Page:Drug Use and Nightlife： More than Just Dance Music.pdf/4

Van Havere et al. Substance Abuse Treatment, Prevention, and Policy 2011, 6:18http://www.substanceabusepolicy.com/content/6/1/18 Fisher Exact Probability Tests with * p < 0.05, ** p < 0.01 and *** p < 0.001.

funky music” included Salsa, Latino and R&B, hip-hop and rap, disco, reggae and ragga.

Given the variety of substances included in the questionnaire, we limited our findings to the substances most frequently used by those participating in Belgian nightlife: alcohol, cannabis, MDMA, amphetamines and cocaine [24].

To determine the relation between substance use and nightlife variables, two types of analyses were performed. First, to determine whether the odds of being an illegal substance user are higher for certain music and nightliferelated variables (i.e. music preference: rock, dance and southern/funky music; last month visits: clubs, pubs and goa parties; sampling venues: dance events, rock festivals and clubs), we calculated odds ratios for the subsample of respondents who claimed to have used an illegal drug during the last year and the subsample of those who did not. Focusing on use last year gives a more reliable insight than focusing on more recent use, because the latter category could be influenced by the timing of the survey: during holidays and free of responsibilities, young people tend to use more substances than during the school year [30]. We compared last year illegal drug use (yes/no) with the nightlife related variables (yes/no variables for dance, rock and southern/funky music, going to pubs, clubs, goa parties, sampled at dance events, rock festivals or in clubs). Second, to investigate the association between the frequency of use for specific types of drugs and the various independent variables, we performed five separate ordinal regression analyses using a proportional odds model [cf. [31]]. In each of these five analyses, the frequency of using a specific substance during the last year (alcohol, cannabis, amphetamines, MDMA or cocaine) was regressed on age (entered as continuous variable), gender, music preference (yes/no for dance music preference, southern and funky music preference, rock music preference), number of visits to clubs, pubs and goa parties within the last month (entered as continuous variables), and sampling venue. To interpret venue effects with regard to the grand mean, the original venue variable consisting of three categories was recoded using an effect coding scheme [32].

For practical reasons (because more categories increase the difficulty of data interpretation from ordinal regression analyses), the original dependent variable (frequency of last year use) consisting of 7 categories was reduced into a variable with three ordered categories: (1) No use :people who never used this drug or have used it, but not within the last year; (2) Occasional use: people who recently used this drug, on a monthly basis or less frequently; and (3) Regular use: people who used this substance at least weekly. As opposed to simple logistic regression, using these ordered categories enables us to investigate the frequency of last year use instead of simply having used a certain drug the last year (yes/no). In the parameterization used by Stata for the proportional odds model, a positive value for b indicates that with increasing values for the predictor, the odds increase of being above a given value of k (with k = 1, ..., number of ordinal categories - 1). In other words, a positive coefficient implies increasing probability of being in higher-numbered categories (of the dependent variable Y) with increasing values for the predictor (holding all other independent variables fixed) [32]. For the present analyses this means that a positive coefficient points to an increased probability of being a high frequency user.

The proportional odds assumption was not fulfilled for models with alcohol, cannabis and MDMA use as the dependent variable. However, fitting partial proportional odds models for these cases did not alter data