Measures and procedures
The measurements included items related to alcohol drinking behaviour and socioeconomic variables that were collected by means of an online questionnaire administered during school hours. Information was collected at two different points in time (0–4 months, or at baseline, and at the 4-month follow-up period). This questionnaire was divided in the following sections: socioeconomic variables (age, sex, nationality, religion, economic situation at home, parents’ education level, weekly pocket money, etc.), alcohol use (binge drinking occasions in the last month, alcohol use in the last week, etc.), social influences (parents, sibling, friend´s alcohol use, etc.) and family functionality [34]. Only the number of BD occasions in the last 30 days and variables related to alcohol use were assessed at two assessment points, pre- and post-intervention. The rest of the variables (sociodemographic variables such as gender, age, religion, nationality, family composition, parents’ education level and economic situation at home) were assessed only at baseline or in the pre-intervention period.
The model’s endogenous variable was the number of occasions of binge drinking in the last 30 days, obtained directly from the answers given by the adolescents to the question: “During the last month, how many times did you drink 4 glasses or more of alcohol (if you are a girl) or 5 glasses or more of alcohol (if you are a boy) on one single occasion (e.g. in a bar, at a party etc.)?”. This variable was self-reported in both pre-intervention and post-intervention assessments. The units of this measure were event counts. The definition of binge drinking used in this study is consistent with the Spanish Survey on Drug Use in Secondary Schools (ESTUDES) [16] as well as the definition used by Jander et al. [35], as Alerta Alcohol is an adaptation of the Dutch programme Alcohol Alert [33]. One alcoholic drink equivalent is defined as a drink containing 9.9 grams of pure alcohol, similar to Spain, where one standard unit of alcohol is equivalent to 10 grams of alcohol. An image was provided in the questionnaire to help respondents understand the meaning of a standard beverage unit or glass of alcohol.
Age was calculated by dividing the difference between the date on which the subject completed the pre-intervention questionnaire and the subject’s birth date by 365.25. Family composition was determined by means of the question “What is the composition of your family?”, with multiple answers being allowed from the following response options: mother, father, brother(s)/sister(s) who live(s) at home, brother(s) who do(es)/do(es) not live at home, sister(s) who do(es)/do(es) not live at home, other (nominal response). We created a new categorical variable called “family composition”, which classified families as: nuclear, extended nuclear, divorced, extended divorced, reconstituted and other, but due to the large proportion of nuclear families in the sample, we dichotomized this into nuclear and others.
Nationality was a dichotomous variable with two response options: value of 1 if you were “Spanish” and, value of 0 if you had “other/s” nationality. When adolescents answered “others”, they specified the additional nationality/es. Due to the proportion of responses for each nationality, this variable was used in the analyses only as a dichotomous variable.
The parents’ educational level was calculated according to years of schooling. Initially, we collected the educational level with a categorical variable asking "What is the highest level of education achieved by your father/mother?”.
The economic situation at home was obtained using the question “Of the following situations, which one would you identify with the most?”. The response options were: (a) We have many economics problems at home and we don’t make it to the end of the month; (b) We manage economically at home, but we have trouble making it to the end of the month; (c) We are pretty well off economically and we make it to the end of the month; and (d) We are very well off economically. This variable was converted into a dummy variable, which indicated a value of 1 “good economic situation at home” (including options (c) and (d), and a value of 0 “other economic situation” (including options (a) and (b). This question was developed ad hoc and used in another study carried out by Lima-Serrano et al. [36].
To measure weekly pocket money, we asked about the amount of money the subject had available to spend on how his/her appearance weekly (not including money for clothing or his/her savings). Response options were: “0 euros”, “up to 10 euros”, “between 11 and 20 euros”, “between 21 and 30 euros”, “more than 30 euros”. This categorical variable was converted to a continuous variable using the mean number of euros for each option.
Other new variables were created to indicate how near the subject was to the most popular local events in each city (there were a large number of events during the study period), as well as how much time had elapsed since last weekend when the subject completed the questionnaire. To obtain the first variable, we codified the date on which the questionnaire was completed and the date of the closest local event for each city and then calculated the difference between the two codes. For the second variable, we recoded the date on which the subject completed the questionnaire as the corresponding day of the week. The variable was then dichotomized into subjects who completed the questionnaire on Monday or Tuesday and subjects who completed the questionnaire later in the week. In relation to the first session, all participants were influenced to answer the initial questionnaire for the same event, which was Christmas.
The variable for family alcohol consumption was calculated on the basis of engaging in binge drinking for family members, i.e., the mother, father and siblings. Three items were taken from the questionnaire that asked about the frequency of those family members who consumed 4/5 glasses of alcohol or more on a single occasion (Response options: “Never”, “Almost never”, “Occasionally”, “More frequently”). Subsequently those items were dichotomized, with the value being 1 when the mother/father or siblings occasionally or more frequently consumed 4/5 glasses of alcohol or more on a single occasion, and with the value being 0 when they never or almost never consumed 4/5 glasses of alcohol or more on a single occasion. Then, the items were combined in a only ordinal variable with four response options (0: Mother, father and siblings did not engage in binge drinking, 1: Mother or father or siblings engaged in binge drinking, 2: Two members of the family (mother, father or siblings) engaged in binge drinking, 3: Mother, father and siblings engaged binge drinking).
Family functionality was measured using the family APGAR questionnaire [37–40], which is a tool frequently utilized in primary care and general medicine settings to assess family function through a 5-item questionnaire measuring five constructs (adaptability, partnership, growth, affection and resolve). Religion was determined through a categorical variable with several answers, and tobacco use was a categorical variable (with nine response options from “never smoke” to “daily”) that was converted to a dummy variable, with a value of 1 for “smoker” and a value of 0 for “non-smoker”. Those who reported being smokers were asked about the number of cigarettes and shishas they consumed in terms of a numerical variable.
Data analysis
We used panel count data for the empirical analysis. Additional variables were created in order to carry out the analysis. The latter included variables that indicated how near the subject was to the most popular local events and how much time had elapsed since the last weekend when the subject completed the questionnaire.
The analysis to determine the factors associated with binge drinking in the sample studied was conducted using three econometric procedures: a negative binomial, a finite mixture model and a two-part model. Given the nature of the endogenous variable (count data), we used a negative binomial specification, which would resolve the main drawback of the Poisson model in which the variance is equal to the mean. The data showed greater variance than average due to multiple causes, notably the high frequency of zeros.
The basic idea of these models (negative binomial regression models) is that the zeros (all or part of them) do not come from the same data-generating process as the rest of the values. For instance, in this study, for those who had engaged in binge drinking zero times, the reason might be because they do not drink alcohol or are not binge drinkers, or it might be because, although they are binge drinkers, they had not engaged in binge drinking in the last month.
The two-part model, along with the finite mixture model, was used to explain the probability of not reporting binge drinking and how often it happened, given that the two variables could be completely independent of each other. Finite mixture models have received increased attention in recent years due to their usefulness for modelling heterogeneous data with a finite number of unobserved sub-population and the probability of belonging to each unobserved group in order to estimate distinct parameters of a regression model or distribution in each group, to classify individuals into the groups and to draw inferences about how each group behaves [41].
The aim in using various models other than the Poisson model was to allow for greater flexibility, bearing in mind the main disadvantage of the Poisson model, which is the difficultly in capturing overdispersion—i.e., when the conditional variance exceeds the conditional average. Specifically, the first part of the two-part model is estimated using a logit regression model and the second part is specified as a generalised linear model (GLM) panel regression. This model was used because of the presence of a large proportion of zero count observations [42]. In the dataset, the number of binge drinking occasions was zero for 60.95% pre-test and 67.32% post-test.
Additionally, the introducing interactions between socio-economic variables and the intervention were estimated. Notwithstanding, none of these interactions turned out to be statistically significant, so we decided to use the entirely sample for this analysis.
The analysis was conducted using Stata (version 16.0; StataCorp, College Station, TX, USA).
Ethics approval
The study received approval from the Bioethics Committee of Andalusia. The clinical trial registration number is: PI- 0031–2014. Registration date: 04 August 2015. Written informed consent was obtained from parents and students prior to participation in the study. The questionnaires were self-completed by the adolescents and confidentiality was ensured.