Regional differences in prescription behaviour: Inducing physicians or demanding patients?
In this project we estimate the marginal treatment effects of an invasive heart attack treatment option compared to conservative treatment. Regional variations in access to invasive heart attack procedures are used as an instrumental variable. We use administrative data of the universe of all hospital inpatients in Germany from 2005 to 2007.
Estimation of the effectiveness of medical interventions is at the heart of health economics. While the randomized controlled trial is typically considered the “gold-standard” of evaluation, often it is not feasible. Other techniques, like instrumental variables estimation (IV), are then used to estimate the causal effect of a medical intervention on outcomes like health or costs. These techniques account for the problem that – if individuals are not randomized into the treatment – recipients of the treatment often differ from those who do not receive it, based on unobservable characteristics.
IV estimation may provide a high internal validity of estimated effects. Yet, a growing literature criticizes a rather limited external validity. If the treatment effect differs by individuals – for instance, some strongly benefit, others do not or are even worse off – IV estimates identify a local average treatment effect (LATE), an average effect of a certain subgroup of individuals which differs from the subgroup of the treated.
This might induce two problems: First, policy makers are usually interested in average treatment effects of the entire population (ATE) or of the treated (ATT), numbers that might differ from the local average treatment effect. Second, effect heterogeneities play an important role, particularly in the evaluation of medical interventions. Positive average effects might not be sufficient if there is also a large fraction of individuals that is actually harmed by the treatment.
The estimation of marginal treatment effects deals with these two issues, allows to inspect heterogeneities and average treatment effects.