Design, participants, & procedure
This study had a longitudinal design and used baseline (T0) and one year follow-up (T1) data from a smoking prevention intervention trial called “Fun without Smokes” [
30]. The “Fun without Smokes” study is a web-based, computer-tailored smoking prevention intervention that was evaluated in a cluster-randomized controlled trial. At T0 (October 2011) and T1 (October 2012), children completed a web-based questionnaire concerning their smoking behavior, intention, attitude, social influence, and self-efficacy expectations toward (non-) smoking. After completion of the baseline assessment, children in the experimental group received computer-tailored feedback letters via email and at the “Fun without Smokes” website. Children in the control group did not receive feedback letters. A detailed description of the “Fun without Smokes” intervention study is available elsewhere [
30].
Children in the “Fun without Smokes” study were recruited through primary schools by Municipal Health Promotion Organizations and Maastricht University. Children in grade 7 (aged 10–11 years) were eligible to participate in the intervention study. Approximately 3,500 Dutch primary schools were approached for participation in the smoking prevention study, but only 162 primary schools decided to participate (N = 3,213 children). In the present study a passive informed consent procedure was used in which all children of the participating schools received informed consent letters for their parents or guardians. If children, parents, or guardians refused to be involved in the “Fun without Smokes” study they were able to sign the informed consent letter and return it to the children’s teacher (1.7 % refused). Subsequently, the teachers informed the research team about the children that refused participation. At T1, 2,146 children (33.2 % drop-out rate) from 133 primary schools filled out the follow-up measurement with the same group of children now in grade 8 (aged 11–12 years). Both baseline and follow-up measurements were completed in the classroom under teacher supervision. In the present study, only the responses from children who had completed both measurements and had provided a verifiable postal code were included in the analyses.
After one year of follow-up, no intervention effects were observed; therefore, it was possible to use the longitudinal data of “Fun without Smokes” for the present study.
Measurements
The primary outcome measure was intention to smoke. The intention to smoke was assessed by self-reports using a previously used question [
31,
32]: “Do you intend to start smoking in the future?” A five-point Likert scale was used in the answer format (1 = certainly yes; 2 = yes; 3 = I don’t know; 4 = no; 5 = certainly not). Children who indicated ‘no’ or ‘certainly not’ were categorized as not having the intention to smoke (coded as 0). Otherwise, children were categorized as having the intention to smoke (coded as 1). It was expected that children who were undecided about smoking were more inclined to engage in smoking compared to children who were certain about not smoking. For that reason, children who indicated ‘I don’t know’ were also categorized as having the intention to smoke.
SES of the participating children was based on the SES index score of the areas in which they live, as determined by their postal code. All postal codes have an SES index score. This SES index score was retrieved from the Netherlands Institute for Social Research (a Dutch government agency that conducts research into the social aspects of all areas of government policy), which gathers information from all Dutch inhabitants concerning their income, occupation, and education. These indicators were used to calculate an SES index score for the 4-digit postal code areas. Thus, SES index scores indicated social status at a neighborhood level [
33,
34]. The SES index score ranges from +3.4 to–5.2 and is based on Dutch inhabitants’ income, occupation, and education. All scores higher than zero were indicated as high SES. The higher this SES index score, the higher the SES of the child. SES index lower than or equal to zero indicated low SES. The lower the SES index score, the lower the SES of the child. In the present study, children from an LSES neighborhood were coded with a ‘0’ and children from an HSES neighborhood were coded with a ‘1’.
Background variables included the age (in years), gender (1 = boy; 2 = girl) and ethnicity of the participating children. In line with the guidelines of Statistics Netherlands, a child was deemed to have a Western ethnic background (scored as 1) if he/she and both parents had been born in the Netherlands, another European country, North America, Oceania, Indonesia (a former Dutch colony), or Japan. Otherwise the child was deemed to have a non-Western ethnic background (scored as 2) [
35].
The socio-cognitive constructs were derived from the integrated model for exploring motivation and behavioral change (I-Change model) [
36].
The attitude dimension advantage was measured by assessing the positively perceived consequences of smoking using nine items. Participants answered these questions by using their perception of the various benefits of smoking, such as feeling more mature, sociable, cool, or receiving more attention from friends. Children were asked to complete the following question “If I smoke….” with a four-point answer category ranging (for example) from 4 = ‘I will feel very mature’ to 1 = ‘I will not feel mature’ (Cronbach’s alpha = 0.84).
The attitude dimension disadvantage was measured using ten different negatively perceived consequences of smoking, such as I will become less physically fit, I will become ill or, I will become addicted. Children provided an answer on a four-point scale ranging (for example) from 4 = ‘I will become very ill’ to 1 = ‘I will not become ill’ (Cronbach’s alpha = 0.80).
Social norm was assessed through the perceptions of smoking norms of important people in the child’s environment. Children had to complete seven questions addressing their father, mother, brother (s), sister (s), friends, best friend, and most people important to them. For example, “My mother thinks that I….”. These questions could be scored on a five-point Likert scale ranging from +2 = ‘definitely should not smoke’ to -2 = ‘definitely should smoke’ (Cronbach’s alpha = 0.69).
Modeling was measured by assessing the smoking behavior of parents, siblings, family, and friends. A total of eight questions were asked, such as: “Does your mother/father/brother (s)/sister (s)/best friend smoke?” (the five-point answer formats ranged from 5 = ‘often’ to 1 = ‘never’) and “How many of your friends/other family members/classmates smoke?” (the five-point answer scales ranged from 5 = ‘(almost) all’ to 1 = ‘(almost) none’). Children could also indicate that they had no parents, siblings, family, or friends or that they did not know if people in their social environment smoked; these answers were also categorized as ‘1’. To create a single variable for modeling, an average score was calculated: the scores of all individual items (best friend, mother, father, brother (s), sister (s), friends, other family members, and classmates) were added and divided by the number of questions. Therefore, a higher score on this scale indicated that more people smoked in the child’s social environment.
Self-efficacy expectations were measured with ten questions to assess the child’s ability to refuse cigarettes in different situations. For example, the question was posed “When others smoke it is….for me not to smoke”; this question was answered using a five-point Likert scale ranging from +2 = ‘very easy’ to -2 = ‘very difficult’ (Cronbach’s alpha = 0.93).
Analyses
Descriptive analyses (means and percentages) were performed to describe the sample under study. This sample included only non-smoking children. Potential differences between boys and girls of HSES and LSES neighborhoods were assessed through an analysis of variance (ANOVA) using Gabriel’s pairwise comparison test (GABRIEL). This post-hoc test is suitable for unequal sample sizes [
37] and adjusts for multiple testing. Significant differences observed between the subgroups indicate that they potentially modify the effects of the socio-cognitive factors. Therefore, factors that differed significantly between subgroups were included as interaction terms (with attitude, social influence and self-efficacy) in the analyses to test for effect modification. If those interaction terms were found to be significant, they were included in the main analyses regarding the subgroups. The influence of school and class level on the smoking intention of participating children was analyzed to test for possible nesting effects. The variance of the random intercept of both the school and class level was zero, indicating that multilevel analyses were not warranted.
To identify whether SES and gender moderated the association of predictor variables (i.e. attitude, social influence and self-efficacy expectations) with the outcomes 3-way interaction terms (SES by gender by predictor) were included in a logistic regression analysis. The analyses were adjusted for age and ethnicity. If a 3-way interaction effect was determined to be present, subgroup analyses were carried out for boys and girls living in HSES or LSES environments. All analyses were performed in SPSS 20.0. The significance level was set at p ≤ 0.05. To reduce potential type I errors, the interaction effects were considered significant if the p-value was equal to or lower than 0.10.