The ontology of critical realism, with its identification of numerous causal mechanisms interacting in different ways in different contexts to produce different outcomes, has influenced the development of realistic evaluation, as advanced by Pawson and Tilley (1997). The aim of realistic evaluation is to explain the processes involved between the introduction of an intervention and the outcomes that are produced. In other words, it assumes that the characteristics of the intervention itself are only part of the story, and that the social processes involved in its implementation have to be understood as well if we are going to have an adequate understanding of why observed outcomes come about. In contrast to the assumptions of constant conjunction, realistic evaluation posits the alternative formula of:
Mechanism + contex = outcome
In any given context, there will in all likelihood be a number of causal mechanisms in operation, their relationship differing from context to context. The aim of realistic evaluation is to discover if, how and why interventions have the potential to cause beneficial change. To do this, it is necessary to penetrate beneath the surface of observable inputs and outputs in order to uncover how mechanisms which cause problems are removed or countered by alternative mechanisms introduced in the intervention. In turn, this requires an understanding of the contexts within which problem mechanisms operate, and in which intervention mechanisms can be successfully fired. In other words, realistic evaluators take the middle line between positivism and relativism, in that positivism’s search for a single cause is seen as too simplistic, while relativism’s abandonment of any sort of generalisable explanation is seen as needlessly pessimistic. In contrast to these two poles, realists argue that it is possible to identify tendencies in outcomes that are the result of combinations of causal mechanisms, and to make reasonable predictions as to the sorts of contexts that will be most auspicious for the success of health-promoting mechanisms. The confidence of prediction can be increased through comparison of different cases (i.e. different contexts) in that concentration on context–mechanism–outcome configurations allows for the development of transferable and cumulative lessons about the nature of these configurations.