3.3 Independent variables

Models were estimated using independent variables described in S1 Table. These variables were chosen through an analysis of relevant literature. Students’ basic characteristics, such as age, sex, and ethnicity, were included to capture possible demographic trends related to alcohol consumption initiation and binge drinking. The authors choose to use age groups rather than a continuous age variable because a student’s age information in PeNSE 2015 is truncated at eleven and nineteen years old. For more details, see Pechansky et al. [49]. Likewise, we controlled for the possible influence of extracurricular activities, like a paid or unpaid occupation (such as internships, part-time jobs, etc.) and the number of hours spent on physical activities not related to physical education (PE) class. The estimated models controlled for the consumption of tobacco-based products or illicit drugs in the previous thirty days and also for the student’s perception of how many of his or her friends consumed alcoholic beverages.

The estimated models also controlled for some household and school characteristics as Brazil is a country with high social inequality (the 2015 GINI index is 0.491, according to the Brazilian Institute of Geography and Statistics–IBGE [50]. Owing to inequality, students were exposed to different environments that offered varying levels of favorability for the consumption of alcoholic beverages. Factors related to social environment included whether the student studied in a public or full-time school, the school’s geographic location, living with one or both parents or guardians, and the number of people with whom the student shared his/her residence.

Three variables listed in S1 Table were constructed through principal component analysis (PCA): household economic status, parental supervision, and mental health condition. These analyses were made with the aim of capturing some otherwise non-observable information and summarizing the influence of a group of characteristics related to the described dimensions of PeNSE 2015 in the student’s life.

The household economic status variable was based on whether the student’s household had a landline telephone, a computer, internet access, car, motorcycle, a housekeeper, and on whether the student had a personal mobile phone exclusively for his or her own use. The parental supervision variable was based on how often (never, rarely, sometimes, most of the time, always) the student’s parents knew what the student was doing during free time, checked their homework, and went through their belongings. The mental health condition was measured based on how often (never, rarely, sometimes, most of the time, always) the student felt upset, bothered, hurt, offended, or humiliated by his or her classmates, felt lonely, and could not sleep at night because something was bothering him or her. The generated variables have a distribution with a mean of zero and a standard deviation of two. Which generates an indicator with positive and negative numbers. A higher value on these variables indicates a student having better household economic status, more protective parents, and a worsened mental health condition. Only the principal component of each these PCAs was used to resume the respective dimension and the results of these analyses are presented in the S2 Table.

It is important to point out that the variables listed in S1 Table are based on students’ answers to the 2015 PeNSE questionnaire. Consequently, it is not possible to ignore the presence of tentative answers due to intentional omission of information or failure to accurately record some events.