Hester van Eeren

| Chapter 5 5 | 100 Three secondary outcome measures were assessed at the end of the treatment: 1) whether or not the youth was living at home (i.e., the adolescent had not been placed out of home); 2) whether or not the adolescent was engaged in school or work for at least 20 hrs/week at the end of the treatment; and 3) whether or not the adolescent had new police contact due to inappropriate or illegal behavior during the treatment period. The therapist registered these treatment outcomes after treatment and in consultation with the primary caregiver. These three outcomes have been operationalized and standardized by MST Services to ensure that these outcomes are scored identically by all therapists (Institute, 2016). This scoring procedure was also followed by FFT. The quality assurance systems of both treatments ensure that their ultimate outcomes are monitored by the therapist, the team supervisor, and the team consultant. Statistical analysis Development of the propensity score The PS is defined as the conditional probability of assignment to an intervention given a set of observed, pre-treatment variables (Rosenbaum & Rubin, 1983). Moreover, the PS is a balancing score which can be used to achieve a balanced distribution for the observed covariates of the treated and control group (Austin, 2011; Rubin, 2001). The PS was estimated in a univariate logistic regression function for the intervention groups. Here, MST is considered the treated group (coded as 1), and FFT the comparison group (coded as 0). This is because, according to the RNR model, adolescents should be assigned to FFT unless there are serious indications to assign an adolescent to MST (Oudhof et al., 2009). The observed pre-treatment variables of adolescents are the independent variables added to the model (Austin, 2011; D'Agostino R.B., 1998; Thoemmes & Kim, 2011). These variables, the potential confounders, were selected for the PS model based on clinical knowledge and their expected relation to at least the outcome, and possibly to the treatment itself (Ali et al., 2015; Austin, 2011; Brookhart et al., 2006; Stuart, 2010). Variables solely related to treatment assignment or influenced by treatment should not be included in a PS model (Ali et al., 2015; Austin, 2011; Brookhart et al., 2006). Weighting by the propensity score The PS was applied by weighting groups by the odds of their estimated PS score (Stuart, 2010). Weighting by their odds was preferred because there were more treated MST cases than control FFT cases and the interest lies in the average treatment effect in the treated (ATT) rather than the average treatment effect (ATE) (Stuart, 2010). The ATT is the average effect that would be found if all adolescents treated with MST had been treated with FFT. The ATE, however, estimates the average effect if all adolescents (MST and FFT) had received MST compared to all of them received FFT (Harder, Stuart, & Anthony, 2010). In other words, because the ATT is estimated, treatment effects for adolescents who received MST are compared with treatment effects that would have been found had they received FFT (Harder et al., 2010; Stuart, 2010). The MST group was therefore weighted with 1, while the FFT group was weighted with the odds of the PS, that is, the