Family structure as a predictor of screen time among youth

Study design and population

Study data are based on the nationally representative cross-sectional 2009/2010 Canadian Health Behaviour in School-aged Children Survey (HBSC). The HBSC is conducted every 4 years in 43 countries in collaboration with the World Health Organization (Currie, Nic Gabhainn & Godeau, 2009; Freeman, King & Pickett, 2012). This study is limited to the Canadian data. The HBSC consists of a standardized self-report survey filled out in a classroom setting, with the goal of determining the prevalence and distribution of a wide range of psychological, social and physical determinants of health in adolescents. All items in the HBSC study are continuously developed, piloted, and validated by the HBSC international network (Currie, Nic Gabhainn & Godeau, 2009; Freeman, King & Pickett, 2012).

Ethics approval for the 2009/2010 Canadian HBSC was given from Health Canada and the General Research Ethics Board at Queen’s University. Depending on the jurisdictional requirements at the participating schools, informed consent from the parents/guardians of participating students could have been obtained in an active or passive manner. In both situations a letter of information was sent home to parents/guardians. If active consent was required, the parent/guardian had to sign the letter and it had to be returned to the school or the student was not allowed to participate. If passive consent was required, the parent did not have to sign or return the letter if they allowed their child to participate. Conversely, they were asked to sign and return the letter if they did not want their child to participate. Irrespective of whether active or passive consent was used, student participants had to provide their informed consent, which was demonstrated by their willingness to complete the survey.

The 2009/2010 Canadian HBSC had a 77% response rate. The final sample consisted of 26,068 students in grades 6–10 (approximate ages 11-15 years) from 436 public schools across Canada. All provinces and territories participated, with the exceptions of New Brunswick and Prince Edward Island. The provincial samples were obtained using a two-tiered cluster-sampling procedure to sample entire classrooms for participation; all students living in the three territories were invited to participate if they met the study inclusion criteria to ensure adequate representation. Students attending private, on-reserve, special needs or home-based schools were excluded, as were those who were absent from school on the day of the survey.

Exposure (family structure)

Information on family structure was derived from two questions on the HBSC. The first asked participants to check off the adults who live in the home “where [they] live all or most of the time” from a list of choices including mother, father, stepmother (or father’s girlfriend) and stepfather (or mother’s boyfriend). The second asked participants to indicate whether they had a second home and, if they did, to identify how often they stayed there (“half the time,” “regularly but less than half the time,” “sometimes” or “hardly ever”). Families were defined as traditional (includes both a mother and a father), single parent (includes either a mother or a father), or reconstituted (includes either a mother or a father and either a stepmother/father’s girlfriend or stepfather/mother’s boyfriend). Adolescents from non-traditional families were further defined as having “regular visitation” with a second parent if they had a second home and reported visiting it “half the time” or “regularly but less than half the time” and “irregular visitation” if the adolescent reported not having a second home, or having a second home but visiting it “sometimes” or “hardly ever.” Youths who reported that neither their mother nor their father lived in their primary home constituted ∼4% of the sample and were excluded from the analysis.

Outcome (screen time)

Information on the three screen time behaviors of interest was obtained by asking participants to report how many hours in their free time on a typical weekday and weekend day they usually “watch television (including videos and DVDs),” “play games on a computer or games console (Playstation, Xbox, Gamecube, etc.)” or “use a computer for chatting on-line, internet, emailing, homework, etc.” For each question there were 9 ordinal response options, ranging from “none at all” to “about 7 or more hours a day.” A weighted average of weekly time spent in each screen time behavior was calculated. A validation study of a similar questionnaire measuring weekly television use in adolescents showed that participants’ responses were significantly correlated (r = 0.47) with television viewing time as measured using a detailed 7-day log (Schmitz et al., 2004). Screen time was dichotomized so that those in the highest quartile of weekly time spent in each screen time behavior could be compared to those in the lower three quartiles. Separate analyses were performed comparing those who spent >2 h in each screen time behavior to those who spent ≤2 h, as screen time guidelines have recommended this threshold (American Academy of Pediatrics, 2001; Tremblay et al., 2011a).

Covariates

Potential covariates for the model of the relationship between family structure and screen time were selected based on previous literature and their availability within the HBSC. These included gender, grade, ethnicity (Canadian, which includes those who self-identify as both Caucasian and Aboriginal ethnicity, East and Southeast Asian, South Asian, Black, Arab, or other, which includes those of mixed ethnicity and who also who self-identify as other), immigration status (non-immigrant/immigrated >5 years ago or immigrated ≤5 years ago), presence of siblings (yes or no), and family affluence. Family affluence was measured through a self-report item on the HBSC that asked students to report “how well off [they] think [their] family is,” with five ordinal responses ranging from “not at all well off” to “very well off.” The top and bottom two responses were combined to create a three-level ordinal measure of perceived family affluence.

Statistical analysis

All analyses used survey procedures in SAS 9.4 to account for the complex sampling design used by the HBSC, including clustering at the classroom and provincial levels and sampling weights. All analyses were stratified by gender. The HBSC sample population was characterized using simple descriptive statistics. Cross-tabulations of the average amount of time spent in each screen time behavior by family structure were calculated, and differences between family structures were assessed using analysis of covariance after adjusting for relevant confounders. A Bonferroni correction was used to account for 10 multiple group comparisons. Note that the 95% confidence intervals (CI) for the group means that are presented in the manuscript were not adjusted to account for multiple group comparisons; we only corrected the P values used to denote statistical group differences.

Multiple logistic regression was used to further quantify associations between family structure and screen time behavior. Final covariate selection for the multivariate models was performed through backwards deletion using a 10% change-in-estimate threshold (Rothman & Greenland, 1998). If a covariate significantly changed the regression coefficient of at least one non-traditional family structure in at least one of the models, it was included in all final models. All covariates of interest met this criterion.

Approximately 15% of participants were missing data for at least one relevant exposure, outcome, or confounding variable. Because we were concerned that this would bias our results, we performed two forms of multiple imputation for the missing data: one based on fully conditional specification and the other based on the use of Monte Carlo Markov Chains. Because the two imputation methods produced consistent and similar results, only the results from the fully conditional specification-based imputation are shown in the manuscript. This method is more appropriate for the imputation of categorical data because it does not assume normality of imputed variables (Lee & Carlin, 2010). In both imputation procedures, all variables were included for the multiple imputation, and 50 concatenated datasets were created rather than the typical 5 in an effort to reduce any potential bias caused by rounding.

Sample characteristics

Demographic characteristics of the sample are in Table 1. The majority of participants had lived in Canada for more than 5 years (94.4%), were of Canadian (Caucasian or Aboriginal) ethnicity (75.3%), and considered themselves to be of higher than average socioeconomic status (53.7%). Most participants lived in traditional, dual-parent families (65.0%), followed by single parent families with irregular visitation with their non-custodial parent (13.4%), reconstituted families with irregular visitation (6.6%), single parent families with regular visitation (4.2%) and finally reconstituted families with regular visitation (2.7%). Most participants exceeded Canada’s recreational screen time guidelines, with 80.6% reporting that they spent a cumulative total of more than 2 h per day watching television, playing video games, and using a computer. The average weekly screen time was 59.4 h in boys and 53.4 h in girls.

Relationships between family structure and screen time

As shown in Table 2, youth from non-traditional families had slightly higher screen time values than youth from traditional families, although few differences were statistically significant and even the significant differences were small in magnitude (e.g., <5 h/week difference between the mean screen time in the non-traditional and traditional).

Tables 3 to 6 show odds ratios for being in the highest quartile of weekly screen time by family structure, with traditional families serving as the referent group. Statistically significant (p < 0.05, after Bonferroni correction) relationships are shown in bold. There were several significant relationships in the bivariate analyses. However, these significant relationships were weak in strength (i.e., odds ratios <1.25) and were not consistent across gender or the three screen time behaviors. Furthermore, these relationships were no longer significant after controlling for the confounding variables.

The associations between family structure and screen time did not change in a meaningful way when the definition of elevated screen time was changed from being in the highest quartile of screen usage to exceeding a 2 h/day threshold (data not shown). The associations also did not change based on the choice of imputation methods (data not shown). The relationships observed in the imputed data were, however, different from those estimated using a full case analysis in which observations with missing data were simply deleted. For example, the odds ratios for boys from single parent families being in the highest quartile of video game usage were significant in the full case analysis (OR 1.40, 95% CI [1.22–1.60] for those with irregular visitation and OR 1.48, 95% CI [1.05–2.09] for those with regular visitation), indicating that boys from single parent families watched more television than those from traditional families regardless of visitation. However, these odds ratios were not significant when using the imputation models to reduce biases resulting from missing data (OR 1.13, 95% CI [0.96–1.34] and OR 1.17, 95% CI [0.91–1.51], respectively).