Study
The data presented here were collected as part of an ongoing 6-year multi-modal study designed to assess the impact of Scottish legislation banning tobacco POS displays on young people’s smoking-related attitudes and behaviours [23]. The information on e-cigarette advertising exposure, e-cigarette use and smoking status were gathered through a school-based survey conducted in four purposively selected communities. Schools were selected to reflect two levels of urbanisation (urban vs. small town) and two levels of social deprivation (high vs. medium/low).
The results reported here are based on data collected between January and March 2015 after the implementation of the tobacco point of sale ban in large supermarkets but before its implementation in small shops. Questionnaires were administered to pupils in Secondary 1 through to Secondary 6 (age range 10.83–18.67 years, mean 14.71 sd 1.65) by teachers during class time under examination conditions. Information regarding the survey was sent out to parents two weeks prior to the survey date. Parents could withdraw their child from the survey by returning the opt-out form to the school. Young people could decline to take part in the survey or withdraw from the survey at any point. Ethical approval for the study, including the consent procedures, was obtained from the University of St Andrews, School of Medicine Ethics Committee.
Analysis approach
This study aimed to examine the relationship between e-cigarette point of sale recall (in large supermarkets and small shops) and e-cigarette use. Previous e-cigarette use and intention to use e-cigarettes in the next 6 months were the dependent variables. E-cigarette POS recall in large supermarkets and small shops were the explanatory variables. Recall of other sources of e-cigarette advertisement was included in the adjusted models to control for the potentially confounding effects of exposure to other types of e-cigarette advertisement. In addition, demographic factors such as age, sex, and family affluence were included as control variables. Tobacco smoking is known to be a predisposing factor to e-cigarette use therefore previous tobacco use was also included in the adjusted models as a control variable in order to test whether e-cigarette POS sale influenced e-cigarette uptake in never smokers (never tried tobacco cigarettes).
Demographic variables
Respondents were asked their gender, ethnic group, and date of birth. Ethnic group was dichotomised to white ethnic group versus other responses. Individual family material well-being was assessed through the Family Affluence Scale (FAS) [24] which is a validated measure consisting of six questions (own bedroom, number of family cars, number of computers, number of family holidays abroad per year, owning a dishwasher and number of bathrooms). The FAS raw scores were transformed though categorical principal component analysis into single dimensional scores that were then divided into tertiles of high, medium, and low FAS scores. This asset based measure of family affluence has been shown to be strongly related to household income (eta squared around 0.30 in most countries [25]).
E-cigarette use
Respondents were asked whether they had heard of e-cigarettes and could respond ‘yes’, ‘no’, or ‘don’t know’. Respondents who answered ‘yes’ were directed to the question ‘which ONE of the following is closest to describing your experience of e-cigarettes?’, to which they could respond ‘I have never used them’, ‘I have tried them once or twice’, ‘I use them sometimes (more than once a month)’, or ‘I use them often (more than once a week)’. This variable was dichotomised to ‘ever tried-1’ versus ‘never tried-0’. The next question was ‘Do you think that you will try e-cigarettes in the next 6 months?’ to which they could reply ‘yes I do’, ‘no I don’t’ or ‘don’t know’. This response was also dichotomised to ‘yes’ -1 versus ‘no’ or ‘don’t know’-0. In the original coding of the e-cigarette variables participants who responded that they did not know what an e-cigarette was or missed this question were coded as ‘missing’ on all other e-cigarette questions. Another alternative coding method was employed to minimise missing data where participants who did not know what an e-cigarette was, were assumed not to have tried one and not to intend to try one. Results are presented for both versions of the e-cigarette dependent variables.
Cigarette smoking
Ever smoking was assessed with the question: ‘Have you ever smoked cigarettes, even if it is just one puff?’ to which they could respond ‘yes’ or ‘no’. Negative responses were used as our variable for ‘never smoked’. To assess intention to smoke respondents were asked ‘Do you think you will smoke a cigarette or hand-rolled cigarettes (roll-ups) at any time during the next year?’ To which they could answer ‘definitely yes’ ‘probably yes’ ‘probably not’ ‘definitely not’. This was dichotomised to ‘yes’-1 or ‘no’-0.
Advertising
Respondents were asked ‘During the past 30 days, have you noticed any adverts (e.g., shops, shopping centres, TV, radio, billboards, newspapers, magazines, etc.) for (a) cigarettes or hand rolled cigarettes (roll-ups), or (b) electronic cigarettes (e-cigarettes)?’ To which respondents could answer ‘Yes’, ‘no’ or ‘don’t know’. This response was also dichotomised to ‘yes’-1 versus ‘no’ or ‘don’t know’-0.
Internet advertising
Respondents were asked ‘When you are using the internet, how often do you see adverts for (a) tobacco products or pictures of people smoking, or (b) electronic cigarettes (e-cigarettes) or people smoking them?’ To this pupils could respond ‘I don’t use the internet’, ‘most of the time’, ‘some of the time’, ‘hardly ever’, ‘never’. Responses of ‘I don’t use the internet’ were recoded as ‘never’. All responses were dichotomised to ‘never’ -0 versus other responses-1.
Recall point of sale displays in supermarkets and shops
Respondents were asked whether in the past 30 days, when they had been in (a) large supermarkets and (b) small shops, they could remember seeing (i) cigarette or tobacco packs or (ii) electronic cigarettes displayed for sale. To this they could respond ‘yes’, ‘no’ or ‘don’t know’. This response was also dichotomised to ‘yes’ -1 versus ‘no’ or ‘don’t know’-0.
Analysis
Data analysis and imputation was conducted in Stata version 14. Analysis was by logistic regression with standard errors adjusted for clustering by school. Where the prevalence of the outcome was low (i.e., near to 10 % for dependent variable ‘intention to try e-cigarettes' -Table 3 and dependent variable intention to try cigarettes-Table 4), the alpha value was adjusted to 0.01 [26] and therefore the 99 % confidence interval for the odds ratio is presented.
Interactions between demographic variables and other predictors were explored for all models and interaction terms retained in the models only if the interaction term was significant.
To test how robust the results were across different methods of handling missing data, we used multiple imputation by chained equations to fully impute the dataset. Multiple imputation by chained equations is an iterative process that imputes multiple variables using a series of univariate chained equations with fully conditional specification of prediction equations. It is based on Monte Carlo simulation techniques for sampling from complicated multivariate distributions. Auxiliary variables were included in the imputation model: whether parents were employed, receipt of free school meals and area level deprivation. A uniform prior distribution was assumed. A burn in of 10 iterations was employed and this was confirmed to lead to convergence through examination of trace plots. Twenty imputations were initially performed and adding additional imputations up to a total of 100 did not significantly change the parameter estimates.