All 617 children in the study had ≥3 days of diary data, but 3 had insufficient sleep data and 2 were missing demographic covariates, resulting in an analytic sample of 612 (99%). Of these children, 55% were male, 18% were from low-income households, and 10% had a bedroom television (). Children with a bedroom television were more likely to be low-income or single-adult households. Sleep problems were common, with 18% of families reporting the child experienced at least 1 of the queried sleep problems 5 to 7 days per week. The most frequently reported problem was difficulty with sleep-onset latency, and children with a bedroom television were more likely to have parent-reported daytime tiredness (8% vs 1%; P < .01). The sleep problem scale ranged from 5 to 12 in this population (mean: 6.4) and was significantly higher in the bedroom television group (P < .01).
Average daily screen time was 72.9 minutes (), with 58.8 minutes in the daytime (before 7:00 pm), and 14.1 minutes in the evening (after 7:00 pm). Compared with daytime use, evening use was more likely to have ratings for older children or adults, to have violent or scary content, and to occur with adult co-use. Although violent content accounted for a greater proportion of evening use, content consumed within categories was similar across time periods; for both violent and nonviolent content, 8 of the top 10 content titles were the same for daytime and evening. Not surprisingly, screen time was significantly greater for children who had a bedroom television, with an additional 40 minutes across the day. Subsequently, children with a bedroom television also consumed significantly more violent or scary content or content rated for older children or adults. However, the proportion of daily content represented by these categories was similar to those in children without a bedroom television.
Media Use Characteristics in Mean (SD) Minutes per Day
In model 1, total screen time was associated with sleep problems, with an increase in the sleep problem score of 0.244 (95% confidence interval [CI]: 0.113 to 0.375) for every additional hour of daily media use (). In model 2, we split total media use into daytime and evening use, and confirmed our hypothesis (P < .01) that evening use has a greater impact (0.819 [95% CI: 0.450 to 1.188]) than daytime use (0.107 [95% CI: −0.047 to 0.260). In model 3, both daytime and evening use were further split into violent and nonviolent content. For daytime use, there was a negative impact of violent media consumption on sleep problem score but no impact of nonviolent daytime media use (P = .048 for the difference in effects). Within evening use, the differences between violent and nonviolent media use were not statistically significant (P = .96), so these 2 categories were recombined for the subsequent model. In model 4, there were significant increases in sleep problem scores with each additional hour of evening media use (0.743 [95% CI: 0.373 to 1.114) or daytime violent media use (0.398 [95% CI: 0.121 to 0.676]). Again, there was no impact of daytime nonviolent media use, which continued to be the case throughout all subsequent models.
Regressions: Change in Sleep Problem Score for Each Additional Hour of Media Use
In model 5, we examined whether co-use with an adult mitigated the effects, and split both evening media use and daytime violent use according to whether it occurred with adult co-use. Using the Wald tests, we found no differences in effect with adult co-use versus without adult co-use for either evening media use (P = .35) or daytime violent use (P = .79). To test the last hypothesis, in model 6 we split evening media use and daytime violent use according to whether the child had a bedroom television. There were no significant differences between evening use with and without a bedroom television (P = .87), so we recombined those for model 7, in which we observed a trend for an increased effect of daytime violent media use in children with a bedroom television (P = .098). Although this trend was intriguing, it did not reach the cutoff for statistical significance; as a result, we chose model 4 as our final model ().
In this final model, we found significant increases in the sleep problem score for violent daytime media (0.398 increase with each additional hour of use [95% CI: 0.121 to 0.676]) and any evening media use (0.743 [95% CI: 0.373 to 1.114]). Conversely, nonviolent daytime media use had no significant impact on the sleep score (0.022 [95% CI: −0.186 to 0.229). Regression diagnostics on this model found no overall problems with nonlinearity, severe outliers, significant multicollinearity, or specification errors. In examining the residuals, we found model fit to be somewhat less specific in children with sleep problem scores in the top 1% than for the rest of the sample.
Our ability to examine effect modification according to child characteristics was limited by our smaller sample in the potentially vulnerable subgroups. No significant differences were observed in effects when media use covariates were split according to internalizing or externalizing status, and neither did the results of the final model change significantly when the internalizing and externalizing scale variables were excluded. There was no effect modification according to gender, but there was a trend (P = .07) toward a greater effect of daytime violent use in children from low-income (0.761 [95% CI: 0.276 to 1.246]) compared with non–low-income (0.239 [95% CI: −0.089 to 0.567]) households.
In secondary analyses raised by the aforementioned findings, we examined the individual components of the sleep problem score and more detailed categorizations of content. Across components, we found a significant effect of evening media use on sleep-onset latency (0.204 [95% CI: 0.01 to 0.403]), nightmares (0.085 [95% CI: 0.005 to 0.165]), and daytime tiredness (0.162 [95% CI: 0.030 to 0.294]), and a trend toward an effect on night wakings and hard time waking in the morning. Although we found no impact of daytime violent use on sleep-onset latency or night wakings, there was an impact on nightmares (0.087 [95% CI: 0.027 to 0.147]), hard time waking in the morning (0.170 [95% CI: 0.061 to 0.279]), and daytime tiredness (0.111 [95% CI: 0.012 to 0.210]).
Although the primary analyses focused on violent content meeting the frequency criteria for episodic or central, we also examined the individual frequency categories. We found no significant difference in effect between content with isolated violence and no violence (P = .61) and likewise no significant difference between episodic and central violence (P = .66). For type of violence, there were no significant differences between fantasy violence, mild/slapstick violence, sports violence, or realistic violence, and no content meeting criteria for gratuitous violence was reported. There were also no significant differences between the effects of animated and live-action violence (P = .37). In addition, although we examined “scary” content as well, the large degree of overlap between “scary” and “violent” content in the media consumed by this age group meant that there was no marginal effect of scary content above and beyond the effect of violence. We also explored to what degree official ratings from the Motion Picture Association of America, TV Parental Guidelines, and Entertainment Software Rating Board were an effective proxy measure for violence in the media content reported. When we replaced the daytime violent and nonviolent categories in the analysis with those split according to ratings category, we found no significant difference in the impact on sleep between content rated as appropriate for young children and those with a nonrated designation, and likewise no significant difference between content rated as appropriate for older children and content rated for adolescents or adults. As with violent content, daytime media use of content rated for older children, teenagers, or adults was associated with a significant increase in the sleep problem score (0.58 [95% CI: 0.15 to 1.01]). As with nonviolent content, daytime use of content rated for young children did not have a significant impact on sleep (0.03 [95% CI: −0.18 to 0.24]).