15 General Introduction A second challenge is related to the causal nature of primary research which is particularly desirable to inform decision-making. Contrary to what happens in precursor movements of evidence-based medicine and health technology assessment, most exposures of interest in health and public policy cannot be studied in RCTs or experiments. This is undoubtedly the case of interventions, programmes and policies targeting social determinants of health, in which promoting randomisation would often be unethical (poverty, unemployment), unfeasible or very costly (neighbourhood environment) or impossible (ethnicity, place of birth). Methodological alternatives to evaluate causal effects of these exposures are therefore extremely important for evidence-based policy. These alternatives are called quasi-experiments3 and are observational studies in which, per opposition to randomisation, the investigator does not control the assignment of individuals to the exposure. Instead, the exposure is somehow – either by “accident of chance” or by design – exogenously attributed to certain (groups of) units and not to others. Conditional on some assumptions, the intervention and control groups resulting from the exogenous assignment can be compared to elicit causal effects. How strong these assumptions are depends on the mechanism behind the assignment. For instance, lotteries or weather shocks are examples of “accidents of chance” producing interesting variations to study cause and effect of getting wealthy or migrating, respectively. In this case, the treatment group is indeed “as if” randomly assigned, as individuals cannot manipulate or select themselves into winning the lottery or suffering from a hurricane. There are other cases in which the assignment is not random, and the possibility of mimicking randomisation comes with additional study design features that require some assumptions. This will often be the case of quasi-experiments trying to leverage program or policy roll-out processes that imply local/staggered implementation which creates opportunities for comparing treated and non-treated units. Quasi-experiments represent a great promise for evidence-based policy on health inequalities, in which most exposures of interest are non-randomisable. Furthermore, most quasi-experimental designs can accommodate either stratified analysis by socioeconomic status or interaction terms between socioeconomic status and exposure. This allows the quantification of distributional effects, examining policy impacts on inequalities [64]. Producing quasi-experimental evidence was one of the objectives of this thesis, to strengthen the scarce body of causal evidence about the distributional effects of mental health policy. We were interested in evaluating exposures that might reduce or (unintendedly) widen the mental health gap. Several interesting interventions and policies could not be studied due to the lack of exogenous variation. Others had some potential but were not pursued due to challenges to the utilization of quasi-experimental 3 Quasi-experiments are often also designated as natural experiments, although some authors will recognise differences between these terms. 1
RkJQdWJsaXNoZXIy MjY0ODMw