Participants and Procedure
Participants were recruited into a large-scale cross-sequential cohort study through their schools (21 primary (n = 453) and secondary (n = 500) schools) in the Netherlands as part of the so-called MyMovez project (41). Active written consent was obtained from the school directors, caretakers and the participants themselves. Participants received a smartphone with the MyMovez research application for nearly a week. They received random invitations to fill out questionnaires on their smartphone each day between 7:00 AM and 7:30 PM (but not during school hours, except for school breaks) (for detailed information about the recruitment procedure and the Wearable Lab, see (41)). Background and precursor variables were assessed during Phase I of the MyMovez project (41). For the present study, we made use of data on frequency of snack consumption, perceived descriptive and injunctive parent and peer norms, and several background variables collected over a period of three years.
Data collection took place around February and March 2016 (T1), 2017 (T2) and 2018 (T3). The participants were allowed to enter and drop out when they wanted (also during a measurement period), because participation was voluntary. In addition, the participants attending the highest grade levels dropped out at T2 and T3, because they left school. From the 953 participants who initially had parental consent, the response rate was 71% at T1, 40% at T2 and 33% at T3. Participants who answered to the set of questions used for this study at least once were included in the study (N = 819; 47.5% primary school children; 46.1% boys; M(SD) age = 11.19 (1.36); > 90% Dutch origin).
Core and non-core snack intake. Participant’s self-reported snack consumption was assessed by a food frequency questionnaire (FFQ) which accounted for Dutch food items based on the Dutch EPIC FFQ (42). Participants were asked to recall every other day (i.e., three times per data wave) how many pieces of food items they consumed on the previous day with answering options ranging from 0 = none to 6 = six or more. The number of units were multiplied by the average kilo caloric value representing each food item category to place relative weight on the food items (e.g., a small cookie does not equal the energy value of a piece of pie) . The participant’s reported consumption was averaged for each food item per wave. Next, a clear distinction was made between core and non-core items based on previous literature and nutritional guidelines of the Dutch Nutrition Centre (39, 40, 43). This resulted in three snack food item categories: 1) core: fruit and vegetables; 2) non-core: small, large and wrapped cookies, sweet pastry, chocolates, chocolate bars, candy and liquorice, savory and warm (pastry) snacks, chips, ice cream, and 3) an in-between category: (skimmed) milk, cottage cheese, nuts. A proportional snack intake value was then calculated by adding up the three categories for a total snack consumption score and dividing the kilo caloric values for core and non-core snacks by the total snack consumption score. The two core and non-core measures were used in the analyses.
Descriptive norms. Perceived descriptive norms on fruit and vegetable snack intake were assessed with two separate items about parents and friends: ‘How often do your parents/friends eat fruits and vegetables as a snack?’(31). Response items ranged from 1 (‘never’) – 6 (‘always’).
Injunctive norms. Participants’ perceptions of social pressure on their fruit and vegetable snack intake were assessed with two separate items for parents and friends: ‘Do you think that your parents/friends believe you should eat fruits and vegetables as a snack?’(31). Response options ranged from 1 (‘no, certainly do not’) – 6 (‘yes, certainly do’).
Covariates. Age, sex and weight status were mentioned in empirical studies as potential confounders in the relationship between social norms and snack intake (12, 16, 44). Demographic variables age and sex were supplied by the schools’ administration offices. Height and length was measured each year by trained researchers following standard procedures. Standardized BMI scores were calculated accounting for variations in growth curves of children and adolescents (45).
Strategy of Analyses
Descriptive statistics were calculated to examine the distribution (minimum, maximum and means) of all model items and the differences between time points T1-T3. In addition, bivariate correlations among all model items were computed.
The primary analyses consisted of a series of competing cross-lagged autoregressive models specified in a structural equation modeling framework in Mplus version 7.2 (46, 47). To examine our research questions, 2 sets of four models were compared for descriptive and injunctive norms on snack intake. Model 1 (answering RQ1) was the baseline or stability model, which only included autoregressive paths (estimating intra-individual stability) between the three assessments and all concurrent correlations among constructs. Model 2 (RQ2) examined social norms as predictors of snacking behaviors, and included the same parameters as Model 1, but also included cross-lagged paths from T1 and T2 social norms to T2 and T3 snack intake, respectively. Model 3 (RQ3) examined snacking behaviors as predictors of social norms, and included the same parameters as Model 1, but also included cross-lagged paths from T1 and T2 snack intake to T2 and T3 social norms, respectively. Model 4 (RQ4) examined bidirectional associations between social norms and snacking behaviors, and included all parameters specified in Models 1, 2 and 3. Figure 1 presents an overview of the four models. Age, sex, and BMI were included as covariates in the models on T1 based on their significant correlations with the model variables under investigation (see Table 2). The parameters in the models were estimated using (Full-Information) Maximum Likelihood estimation with robust standard errors (MLR in Mplus) to account for missing values and potential deviations from multivariate normality. In the additional analysis, we combined all variables (i.e., descriptive and injunctive parent and peer norms) in one model to examine the bi-directional relationship between norms and snacking behavior.
For each model, model fit information was assessed by the following fit indices: the χ2 test of model fit, the Standardized Root Mean Square Error of Approximation (RMSEA; satisfactory values below .06) (Hu & Bentler, 1999), the Comparative Fit Index (CFI) and the Tucker-Lewis Index (cut-off values close to or above .90). Next, Chi-square difference tests with MLR scaling correction were assessed to assess which model(s) provided a significantly better fit to the data.
 T3 also served as a baseline measurement for the MyMovez intervention at the end of the project during Phase II. In order to schedule all required questionnaires without overburdening the participants, the FFQ was assessed once.
 Three members of the MyMovez team independently scored kilo caloric values for each of the food items. Inter-coder reliability was consistently high (ĸ=.97).