Teun Remmers

90 | Chapter 6 were the primary (fixed) explanatory variables. Analyses were generally adjusted for accelerometer daily weartime, school term, gender, age and BMI. Our primary aim was to analyze day-to-day PA variation within individual children (intra- individual variation). Therefore we adjusted for differences between individual children in our model (such as variation related to differences between schools) by allowing multilevel linear mixed models to compute a random intercept for each child, and a repeated term for each child across each day of measurement. Theoretically, as a result of investigating day-to-day variation within children using the model specified, additional adjustment for between-subject variables (such as school or gender) would not improve the model, nor the estimation of our relationship of interest. We found homogenous variances but decreasing covariances with increasing distance between repeated measurements; therefore we accounted for the covariance between repeated measures by specifying an autoregressive (AR1) covariance structure. Linearity was evaluated for each association, and in the case of curvilinear relationships, quadratic terms were fitted (21). Multi-collinearity was investigated using linear regression analyses. With Variance Inflation Factors (VIF) ranging between 3.9 and 4.4, day length and temperature revealed highest multi-collinearity statistics, followed by solar radiation (VIF 3.0). Only removing day length from the model led to notable reduction in multicollinearity statistics (VIF’s 2.8 and 2.3 for solar and temperature, respectively). We decided to retain all variables in subsequent analyses, as all Variance Inflation Factors were under 4.4 (20). Potential moderation of gender, age, and BMI, as well as mutual moderation between weather variables (e.g. temperature and relative humidity) was evaluated by systematically investigating interactions and changes in model fit. BMI was dropped from the models investigating MPA as its initial main effect was not statistically significant. Advancing from the multi-collinearity statistics computed earlier, we compared results from models with- and without day length, to investigate whether potential collinearity between temperature and day length would have influenced these interactions. However, as both model-variations were highly comparable (data not shown), we retained day length in subsequent interaction models. Finally, we repeated our analyses for light PA and sedentary time. For all analyses, SPSS 21.0 for Windows (IBM SPSS Inc., Armonk, NY) was used, and p <0.05 indicated statistical significance. Results In total, 307 of 326 children (52% girls) provided valid PA data (≥1 day) in at least one school term and were included in the analyses. The average age was 11.14 years (SD=0.67, range=8.7-12.8) among boys and 11.10 years (SD = 0.68) among girls. In total, 24.7% of the boys and 15.9% of the girls were classified as overweight and 5.3% of the boys and 7.6% of the girls were classified as obese, according to International Obesity Taskforce thresholds for BMI (7).