A Meta-analysis of Interventions That Target Children's Screen Time for Reduction
- Dayna M. Maniccia, DrPH, MSa,
- Kirsten K. Davison, PhD, MSa,
- Simon J. Marshall, PhD, MAb,
- Jennifer A. Manganello, PhD, MPHa, and
- Barbara A. Dennison, MDc,d
- Departments of aHealth Policy, Management, and Behavior and
- cEpidemiology, School of Public Health, University at Albany, State University of New York, Albany, New York;
- bSchool of Exercise and Nutritional Sciences, San Diego State University, San Diego, California; and
- dDivision of Chronic Disease and Injury Prevention, New York State Department of Health, Albany, New York
BACKGROUND: Screen time, especially television viewing, is associated with risk of overweight and obesity in children. Although several interventions have been developed to reduce children's screen time, no systematic review of these interventions exists to date.
OBJECTIVE: This is a systematic review and meta-analysis of interventions targeting a reduction in children's screen time.
METHODS: Effect sizes and associated 95% confidence intervals (CIs) were calculated by using a random-effects model. Heterogeneity tests, moderator analyses, assessment of bias, and sensitivity analyses were conducted. Reliability was assessed with Cohen's κ.
RESULTS: The systematic search identified 3002 documents; 33 were eligible for inclusion, and 29 were included in analyses. Most reported preintervention and postintervention data and were published in peer-reviewed journals. Although heterogeneity was present, no moderators were identified. Overall Hedges g (−0.144 [95% CI: −0.217 to −0.072]) and standard mean difference (SMD) (−0.148 [95% CI: −0.224 to −0.071]) indicated that interventions were linked with small but statistically significant reductions in screen time in children. The results were robust; the failsafe N was large, and the funnel plot and trim-and-fill methods identified few missing studies.
CONCLUSIONS: Results show that interventions to reduce children's screen time have a small but statistically significant effect. As the evidence base expands, and the number of screen-time interventions increases, future research can expand on these findings by examining the clinical relevance and sustainability of effects, conducting a more thorough analysis of effect modifiers, and identifying critical components of effective interventions.
- mass media
Although the American Academy of Pediatrics recommends no more than 2 hours/day of screen time (watching television or videos/DVDs, playing video or computer games, and using a computer for purposes other than school work) for children aged 2 years old and older,1,2 47% of children aged 2 to 15 years spend 2 or more hours/day using screen media.3 A recent study4 found that approximately two-thirds of a national sample of 3-year-old children watched ∼3 hours/day of television and were exposed to more than 5 additional hours of indirect (ie, in the background) television daily. Contrary to the American Academy of Pediatrics recommendations, 33% of children aged 6 years and younger5 and 71% of children aged 8 to 18 years6 have televisions in their bedrooms, a characteristic associated with increased viewing.5,–,7
Time spent watching television is associated with a number of negative health behaviors and outcomes among children, including overweight,8,–,11 irregular sleep,12 insufficient consumption of fruits and vegetables,13 and disordered eating.14,15 Excessive television viewing has detrimental effects on prosocial behaviors, such as spending time with parents and siblings, doing homework, or engaging in creative play,16 and is linked with getting lower grades, getting into trouble, and feeling sadness and boredom.6 In addition, children who watch more television may be exposed to more content that negatively influences healthy behavior. In particular, television and video-game violence is associated with violence and aggression4,17 in thought and deed, and exposure to tragedies and disasters on television negatively impacts children's mental health.18
Before 1985, when Dietz and Gortmaker19 published their seminal article linking television viewing to obesity, the majority of the research on media effects on children focused on the impact of television content or media policy on children's behavior.20,–,27, 28, 29,–,41 Recently, research has included developing and testing interventions that attempt to reduce children's screen time.42,–,47 Although meta-analysis has been used to study the relationship between media use and weight and physical activity,11 to date no systematic review and meta-analysis of interventions to reduce children's screen time has been completed. In this article we present the results of a systematic review and meta-analysis conducted to identify and examine the effectiveness of interventions to change children's screen time. The broader behavior of screen time was selected because, although television use has remained relatively stable for more than 50 years,48 total screen time has increased over the past 10 years,6 and the American Academy of Pediatrics recommends limiting children's total screen time.2,49
Studies were identified with a systematic search of research databases, review of the table of contents of journals not included in searchable databases, review of the reference lists of relevant publications, and a search of the National Institutes of Health's Computer Retrieval of Information on Scientific Projects (CRISP) funding database (currently “NIH RePORT”).
Between November 24 and December 6, 2008, 8 databases available through the State University of New York at Albany Library were searched for the words “television,” “media use,” “recreational media,” and “screen time,” combined with “trial,” “program,” “intervention,” and “experiment” (see Appendix 1 for a sample search strategy). Search results were limited to works published after 1985 (the year Dietz and Gortmaker19 described the association between children's television viewing and weight). The Web-based Cochrane Library50 and the Centre for Reviews and Dissemination51 databases also were searched using the terms listed above. The table of contents of 2 journals that have only recently been indexed in Medline, Cyberpsychology and Behavior and Pediatric Exercise Science, also were searched by accessing the journal's Web site.
The reference lists of the Cochrane Collaboration reviews of obesity interventions52,53 and articles selected for inclusion in the meta-analysis were reviewed for eligible references. In an attempt to minimize the impact of the file-drawer problem,54 the Academy of Medicine's Gray Literature Report database; the National Institutes of Health's funding database, CRISP; and the Web site worldwidescience.org were searched. Researchers identified in the CRISP search were contacted and asked to provide information about their National Institutes of Health–funded project if they believed it met the meta-analysis inclusion criteria.
Identification of Eligible Studies
Articles identified by the systematic search were screened for eligibility using a 2-step process. First, references were identified as eligible for full review if the title or abstract-stated screen time was measured and targeted for change. If no abstract was available, the complete reference was reviewed. Second, articles eligible for full review were screened to determine eligibility for inclusion in the meta-analysis using the PICO (populations, interventions, comparisons, and outcomes) framework.55 A study was eligible for inclusion if it met all of the following criteria: (1) described an intervention or program to change behavior in children between the ages of 0 and 18 years; (2) outlined the results of an intervention to reduce screen time; (3) compared a nontreatment control, comparison group, or preintervention period with an intervention group or period; (4) included screen time (watching television or videos/DVDs, playing video or computer games, and using a computer for purposes other than school work alone or in combination) as an outcome variable; and (5) measured television viewing alone or with any combination of video viewing, computer time, or video-game use. Only articles published in English were eligible for inclusion in the meta-analysis.
Data Extraction and Coding
Information extracted from each article included sample characteristics (age, race/ethnicity, gender, and study location), intervention setting (home, school, or other), theoretical framework, intervention components (eg, goal setting, behavioral monitoring, and policy change), and behavior targeting (see Table 2). Study design information extracted included sampling and group-assignment procedures, timing of assessment, type of comparison group, and whether a validated instrument had been used to measure the outcome (see Table 3). An instrument was considered valid if the author stated that the instrument was validated or on the basis of a previously validated instrument. Whether the study targeted high-risk individuals also was recorded; studies targeting high-risk individuals only included participants with high BMI and/or excessive screen-time behavior. The sample size at group assignment and each assessment point and the number of subjects included in the analysis also were recorded. Finally, information about study outcomes, including means and associated SDs and mean change from baseline to posttest, were extracted for use in calculating effect sizes. Data were extracted using a standard data extraction instrument developed specifically for this study. The instrument was based on the Community Guide methods and instrument,56,–,60 other review instruments,55,61,–,64 and accepted methodologies.65,66 A copy of the data collection tool and associated code book are available on request.
Whenever possible, preintervention and postintervention means (SD) were used to calculate the study effect size. When unavailable, postintervention means (SD), mean change in each group, or adjusted differences after the intervention were used to calculate the study effect size. If an exact P value was not provided, a conservative approach of using an estimate closest to the significance level provided was used (eg, if P < .01 was provided, P = .009 was used to calculate the effect size).63,67
Given that the intent of the meta-analysis was to determine the overall effectiveness of the interventions, a mean study effect size was calculated when multiple effect-size statistics could be calculated55,68 (for example, when data were presented from several subsamples within a study or when multiple screen-time measures were presented). When key information was missing from an article, as was the case in 19 of 31 eligible studies, 3 attempts were made to reach the corresponding author via e-mail before using the available data or eliminating the study from the analysis.
Because of the subjectivity and the difficulty with assessing the overall quality of each study (because of insufficient information), a single index of study quality was not included in the analyses. Instead, study characteristics that are reflective of study quality (eg, number of groups, random assignment to groups) were examined as potential moderators, as outlined in greater detail below. An assessment of the heterogeneity of effect sizes was conducted to confirm the appropriateness of a random-effects model. Sensitivity and publication bias analyses also were conducted to improve accuracy and assess the robustness of the results.
A random 25% subsample of articles (n = 8) eligible for inclusion in the meta-analysis was reviewed by a second reviewer to assess coding reliability. κ was calculated by using Stata 9.2 (Stata Corp, College Station, TX) to determine coder agreement beyond chance.69,70 The percentage of agreement between coders also was calculated. All items included in the analyses had at least an 80% coder agreement or a κ value of >0.70. A κ value of >0.60 has been deemed good69,71 and 0.80 has been deemed acceptable in many situations.70 Any disagreements were settled via consensus. After data entry, all studies were reviewed a second time to verify data extraction and entry.
Data were entered into and analyzed with Comprehensive Meta-analysis 2 (Biostat, Englewood, NJ).72 Two measures of effect, the SMD and Hedges g, and associated 95% confidence intervals (CIs) were calculated. The data were coded such that a negative effect size indicated a greater reduction in screen time in the intervention group relative to the control or comparison group or preintervention period. A generally accepted criteria for effect-size magnitude was adopted (0.2 = small; 0.5 = medium; 0.8 = large).55,73,74
A random-effects model was used.55,75 Separate models were calculated for studies that measured the amount of screen time during the intervention period (n = 5) and studies that reported postintervention data (n = 27). Three studies were included in both data sets. When data were provided for multiple time points after the intervention (eg, immediately after and 6 months after the intervention), data from the time point most proximal to the intervention period were included in the analyses. At least 10 studies are required for moderator analyses to be conducted.55 Because too few studies reported data collected during the intervention period (n = 5) moderator analyses were conducted only with studies that reported postintervention data.
Heterogeneity and Moderator Analyses
To assess heterogeneity and identify possible moderator variables, effect-size estimates were calculated for subgroups and associated CIs were examined for overlap. The between-groups Q test, with a significance level of P ≤ .05, was used to assess significant differences between subgroups. Potential moderators were determined a priori68,76 and included age, gender, racial distribution (up to 50% nonwhite, >50% nonwhite), country, type of data (adjusted, raw), number of study groups, population risk, use of a television control device, outcome (screen time, television alone), intervention setting (school, home, other), and theoretical model. Because the Q statistic does not provide an assessment of the magnitude of heterogeneity,77 the I2 value68,78,–,80 was calculated to assess the proportion of variance that reflects real difference in effect sizes.68,79
Publication bias was assessed using 3 techniques, the funnel plot,81 failsafe N,82 and the Duval and Tweedie trim-and-fill method.83,84 In the absence of publication bias, the distribution of effect sizes in the funnel plot are symmetrical and take on an inverted funnel shape.63,85 The failsafe N or file-drawer number estimates the number of studies that would need to be included in the meta-analysis to change the overall results.55,82 The Duval and Tweedie trim-and-fill method assumes that the most undesirable studies are missing.86 An asymmetric appearance of many missing studies suggest publication or small-sample bias.55
Several sensitivity analyses were conducted to assess the robustness of the results: the SMD and Hedges g were compared; the impact of including adjusted data was assessed; and the impact of individual studies was determined.55 To assess the impact of including studies that provided an adjusted value for changes in television viewing behavior or postintervention means, the effect size was calculated with and without studies that provided adjusted data (n = 4). The overall effect size also was calculated repeatedly, with 1 study excluded each time to determine whether the overall effect size was strongly influenced by 1 particular study.
Systematic Literature Search
Fourteen databases were searched, resulting in 3002 potential studies. Thirty-three studies were eligible for inclusion in the meta-analysis (Fig 1). Four studies were excluded from the analyses; 1 study was excluded because it presented data on 2 cross-sectional samples,87 1 study was excluded because the outcome was binary (<1 hour/day of television)88 compared with all the other studies that presented continuous data (eg, hours per week of screen time), and 2 studies were excluded because data were unavailable.89,90 Twenty-nine studies were included in the analyses. Although 1 of the predefined exclusion criteria was non-English language, no articles written in a language other than English were identified. Fifteen authors were sent e-mails requesting additional information, and all but 1 responded. Eight authors provided data, 4 authors were unable to provide the data requested, and 2 authors stated that their work did not meet the study inclusion criteria. Search of the CRISP database identified 53 records that included the search terms. Only 4 of these studies were eligible for inclusion; 1 author responded to the request for additional information stating that data were not available at the time. The vast majority (n = 26) of studies were published in peer-reviewed journals, and 3 studies were doctoral dissertations.
Sample and Intervention Characteristics
Table 1 briefly describes the interventions eligible for inclusion in the meta-analysis and summarizes the study outcomes specifically related to screen time (n = 33). Table 2 provides additional information about the sample and intervention characteristics for interventions included in the analyses (n = 29). More than one-half of the interventions included in the analysis (18 of 29) were theory based (social cognitive theory was used most frequently). The school was the most common intervention setting (13 of 29), followed by the home (8 of 29). Many programs included both parents and children in the intervention. With regard to intervention components, approximately one-third (9 of 29) of the interventions facilitated behavior change by controlling the environment with a television-control device. Many interventions facilitated behavior change by setting goals and planning media use; often children participated in this process. Another common feature of the interventions was a behavioral contract in which children agreed to a specified amount of time in front of a screen. Often, a reward was provided if screen-time targets were met. Several interventions included increasing awareness by having children monitor and record their own screen time. Finally, only 1 intervention made television viewing contingent on physical activity.
The primary outcome in the majority of interventions (19 of 29) was screen time (television, video/DVD, computer, or video-game use alone or in combination). In approximately one-half of the studies (14 of 29), television only was the primary outcome. Only 5 of the interventions measured other screen-related behaviors, such as eating while watching television or having a television in the bedroom (Table 2).
With regard to sample characteristics (see Table 2), the majority of the interventions targeted children between the ages of 5 and 11 years (20 of 29). When demographic information was available, most study populations included 25% to 50% nonwhite participants and 25% to 50% male children. Ten studies targeted high-risk children. Eight studies limited participation to children who were overweight or obese, and 5 excluded children who did not use a predetermined amount of screen time. Most studies (20 of 29) were conducted in the United States.
As shown in Table 3, the vast majority of the studies included an intervention and a comparison group. Five interventions used 1 group with a pretest-posttest design. Among studies with 2 groups of participants, most used randomization to assign group membership. All studies reported baseline values of the outcome variable, 5 studies reported data collected during the intervention period, and only 2 studies reported follow-up data. Four of the studies included in the analyses reported adjusted postintervention data. Most (20 of 29) studies reported using a valid data collection tool.
Heterogeneity and Moderator Analyses
Table 4 contains the effect-size estimates calculated with a random-effects model for the 2 main groupings of studies, those reporting data collected during the intervention period and those reporting data collected after the intervention. The within-group variability was greater than would be expected by chance, signifying possible heterogeneity within each group. The between-groups Q statistic did not support the assumption of homogeneity; no moderators were identified. To assess the magnitude of heterogeneity present within each level of the potential moderators, the I2 statistic was calculated for each subgroup (ie, level) (Table 4). In general, I2 values were moderate to high, using generally accepted I2 values of 25, 50, and 75 (low, medium, and high, respectively).77,79 The presence of heterogeneity within subgroups and the nonsignificance in between-groups χ2 (Q) test supported combining studies using a random-effects model.
Overall Measure of Effect
When intervention effects were assessed after the intervention, most studies showed a small favorable effect of the intervention (Fig 2A). Individual intervention effects ranged from −3.9891 to 0.466.111 Several interventions had large effects with large CIs, likely because of the small sample size. These did not greatly impact the overall mean effect size. The overall mean effect of all interventions included in the analysis was small yet statistically significant. After the intervention, the overall SMD in means effect size was −0.148 (95% CI: −0.224 to −0.071) (Fig 2A) and Hedges g was −0.144 (95% CI: −0.217 to −0.072) (data not shown).
During the intervention period (ie, while the intervention was being delivered), the effect was large and statistically significant (SMD: −1.904 [95% CI: −3.041 to −0.767]) (Fig 2B). Likewise, Hedges g was −1.807 (95% CI: −3.069 to −0.545) (data not shown). Because of a lack of variance, 2 studies (identified with “a” in Fig 2) could not be included in the analyses to calculate Hedges g. Because the SMD and Hedges g differed only slightly and 2 studies could not be included in calculations of Hedges g because of a lack of variance, only the SMDs are presented.
Visual inspection of the funnel plots (Figs 3 and 4) show potential publication bias. On the basis of the Duval and Tweedie trim-and-fill method, 4 studies are missing (illustrated by filled circles in Fig 3). If the missing studies were included in the calculation of the overall postintervention mean effect size, the SMD (95% CI) would be −0.133 (−0.218 to −0.047), which is still small but significant. The trim-and-fill method did not identify any missing studies among the group of studies representing data during the intervention period (Fig 4). Assessment of publication bias with the failsafe N further supports the conclusion that the real effect size is not 0. For the group of studies presenting postintervention data, an additional 255 studies reporting no effect would have to be located and included in the analyses to nullify the existing results. For the second set of studies, those presenting data collected during the intervention period, the failsafe N was 40. Sensitivity analyses did not identify any decisions or studies that greatly impacted the overall conclusions.
This meta-analysis shows that interventions to reduce children's screen time have a small but statistically significant effect after the intervention and a large statistically significant effect during the intervention. Although studies included in the meta-analysis were heterogeneous, there was variability greater than would be expected by chance, no significant moderators were identified to explain this variability. It is possible that because of the number of studies included in the analysis, there was insufficient power to detect effect moderators. It also is possible that including studies with varied primary outcomes impacted the ability to detect moderators. The decision to include all interventions that targeted screen time instead of interventions that only targeted screen time is worthy of further discussion. We hypothesized that the majority of interventions, especially those developed earlier, were developed to address childhood obesity using screen-time reduction as a mechanism to decrease weight status. Although our broad inclusion criteria limit our ability to understand specific mechanisms of how television time uniquely impacts body fatness (ie, using an effect-modifier model), it increases the generalizability and clinical utility of the findings because attempts to reduce body fatness are almost always going to be multicomponent. Thus, it was important to not exclude studies that had a screen-time reduction strategy simply because other strategies also were used. This reflects our philosophical position that it is of greater public health value to know that including a screen-time reduction strategy is efficacious for reducing body fatness than it is to know simply that screen-time reduction works as a stand-alone strategy. These points notwithstanding, the ability to test potential moderators will improve as the evidence base on screen-time interventions grows.
Several methods were used to increase confidence in the study results. On the basis of the results of the sensitivity analyses, the decision to calculate SMD instead of Hedges g did not impact the conclusion. Although an attempt was made to identify unpublished works, none were identified. Visual observation of funnel plots, the Duval and Tweedis trim-and-fill method, and the failsafe N were used to assess the robustness of this study's conclusion. The funnel plot and the trim-and-fill method identified few possible missing studies and the failsafe N was large. Collectively, these methods support the validity of the primary finding.
The Guide to Community Preventive Services59 groups interventions into several categories: interventions that include provision of information only (interventions that try to change knowledge, attitudes, or norms); behavioral interventions (those that try to change behavior by providing skills or materials); environmental interventions (those that try to change the physical and/or social environment); and policy or regulatory interventions. Most of the interventions reviewed include an information provision component.42,–,45,47,91,–,107 This strategy has been recommended by Dennison and Edmunds108 as a means to change behavior. The majority of the interventions are behavioral interventions. They increase skills by having the child and/or parent develop a television-viewing budget or plan, set screen-time goals or monitor viewing,* or have the child identify alternative activities.42,96,100,109
Several of the interventions attempted to change behavior by modifying the environment by restricting access to the television or computer using a television-control device† or by providing opportunities for physical activity.99,103,107,111 It is worth noting that 491,97,98,110 of 5 interventions with large effect sizes after the intervention used a television-control device to help budget screen time. Although the majority of adolescents (aged 8–18 years) have media devices in their bedrooms,6 and 33% of children aged 6 years and younger have televisions in their bedrooms,5 only 1 intervention42 reported changes in the proportion of study subjects with televisions in the bedroom. Finally, none of the interventions included in this review attempted to change children's behavior through regulation. However, 1 intervention,107 included policy makers in the change process by requesting that they provide supportive environments for physical activity in the form of low- or no-cost physical activity opportunities. Given that most interventions used a combination of strategies, intervention strategy was not tested as a moderator.
Children are able to take with them and use media now more than ever. In fact, 20% of children's media consumption is from mobile devices, such as cell phones, iPods/MP3 players, and video-game players.6 Reducing the amount of time children spend with screen media and increasing discriminate media use and media use budgeting are important given the negative health and behavioral implications of excessive screen time. Results from this meta-analysis show that interventions to decrease children's screen time have a small but statistically significant effect. Because excessive screen media use has been associated with many negative behavioral and health consequences, these results support the implementation of screen-time reduction interventions. Many of the interventions reviewed can provide children with the skills needed to decrease screen-media use. Parents and clinicians should incorporate the interventions or the strategies that are common to the effective interventions into their efforts to combat childhood obesity. Even modest effects could result in a positive change in the health status of the population given the large number of children who use screen media and the increasing amount of time children spend with media. As the evidence base expands, and the number of screen-time interventions increases, future research can expand on these findings by examining the clinical relevance and sustainability of effects observed, conducting a more thorough analysis of effect modifiers, and by identifying critical components of effective interventions.
- Accepted March 22, 2011.
- Address correspondence to Dayna M. Maniccia, DrPH, MS, Department of Health Policy, Management, and Behavior, School of Public Health, University at Albany, 1 University Place, Rensselaer, NY 12144. E-mail:
Dr Maniccia contributed to the study conception and design; data collection, analysis, and interpretation; and manuscript writing. Dr Davison contributed to the study design and manuscript writing and revision. Dr Marshall contributed to the study design, statistical analysis design, and manuscript revision. Drs Manganello and Dennison contributed to the study design and manuscript revision.
FINANCIAL DISCLOSURE: The authors have indicated that they have no personal financial relationships relevant to this article to disclose.
- SMD —
- standard mean difference
- CI —
- confidence interval
- Copyright © 2011 by the American Academy of Pediatrics