The interviewed sample was 6,486 at Time 1, 4,995 at Time 3, and 4533 at Time 4. At Time 1, the mean age of youth was 12 (range 10-14); 62% were White, 18% Hispanic, 11% Black, and 9% were Other race. Overall, 31% had a parent with a college degree, and household income ranged from $10,000 or less (8%) to over $75,000 (30%). Analyses of attrition showed that persons who dropped out of the study were somewhat more likely to be of nonwhite race and lower socioeconomic status, and to score higher on rebelliousness and sensation-seeking.(9
) Detailed multivariate analyses of attrition effects showed that the set of study variables accounted for 3-5% of the variance in attrition, so overall the magnitude of attrition effects was moderate and the composition of the continuing sample was similar to that of the baseline sample.
Descriptive data showed that participants reported the following grades for baseline school performance: “excellent” (30%), “good” (42%), “average”/”below average “(28%). There were significant correlations at baseline between the covariates and the media use variables at Time 1 (). For example as compared to youth with better grades, participants with Average/Below average grades were more likely to have a television in the bedroom, to have more hours of television viewing and video game playing, and to watch a higher proportion of movies that were PG-13 or R-rated; and participants with greater screen exposure and more PG13/R movie viewing scored higher on sensation seeking and school problem behavior. For intermediate variables, the proportion of participants who had ever smoked (even a puff) increased from 10% at Time 1 to 18% at Time 3, and the proportion who had ever drunk alcohol increased from 10% to 19%. The mean score for sensation-seeking increased from 7.9 at Time 1 to 8.3 at Time 3, and the mean score for school problem behavior increased from 2.41 at Time 1 to 2.50 at Time 3. At Time 4, participants reported the following grades: “A's & B's” :61%; “Mostly B's” :12%; “B's & C's”: 20%; “Mostly C's” :6%; “Mostly D's & F's”:1%. shows the relationship between baseline covariates and school performance at Time 4.
Relationship between media use and other variables at baseline
Relationship between baseline variables and poor school performance at 24 months.
The final structural model () had chi-square (127 df, N = 6,486) of 1021.50, CFI of 0.92, and RMSEA of 0.033, these parameters generally indicating reasonable fit of the model to the data. Extracurricular activities and parental smoking were nonsignificant and were dropped from the initial model. The residual correlations of Time 3 variables (excluded from the figure for graphical simplicity) were 0.09 between substance use and school problem behavior, 0.16 between school problem behavior and sensation-seeking, and 0.10 between sensation-seeking and substance use. Hypothesized paths from the Time 3 intermediates to Time 4 school performance were all significant (beta = .06 for substance use, beta = .14 for school problem behavior, and beta = .12 for sensation seeking), thus qualifying these as mediating variables. A direct effect from Time 1 screen exposure to poor school performance at Time 4 (beta = .07, t = 1.80, p < .10) was omitted from the figure because it did not meet the criterion for statistical significance; a path from poor school performance at Time 1 to substance use at Time 3 (beta = .05, t = 4.45, p < .0001) was omitted from the figure for graphical simplicity. The prior variables in the model accounted for 20% to 40% of the variance in the hypothesized mediators. Together the variables in the model, including direct effects, indirect pathways, and the stability coefficient for school performance, accounted for 48% of the variance in Time 4 school performance.
Figure 2 Structural model for relation of Time 1 predictors and Time 3 mediators to Time 4 school performance. Analytic N=6,486. Ovals indicate latent constructs, rectangles indicate manifest variables. Values are standardized coefficients; all coefficients are (more ...)
The coefficients in are standardized to make them comparable; they indicate the change in school performance expected for a 1 standard deviation increase in the predictor, adjusted for all other covariates. Regarding hypothesized pathways, there was a significant indirect effect from more Time 1 screen exposure to worsened school performance at Time 4 through an increase in sensation-seeking at Time 3. Also, Time 1 PG-13 movie viewing had indirect effects for worsened school performance at Time 4 through two mediating variables, increased substance use and increased sensation-seeking at Time 3. The Time 1 measure for R movie viewing had indirect effects on worsened school performance through changes in all three mediators from Time 1 to Time 3: increases in substance use, sensation-seeking, and school problem behavior.
With regard to the effects of other Time 1 variables on school performance, maternal responsiveness and monitoring resulted in better school performance because they decreased school problem behavior and substance use, respectively. In addition to a direct effect for (better) school performance, good self-control also improved school performance because it was related to a decrease in sensation-seeking. Sensation-seeking itself was a key predictor variable aside from the role as an intermediate variable for media effects: initial sensation-seeking led to worsened school performance through effects on higher levels of substance use and school problem behavior at Time 3. In addition to the stability coefficient for school performance from Time 1 to Time 4, poor school performance at Time 1 affected subsequent school performance through its links to increases over time in school problem behavior, sensation-seeking, and substance use.
Effects for Time 1 demographic variables, included in the model but excluded from the figure for graphical simplicity, were as follows. Positive relations to change in substance use were noted for older age (beta = 0.14, p < 0.0001), female gender (beta = 0.05, p < 0.001) and White race (beta = 0.06, p < 0.001). Inverse relations to change in school problem behavior from Time 1 to Time 3 were noted for female gender (beta = -0.04, p < 0.01) and household income (beta = -0.07, p < 0.0001). Change in sensation-seeking was positively associated with White race (beta = 0.07, p < 0.0001) and was inversely associated with household income (beta = -0.04, p < 0.01). Female gender (beta = -0.09, p < 0.0001), White race (beta = -0.06, p <0 .01), higher household income (beta = -0.06, p < 0.01), and higher parental education (beta = -0.08, p < 0.0001) were all inversely related to worse school performance.