Social mechanisms
One of the key tenets of realism is the very basic idea that observational evidence alone cannot establish causal uniformities between variables. Rather, it is necessary to explain why the relationships come about; it is necessary to establish what goes on in the system that connects its various inputs and outputs. In this manner, physicists are able fully to understand the relationship between the properties of a gas (as measured by the variables—pressure, temperature and volume) using knowledge about the kinetic action of the constituent molecules. In pharmacology, the term ‘mechanism of action’ refers to the specific biochemical interaction through which a drug substance acts on the body to generate its curative effect. Programme evaluators do not suppose that CCTV (the intervention) causes a fall in crime rates (the outcome). It does so, when it does so, by persuading potential perpetrators of increased risks of detection (the mechanism). In all cases, science delves into the ‘black box’. In all cases, the mechanism is what generates the observed relationship.
Whilst it is possible to recognise the affinities in explanatory structure across these examples, they also demonstrate that the action of the generative mechanisms is quite different, to such an extent indeed that that they defy a simple, unitary definition of their nature and content. Pawson expands on the applications of generative vs successive conceptualisations of causation elsewhere [
2].
Readers of this journal will need no reminding that these paradigms have been debated for many years. Realists see physical and social reality as stratified and emergent. Things that cannot be cast as variables yet are vital to explanation (like kinetic forces, cultural norms and human interpretation or agency) are missing from correlational methods. Causal associations themselves are rarely universal; they are adaptive ‘demi-regularities’, which are always strongly influenced by setting and context. The original sources for these arguments may be found in Hesse [
3], Harré [
4], Pawson [
2,
5], Sayer [
6,
7], Bhaskar [
8], Boudon [
9] and Stinchcombe [
10].
We acknowledge the further cleft between ‘critical realism’ and ‘scientific realism’. The writings of Bhaskar [
8,
11] and Pawson [
2] serve as a reasonable proxy for these two schools. They differ on the matter of whether social science can create ‘closed system’ investigations. For Bhaskar, the closed system, experimental control available to the natural scientist is not achievable in social research because of ever-present emergence, that is to say the unique and unceasing human capacity to change the circumstances in which they live. As a ‘substitute’ for closed system empirical enquiry, he thus proposes the usage of abstract,
a priori reasoning and the admission of a moral lens through which to critically evaluate human actions ([
11], p. 64). Pawson, by contrast, argues much more pragmatically that neither physical science nor social science investigation depends on the achievement of closed systems ([
5], p. 67). There are no crucial experiments (most especially Randomised Controlled Trials) which alone furnish us with social laws. But equally, natural science only ever makes slow and imperfect progress in gathering knowledge of the potentially infinite number of contingencies that can shape a physical system. Investigatory closure is always partial. Again, we are presented with rather different visions, the only contradiction occurring when an investigation claims to be
both normative and scientific.
For Archer [
12], collective, constrained decision making is the underlying mechanism that creates all social outcomes. Society is made by but never under the control of human intentions. At any given time, peoples’ choices are conditioned by pre-existing social structures and organisations. We are thus externally constrained in our actions but always part of human agency is the choice to attempt to change the initial conditions that bear down on us. These adaptive choices, over time, go on to mould novel structures and changed institutions. Collectively, our present decisions congregate to form new systems, which in their own turn, constrain and enable the choices of the next generation. Society is thus patterned and re-patterned by wilful action, but as Archer reminds us, the causal outcomes never conform to anyone’s wishes—even the most powerful.
Most realists would affirm this broad account of the mechanisms of social change, where structures shape actions, which shape structure, which shape actions, and so on. There are, however, some significant differences in where they locate the precise locus of that change. For Bhaskar [
8], causal mechanisms sit primarily within the structural component of the social world. They reside in the power and resources that lie with the great institutional forms of society. For other realists, such as Pawson and Tilley [
1], mechanisms are identified at the level of human reasoning. Thus, mechanisms can have different meanings depending on the scope of the intended explanation. Structural mechanisms come to the fore if the social scientist is attempting to explain large-scale social transformations. If, however, the researcher is attempting to discover whether a particular fitness programme creates healthier participants, it can be assumed that key outcomes will result from the reasoning and responses of the participants.
Mechanisms in evaluation
This brings us to a consideration of mechanisms in evaluation research; here the focus is on developing an explanation of how a particular programme works through changing the reasoning and responses of participants to bring about a set of intended outcomes. There have been a number of different conceptualisations of mechanism within evaluation. Chen and Rossi [
13] were among the first researchers to use the term ‘mechanism’ and highlight its significance in theory-driven evaluation [
14]. In 2005, Chen [
15] broadened our understanding of causal mechanisms by identifying two types: mediating and moderating. He defines these as follows:
“A mediating causal mechanism is a component of a program that intervenes in the relationship between two other components . . . [while] the second type of causal mechanism—moderating—represents a relationship between program components that is enabled, or conditioned, by a third factor.” (pp. 240–241)
Weiss [
16] also reflects on mechanisms, in terms of programme theory. She states that it is important to understand the difference between implementation theory and programme theory. The earlier can be conceptualised as a logic model, whereas the latter:
“. . . deals with the mechanisms that intervene between the delivery of program service and the occurrence of outcomes of interest. It focuses on participants’ responses to program service. The mechanism of change is not the program service per se but the response that the activities generate.” (p. 46)
As Weiss [
16] states, mechanisms are not the programme service but the response it triggers from stakeholders and resulting outcome. For example, Vassilev et al.’s [
17] metasynthesis investigated how social networks can make a considerable contribution to improving health outcomes for people with long-term conditions (specifically, type 2 diabetes). They identified three themes which translated into three ‘network mechanisms’:
network navigation (identifying and connecting with relevant existing resources in a network),
negotiation within networks (re-shaping relationships, roles, expectations, means of engagement and communication between network members) and
collective efficacy (developing a shared perception and capacity to successfully perform behaviour through shared effort, beliefs, influence, perseverance, and objectives). The authors highlight not only resources in these mechanisms but also reasoning; these mechanisms convey the close interdependence between social and psychological processes in long-term conditions management. Furthermore, these network mechanisms are subject to context, as the authors state:
“they are shaped by the environments in which they take place which can be enabling or disabling depending on the capacities they offer for carrying out illness management work and supporting behaviours beneficial for people’s health.” (p. 10)
Despite the many different conceptualisations, e.g. [
9,
13-
16,
18], and applications of mechanisms, e.g. [
17,
19,
20], most in some way have been influenced by the critical realism and scientific realism accounts of causation, e.g. [
1,
21,
22], discussed above. In these schools of thought, mechanisms are usually hidden, sensitive to variations in context and generate outcomes. As Astbury and Leeuw [
14] state, mechanisms in realism are:
“underlying entities, processes, or structures which operate in particular contexts to generate outcomes of interest.” (p. 368)
We survey this broader terrain as a prelude to focussing on the more specific version of mechanism thinking referred to by Pawson and Tilley that has come to play a key role in the evaluation of social interventions, namely realist evaluation [
1], which is the main focus of this article.
Mechanisms in realist evaluation
Within the scientific realism approach, Pawson and Tilley [
1] have provided their own conceptualisation of mechanisms; mechanisms are a combination of resources offered by the social programme under study and stakeholders’ reasoning in response [
1]. They state that mechanisms will only activate in the right conditions, providing a context + mechanism = outcome formula as a guiding principle to realist enquiry [
1]. This article sits within the empirical application of realism in the form of realist evaluation and the usage of mechanisms therein. In particular, we make a case for the explicit disaggregation of resources and reasoning in implementation endeavours, to which task we now turn.
The units of analysis within realist evaluation are programme theories—the ideas and assumptions underlying how, why and in what circumstances complex social interventions work. Many readers will by now be very familiar with programme theories expressed as CMOc and with the fact that data collection and analysis in realist evaluation centres on the process of developing, testing and refining CMOc. In the next section of the paper, we propose a development of this formula, which aims to facilitate the study of implementation processes and interventions.