Propriedades psicométricas de estrutura interna do Teste de Rastreio de Álcool, Tabagismo e Envolvimento com outras Substâncias (ASSIST) quando usado em adolescentes

Abstract

The primary objective of this thesis is to evaluate the psychometric properties of the internal structure of the Alcohol, Smoking, and Substance Involvement Screening Test (ASSIST) when applied to adolescents. The original instrument comprises eight items, of which seven assess the risk of involvement with a specific substance. This study operationalizes these seven items in an ordinal manner, considering sequential scores that allow mapping the propensity for substance misuse and abuse.. The data sets were obtained from a cluster-stratified sample comprising 693 students whose ages range from 15 to 19 from the ninth Administrative Region of Rio de Janeiro. The analysis was restricted to tobacco, alcohol, and cannabis, in addition to individual substance-specific assessments. In the first article, four psychometric properties were examined for each substance: dimensional structure (P1), reliability (P2), absence of item content redundancy (P3), and convergent factorial validity of the scale (P4). The analyses followed four steps: (1) Horn’s parallel analysis, followed by exploratory structural equation modelling (ESEM) with one- and two-factor solutions (P1); (2) confirmatory factor analysis (CFA), focusing on factor loadings and residuals (P2); (3) assessment of residual correlations (P3); and (4) calculation of the average variance extracted (AVE) (P4). To ensure methodological robustness, probit models, WLSMV estimators, and fit indices (RMSEA, CFI, and TLI) were employed, given the ordinal nature of the item metrics. The second article investigated the properties of the scale structure through the application of a non-parametric item response theory model to evaluate item scalability, using (1) Loevinger’s H coefficient and (2) graphical representation of Guttman errors for item pairs. The Horn’s parallel analyses and ESEM indicated a single-factor structure. Subsequent CFA analyses revealed factor loadings exceeding 0.56, residual correlations below 0.35, and average variance extracted (AVE) exceeding 50% for tobacco, alcohol, and cannabis. Furthermore, Loevinger’s H coefficients were above 0.3, and Guttman errors for most item pairs were below 70%, indicating that the scalability of the items was satisfactory. The ASSIST item set was found to be appropriate for assessing the propensity for substance misuse and abuse, as well as for ranking adolescents along a continuum of increasing substance use intensity when consolidated into ordinal or factorial scores. While further research is required, the findings support its recommendation for screening and population monitoring in this age group.