Teun Remmers

Investigating determinants of children's outside play | 53 First, parents were asked how many weekdays and weekend days in an average week their child played outside. Second, parents were asked to indicate the average duration in the morning, noon and evening that their child played outside, again separately for weekdays and weekend days. We computed total minutes of OP in the morning, noon and evening. Responses were multiplied by the indicated number of days that the child played outside, separately for weekdays and weekend days. Finally, we summed weekdays and weekend days in order to arrive at the average minutes of OP. We used the date on which the questionnaire was completed to classify the season (i.e. winter, spring, summer and autumn) of both OP at age five and seven. Season was then used as a confounder in all subsequent analyses. Statistical analysis Characteristics of the population for analysis at child age five were evaluated with descriptive statistics. As PA levels have shown to considerably differ between boys and girls (21), we explored this potential gender difference with chi-square tests and t-tests for categorical and continuous levels of measurements, respectively. Based on the results described in Table 1, we decided not to stratify our subsequent analyses for child gender, as no significant differences were found for the predictors. We fitted general linear regression models with predictors at child age five as independent variables and OP of the child at age five as the dependent variable. First, we tested univariate associations between each predictor and OP. Relevant predictors (p < 0.10) were entered in separate models per category of predictors, namely socio- demographic characteristics (Model 1), family environment (Model 2) and perceived PE (Model 3). In Model 4, all predictors were entered simultaneously. In all models, season at child age five was added as a confounder. We fitted linear regression models between OP at age seven as the dependent variable and selected predictors at child age five as independent variables. Season at age seven was entered as a confounder. All variables were selected and further entered in the model in a similar way as described for the cross-sectional analyses. Linear mixed models were used to evaluate the influence of the family and perceived PE on the development of OP over two years. We entered OP at five and seven (additional row for a repeated observation), with season at age five and seven entered as a random effect. To investigate the influence of predictors on the development of OP (i.e. change over time) we computed time in years to follow-up by subtracting child age at follow-up from child age at baseline. Subsequently, we modelled interactions between “time to follow-up” and individual predictors to evaluate whether there were associations between the predictor and OP development between child age five and seven. We again selected these predictors based on univariate associations.