Table 1 Key principles underpinning realist research [3, 11, 18, 19] and [20]

• Realism asserts a reality exists independently of the observer: both the material and the social worlds are “real”, at least in the sense that anything that can cause observable outcomes is itself real. • Knowing reality through science is unavoidably relative to the researcher: developing knowledge on reality is constrained by perception and cognition, and is socially constructed. Nonetheless, reality constrains the interpretations that are reasonable to make of it, meaning that it is possible to move gradually closer to an understanding that better reflects the reality under study. • According to realism, the world is differentiated and stratified, consisting not only of observable and measurable events, but also of structures, which have powers and liabilities capable of generating events. These structures may be present even where, as in the social world and much of the natural world, they do not generate regular patterns of events [20]. • Causality concerns not a direct relationship between two observable and discrete events, but a relationship between “the ‘causal powers’ or ‘liabilities’ of objects or relations, or more generally, their ways-of-acting of ‘mechanisms’” [20] and the outcomes of those mechanisms. • Context matters—a lot. Contextual conditions may have an influence on the implementation of the intervention, may or may not provide the necessary conditions for the mechanism that will be triggered and thus may have an effect on the observed outcome. An interaction is always present between context and mechanism. Context includes features such as social, economic and political structures; social policies; organisational context; participants; staffing; and geographical and historical context. • The interaction among intervention (strategies), actors, context and mechanism is what creates the intervention’s impacts or outcomes. • Since interventions work differently in different contexts and through different mechanisms, they cannot simply be replicated from one context to another and automatically achieve the same outcomes. Theory-based understandings about the influence of contexts on mechanism and resulting outcomes (i.e. “what works for whom, in what contexts, why and how” are, however, transferable. • Realist research is theory incarnate. It starts with a fragile theory and ends with a fallible model. Theory is to be understood as theories of the middle range, as defined by Merton [21], a bundle of hypotheses that can be tested empirically [22]. As such, these theories are expressed in such a way that they can be supported, refuted or refined against empirically derived data. Often, the term ‘programme theory’ is used to denote the starting point of a realist inquiry, which is one or more testable hypotheses that spell out how intervention strategies are expected to lead to the outcomes, for whom, in what conditions, how and why. • Realist research progresses through choosing methods that will provide the data needed to help “test” the initial programme theory in terms of effectiveness (Did the programme achieve its goal?) and of causal processes (How did the observed results come about, in which context, why and for whom?). Realist research is methodologically promiscuous, potentially using quantitative and qualitative data to test theories. • In summary, in the account of realism we propose here, outcome patterns do not occur directly because of the intervention strategies that are used. Instead, outcomes are caused by invisible mechanisms that are sensitive to context. The interactions and influences between context, mechanism and outcome may be explained by one (or more) middle-range theories. To develop, support, refute or refine these theories, both quantitative and qualitative data may be needed.