Crystal Smit

An Integrated Model 3 55 Covariates In the analyses, participants’ school cohort, sex, body-mass index z score (z-BMI), and hunger/thirst were included as control variables. School cohort was considered in the analysis by rating whether the participant was in 0 = ‘primary school’ or 1 = ‘secondary school’. The height and body weight of the participants were individually measured according to standard procedures (without shoes but fully clothed) at the participating schools during T2 and 1 year later during T4. Height was measured to the nearest 0.1 cm and body weight was measured to the nearest 0.1 kg. The BMI for each child was calculated using the following formula: weight over height squared (kg/m 2 .) Z-BMI was calculated and represented standards for Dutch children (Schönbeck et al., 2011). Hunger and thirst were assessed on a Visual Analogue Scale (0 cm = not hungry/thirsty; 15 cm = very hungry/very thirsty; Bevelander et al., 2012). Strategy of Analyses Descriptive statistics were calculated to examine means and standard deviations of all model items. Next, correlations among all model items were computed to assess bivariate associations. The primary analyses consisted of two structural equation models (SEMs) using Mplus Version 7.2 (Muthén & Muthén, 2012). The first SEM testedwhich predictors from the various theories were related to fruit and vegetable consumption over time; the second SEM tested the various predictors of water consumption (see Figure 3.1). The covariates (sex, school cohort, z-BMI and hunger/thirst) were included in both SEMs as predictors of participants’ Behavioral Intentions and Behavior. For both models, Attitude, Self-Efficacy, and Intrinsic Motivation were included as latent constructs. For the model predicting fruit and vegetable consumption, the two fruit and vegetable consumption items were also used to form latent constructs at each assessment. The models included regression paths for the behaviors from T1 to T2, T2 to T3, T3 to T4, and for Intentions from T1 to T2 to account for interindividual stability in the behavior. The parameters in the models were estimated applying the (full- information) maximum-likelihood estimator with robust standard errors (MLR in