Hester van Eeren

| Chapter 2 2 | 22 were run to demonstrate model behavior under different assumptions. As the transition probabilities constitute important model parameters, a scenario was created in which probabilities for both interventions were equal. Subsequently, the intervention costs are important parameters, since systemic interventions are concerned to be relatively costly (Aos et al., 2004). From a societal perspective, family costs are assumed to be important, therefore it was investigated how exclusion of these costs would influence the results in the third scenario. Results The resulting health economic model for systemic interventions showed that modelling an intervention with a primary aim of decreasing delinquency was feasible. Based on the illustrative comparison of FFT versus TAU, costs and effects could be expressed in costs per CAFY. This section elaborates on the specific characteristics of the resulting decision analytic model. Obviously, the combination using different sources for the inputs of a model is certainly not without problems, but we stress that the emphasis here was on building an illustrative model and demonstrating the model functioning. Model structure Estimates of long-term effects were essential to the analysis and were taken into account in the current model. This required some (informed) assumption regarding the endurance of effects of treatment also taking into account the influence that reaching a certain age or experiencing certain life events may have on criminal behavior (Farrington, 2003). For the current model, information on these parameters was taken from the literature. Moffitt (1993) roughly suggested that after adolescence or at approximately age 30 subjects who are criminal during their entire life, life-course-persistent offenders, will remain criminal and subjects who only show criminal behavior during their adolescence, so-called adolescence-limited offenders, will have returned to non criminal behavior. This implies a stable state of criminal activity among individuals of age 30 and older. To illustrate the option of incorporating earlier theory and evidence on the development of offending and antisocial behavior we integrated parts of the long-term stabilizing effects described by Moffitt (1993) into the current model framework. This effect is implemented in the model by extending the effectiveness of the treatment till the age of 30 years. Consequently youth remain in their current state after that age. Thus after reaching the age of 30, youth reach a stable state in their criminal behavior, which means the transition probabilities in the model are from then on defined by mortality rates only. The time horizon of the model is 50 years. To illustrate how long-term effects may influence model results, Figure 2 and Figure 3 present the percentage of youth in each model state over the time horizon of the model, for FFT and TAU respectively. Figure 2 and Figure 3 demonstrate that a stable state is already reached after about 1 year, which implies that the actual impact of the incorporation of a stabilizing effect, based on the theory of Moffit (1993) is minor in this model. However, as the model results are only as good as the available input used to fill