. The magnitude of these correlations is comparable with that found in other meta-analyses on the association between screen time and child language skills126 and academic performance.127 Moreover, effect sizes derived in this study are similar to those in other meta-analyses examining associations between various family and child factors (eg, language skills128 and socioeconomic status129) and internalizing and externalizing problems. Although the effect sizes found in this study were small, the consequences of screen time at a population level are likely meaningful,37,130 particularly because a recent meta-analysis131 suggested that 75% of children younger than 2 years and 64% of children aged 2 to 5 years exceed screen time guidelines.
Effect sizes were larger for the association between screen time and externalizing problems compared with internalizing problems given nonoverlapping 95% CIs. Future studies testing which discrete mechanisms underlie the association between screen time and externalizing vs internalizing problems are warranted. For example, it is possible that the content viewed (eg, inappropriate or violent) underlies the association between screen time and externalizing problems, whereas indirect effects such as social withdrawal or sleep disruption may underlie the association between screen time and internalizing problems. From a methodological perspective, externalizing problems may be easier for reporters to observe than internalizing problems owing to their outward display, leading to poorer sensitivity for identifying internalizing problems.132
We found stronger correlations between screen time and externalizing problems in boys compared with girls. This finding is consistent with that of previous research in which boys had higher screen use14,133 and more externalizing problems17 than did girls. Externalizing behaviors, such as aggression, may be more readily modeled by boys through sex-stereotypic socialization.134 In addition, we found that studies measuring aggression had larger effect sizes than those measuring attention deficit/hyperactivity disorder symptoms. While viewing screen media, children may be exposed to inappropriate content, aggression, and violence.40,135,136 Consistent with social learning theory,137 children may become desensitized after repeated exposures and model aggressive or violent content toward others.6,40,136 Moreover, as screen time becomes more normalized, it is possible that aggressive behavior within some screen programming does as well. Screen use guidelines138 suggest that parents should monitor screen time, ensure the content their children are viewing is age-appropriate, limit exposure to violent content, communicate with their children about inappropriate on-screen content, and model healthy device habits.139
Correlations of screen time with externalizing behavior problems were stronger in the Middle East compared with any other geographic location; however, sample sizes in studies from the Middle East were small, and 95% CIs were wide, limiting the conclusions that can be made. A dose-response relationship between screen time and externalizing problems41,42,140 may explain the stronger correlations found in countries in the Middle East because early-onset screen use has been increasing at a faster rate in these countries compared with other countries.141,142 More research is needed to understand how cultural differences (eg, different guidelines and beliefs about screen time)143 contribute to associations between screen time and behavior problems.
The mixed findings in the literature on screen time and behavior problems may largely be the result of methodological variability.7 Studies that used more objective methods (eg, observer report) and reporters (eg, peers) of children’s externalizing problems had stronger correlations than those using child, parent, or teacher reports. In addition, studies that used different informants for screen time and internalizing problems had stronger correlations than those using the same informant across measures. Inconsistency between informants for children’s behavior problems is common,144 but informant discrepancies may indicate the contexts and ways in which behavior problems are perceived by the child and others.145 For example, aggression may be noticed more readily by peers owing to children having more opportunities to engage with peers aggressively (eg, bullying).
Variability exists both across (ie, aggression, oppositional behavior, and hyperactivity) and within (eg, demanding, hits others) the broad construct of externalizing behavior problems. However, historically, externalizing problems have been examined as a composite.25,146 This study’s finding that screen time was more strongly correlated with aggression suggests that the use of an externalizing composite may not capture inherent nuances in the association between screen time and behavior problems. Furthermore, as children move into adolescence, attention problems and aggression are often measured as separate constructs.147 Future research should consider disaggregating externalizing behaviors, presenting associations in their subcomponents (eg, aggression, attention deficit/hyperactivity disorder symptoms), and examining which items within the subcomponents are correlated most strongly with screen time.
Given that methodological variability is critical for understanding patterns of associations in this body of research, several methodological implications are suggested for future research. To the extent possible, future research should use multi-informant (eg, peers, child), multimethod (eg, passive sensing apps and questionnaires), and longitudinal approaches to measure the association of screen time with child outcomes. Of importance, self-report screen time measurement may be biased.148 None of the studies identified in our meta-analysis used objective measures (eg, sensing apps) to evaluate screen time, presenting an opportunity for future research to use novel screen-time measurement methods.
Two characteristics of publications were identified as significant moderators of associations between screen time and externalizing problems: study quality and publication year. Associations found between screen time and externalizing problems decreased as study quality increased. Poorer quality studies often have more random error and inflated effects owing to less methodological rigor (eg, a single informant, lack of statistical controls). Study quality is also closely connected to study year, with larger effect sizes found in older, discovery studies typically owing to more reliance on significance testing (P value thresholds), inadequate power, and measurement error.149 Thus, effect sizes also decreased in newer studies, which is consistent with the notion of a decline effect (ie, diminishing effect sizes over time).150,151 The accessibility and use of screens have increased over time.152 As screens become normalized in childhood and contemporary culture, the risks associated with their use may become less consequential for children’s behavior problems. Furthermore, parents today may be more informed and better at monitoring screen time and content owing to awareness of screen time guidelines.131
This study has several limitations. First and of most importance, effect sizes are correlational, not causal. Causal reductionism should be avoided because behavior problems are associated with a complex combination of genetic and environmental factors. Additional research that addresses this complexity is needed (eg, genetically informative designs).153-155 Directionality cannot be inferred; screens may be used as a tool to placate or negotiate with children.156,157 Second, although there was significant heterogeneity for internalizing problems, only 1 of the moderator analyses had significant results. Other important but unexamined moderators, such as genetic susceptibility,155 sleep,43 or language development,126 may have a significant role in associations. Third, with rapid shifts in technology, generational cohorts may not use devices comparably.7 Mobile phones were originally used almost exclusively for calling, and few children had their own; now, smartphones are used to access web pages, play games, stream videos, and socialize, and 69% of US children have their own smartphone by 12 years of age.22 Although we excluded studies conducted during the COVID-19 pandemic, a cohort of children are growing up during the pandemic. Screen time has increased significantly during the pandemic, and this increase may affect child development.26,158,159 In addition, not all screen time is equal in terms of content, subject, and formal features; some screen time may be more consequential than other screen time. Our findings only apply to screen time broadly in terms of duration or quantity of use. More nuanced aspects of screen time, such as screen content (eg, social media, violent video games), context (eg, coviewing vs passive viewing), and quality (eg, educational vs entertainment), should be examined. Future studies should also ensure that screen time is not measured as a single construct across multiple devices and should instead focus on the function of screen use (eg, entertainment, socializing, or education) to examine the way screens are being used without being device-specific.
The association between screen time and children’s mental health has garnered marked attention from academic, health, and public sectors. This systematic review and meta-analysis found that screen time was weakly but significantly correlated with children’s internalizing and externalizing behaviors. Our study identified several important methodological moderators of the association between screen time and behavior problems, such as the type of informants and the measures of screen time used, that highlight the lack of harmony in the screen time literature. It is essential for researchers to use more rigorous methods and approaches to analyzing screen time data16,160 and to further explore the various facets of screen use (ie, content and context) associated with children’s mental health in a rapidly evolving digital world.
Accepted for Publication: January 17, 2022.
Published Online: March 16, 2022. doi:10.1001/jamapsychiatry.2022.0155
Corresponding Author: Sheri Madigan, PhD, Department of Psychology, University of Calgary, Calgary, Alberta T2N 1N4, Canada (email@example.com).
Author Contributions: Ms Eirich and Dr Madigan had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Eirich, Christakis, Madigan.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Eirich, McArthur, McGuinness, Madigan.
Critical revision of the manuscript for important intellectual content: Eirich, Anhorn, McGuinness, Christakis, Madigan.
Statistical analysis: Eirich, McGuinness, Madigan.
Obtained funding: Madigan.
Administrative, technical, or material support: Eirich, Anhorn, McGuinness, Christakis, Madigan.
Supervision: Eirich, Christakis, Madigan.
Conflict of Interest Disclosures: None reported.
Funding/Support: This work was supported by the Social Sciences and Humanities Research Council and the Canada Research Chairs program.
Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Additional Contributions: Cheri Nickel, MLIS, University of Calgary, conducted the literature search for this project and received compensation.