Participants and data collection
The present cross-sectional study was conducted in a cohort of children living in Musashino City, Japan, in 2018 and Kokubunji City, Japan, in 2017. A total of 4,800 potential residents aged 6–12 years were randomly selected from the residential registries of their respective cities. Both Musashino (population: 150,660 in October 2021) and Kokubunji (population: 130,636 in October 2021) are cities in Tokyo, Japan. Potential participants were stratified by sex (boys/girls) and school grade (1st grade: 6–7 years, 2nd grade: 7–8 years, 3rd grade: 8–9 years, 4th grade: 9–10 years, 5th grade: 10–11 years, and 6th grade: 11–12 years). First, invitation letters explaining the study were sent to all potential participants. A questionnaire and accelerometer were sent to those who responded to the invitation letter indicating that they were willing to participate. To encourage a response, potential participants were told that a 1,000-yen book voucher would be offered to those who returned the questionnaire and accelerometer. Non-respondents were sent one reminder about the responses to a questionnaire and accelerometer. A total of 1,772 individuals (36.9% overall response rate; 881 responders [36.7%] from Musashino and 891 responders [37.1%] from Kokubunji) responded to the invitation. Then, self-administered questionnaires, which included questions about sociodemographic variables, screen time, and children’s height and weight, were mailed to those who responded that they would be willing to participate (620 individuals; 12.9% of those who said they would be willing to participate, 310 people in each city). A total of 484 individuals completed both the questionnaire and accelerometer measurements (78.1% overall response rate; 81.0% from Musashino, and 71.3% from Kokubunji). Data from 283 children, mothers, and fathers who fully completed both questionnaires were included in the analysis. Participation was voluntary, and confidentiality was ensured. A previous study [31] suggested that children younger than 10 years of age are unable to report their activity patterns accurately or reliably. Alternatively, parental reports of physical activity among 6-year-olds have been shown to strongly correlate with heart rate measures during physical activity [32]. Therefore, parents or guardians of the children were asked to complete the questionnaire with their children.
Standard protocol approvals, registrations, and patient consent
All children, mother, and father signed an informed consent form before answering the questionnaire. The Ethics Committee of Waseda University, Japan, approved the study prior to its commencement (2017-245). The present study was conducted in accordance with the principles of the 2013 Declaration of Helsinki.
Measures
Self-reported screen time
Domain-specific sedentary behaviors were assessed using a questionnaire. For children, sedentary behavior was divided into six domains [10] (1) reading or listening to music, (2) TV or video viewing, (3) TV game use, (4) internet or e-mail (computer or tablet) use outside of class, (5) doing homework or assignments, and (6) car travel for transport. Participants were asked how many days on average per week and how much time (hours and minutes) on average per day they engaged in these sedentary behaviors during weekdays and weekends in each domain. Then the frequency per week was multiplied by the number of minutes per day. Each domain-specific sedentary behavior was examined separately, and we calculated the average total number of minutes for each school week (Monday–Friday) and weekends (days × minutes per day). Screen time was calculated using the total of domains (2), (3), and (4). The mothers and fathers were asked to report daily average sedentary time (hours and minutes) over the past 7 days, separately for workdays (weekdays for non-employed) and non-workdays (weekend for non-employed) across the following six domains: (1) being transported to and from a place by car; (2) using public transport; (3) at work; (4) watching television, videos, and DVDs; (5) using a computer, cell phone, and tablet PC outside of working hours; and (6) during leisure time (excluding watching television, videos, and DVDs) [33]. The total minutes of daily average screen time was calculated by summing (4) and (5), separately for workdays and non-workdays. Average daily values of sedentary time were calculated with weighting to account for the number of weekdays and weekend days.
Sociodemographic factors
Data on children’s age and sex were collected from the residential registries. The children’s weights and heights were obtained from the questionnaire responses. BMI was calculated from the height and weight data (BMI = weight/height [2]). Children’s BMI percentiles were calculated using the metric system of the Centers for Disease Control and Prevention [34]. Children were classified according to the recommended BMI-for-age cutoffs [34]: <5th percentile, underweight; 5–85th percentile, normal BMI; and ≥85th percentile, overweight or obese. Additionally, household income level per year (<3, ≥3–<5, ≥5–<7, ≥7–<10, or ≥10 million yen) was assessed from parents or guardians.
Statistical analyses
The data analysis involved assessing the replies from the 283 children, fathers, and mothers who had fully responded. Path analyses were conducted to determine the presence of any associations between the children’s, fathers’, and mothers’ variables. It was hypothesized that after controlling for household income and children’s sex and age, mothers’ and fathers’ screen time during the weekdays and the weekend would be related to children’s screen time during the weekdays and the weekend, respectively; it was also hypothesized that children’s weekday and weekend screen time would be related to BMI. Path coefficients and correlations are reported as standardized estimates. The model was assessed using the goodness-of-fit statistic (GFI), adjusted goodness-of-fit statistic (AGFI), root mean square error of approximation (RMSEA), and Akaike information criterion (AIC). GFI and AGFI indices were used to measure how well the model fit the data. Values of 0.90 or greater indicated a good model fit [35]. RMSEA is a measure of the descriptive measures of overall model fit. An RMSEA score values lower than 0.05 indicated a good fit [36]. A lower AIC value for a model indicated a better fit than the other models [37]. A model was considered to fit the data well when the following criteria were met: GFI >0.90, AGFI >0.90, RMSEA <0.06, and a lower AIC value compared with competing models. Statistical significance was set at p < 0.05. The data were analyzed using path analyses estimated using IBM SPSS AMOS 27.0J for Windows (IBM Corp., Armonk, N.Y., USA).