Realist trials and the testing of context-mechanism-outcome configurations: a response to Van Belle et al.

BACKGROUND: Van Belle et al. argue that our attempt to pursue realist evaluation via a randomised trial will be fruitless because we misunderstand realist ontology (confusing intervention mechanisms with intervention activities and with statistical mediation analyses) and because RCTs cannot comprehensively examine how and why outcome patterns are caused by mechanisms triggered in specific contexts. METHODS: Through further consideration of our trial methods, we explain more fully how we believe complex social interventions work and what realist evaluation should aim to do within a trial. RESULTS: Like other realists, those undertaking realist trials assume that: social interventions provide resources which local actors may draw on in actions that can trigger mechanisms; these mechanisms may interact with contextual factors to generate outcomes; and data in the 'empirical' realm can be used to test hypotheses about mechanisms in the 'real' realm. Whether or not there is sufficient contextual diversity to test such hypotheses is a contingent not a necessary feature of trials. Previous exemplars of realist evaluation have compared empirical data from intervention and control groups to test hypotheses about real mechanisms. There is no inevitable reason why randomised trials should not also be able to do so. Random allocation merely ensures the comparability of such groups without necessarily causing evaluation to lapse from a realist into a 'positivist' or 'post-positivist' paradigm. CONCLUSIONS: Realist trials are ontologically and epistemologically plausible. Further work is required to assess whether they are feasible and useful but such work should not be halted on spurious philosophical grounds.