Abstract
Understanding the mechanisms and contexts that drive the success of complex health interventions remains a challenge. Realist methods, grounded in scientific realism, generate context-sensitive programme theories to explain how, why and for whom interventions work, but these theories often lack structured operationalisation to inform comparison or intervention design. Behavioural science, by contrast, systematically identifies and modifies behaviour change mechanisms using theory-driven frameworks, but has been criticised for insufficiently considering context. Integrating these may enhance the precision, standardisation and applicability of realist programme theories. This novel approach leverages behavioural science concepts such as behaviour change techniques and mechanisms of action to clarify mechanisms, and uses the idea of behavioural settings to explicate context. Together, this establishes a common language for programme theory formulation, making them more structured, testable and transferable. A five-step framework for integration is proposed for realist studies, facilitating more precise and transferable theories that support intervention design and policy translation.
Keywords
Background
Many interventions in public policy and/or health services research are developed by healthcare or public health staff who may not have formal training in intervention design or evaluation methodologies, nor behavioural or implementation science (Skivington et al., 2021). This often results in complex interventions, seeking to change behaviour, that are opaque to researchers or evaluators who try to understand how, why or if they have worked. To tackle this issue, both behavioural science and realist methods have been used differently in syntheses and evaluations to try to unpick what interventions have been doing and how, why and for whom they may work in particular contexts (Crawshaw et al., 2023; Maben et al., 2023). While behavioural science is intended for use both in intervention design and in reviews (Bryan et al., 2021; Crawshaw et al., 2023; Michie and Johnston, 2012), it has been criticised for insufficiently considering the role of context, limiting understanding of how interventions may work in different circumstances. Meanwhile, realist methods consider context at their core, but, in practice, the bespoke nature of realist programme theories has limited ability to transfer and synthesise insights. This paper sets out a framework for how these two seemingly different approaches can be used together to address their respective weaknesses.
Introduction to realist methods
Both scientific and critical realism have contributed to modern use of realist methods, which comprise both realist evaluations and syntheses (Mukumbang et al., 2023). Pawson and Tilley originally grounded these methods in scientific realism, which argues that the universe described by our scientific theories exists objectively and independently of human minds, beliefs, perceptions or conceptual schemes. However, the entities and processes posited by mature scientific theories are often not directly observable (like electrons, genes or black holes), but nevertheless exist and have the same ontological status as observable things. The theoretical claims of science are considered genuine assertions about the world that are either literally true or false as they purport to describe the real world. The ability of science to predict, explain and manipulate phenomena is therefore best explained by the fact that it is uncovering true (or nearly true) knowledge about reality (Boyd, 1980, 1983; Psillos, 2005).
Thus, realist evaluation and realist synthesis aim to uncover not just whether an intervention works, but how, for whom and under what circumstances (Pawson and Tilley, 1997). Realist methods treat interventions as complex, contingent processes where mechanisms are triggered in specific contexts to generate particular outcomes in a process known as ‘generative causation’ (Dalkin et al., 2015). In complex cases, multiple causes may intertwine, which is called configurational causation, as causal factors can work in tandem as ‘configurations’ to produce outcomes (Ragin, 2014). Programme theories (PTs) are developed in both realist evaluation and synthesis, to explain how, why, for whom and in what circumstances the programme causes particular outcomes. These are typically summarised in a heuristic known as a CMOC or context-mechanism-outcome configuration. This perspective has made realism particularly influential in health services and public health research, where interventions often interact in a complex manner with local conditions, requiring explanatory depth beyond simple cause-and-effect models (Jagosh et al., 2022; Marchal et al., 2012).
Focus of behavioural science
While realist methods seek to explain how and why interventions produce varied outcomes across different contexts, behavioural science focuses on identifying the mechanisms that drive individual behaviours and designing interventions to modify them (Michie et al., 2021a; Michie and Johnston, 2012). Rooted in psychology, cognitive science, and behavioural economics, behavioural science applies structured, theory-driven approaches to behaviour change, often through predictive models and experimental testing (Michie et al., 2014). Unlike realist methods, which explore how mechanisms interact with social and organisational contexts, behavioural science typically assumes that certain behavioural mechanisms operate consistently across settings. This is primarily due to a lack of explicit consideration of the environmental and institutional contexts of behaviour. A common limitation of traditional approaches in behavioural science (e.g. social cognitive theory (Bandura, 1986), health belief model (Rosenstock, 1974) and transtheoretical model (Prochaska et al., 1992), as well as ‘nudge’ theory (Thaler and Sunstein, 2009)) has been an almost exclusive reliance on identifying the psychological factors associated with behavioural production, epitomised by their ‘methodological individualist’ stance. More recently, however, an awareness that all behaviour is situated in some sort of environmental context has crept into theorising about behaviour, thus moving beyond individual psychology towards addressing structural drivers (Bronfenbrenner, 1979; Hayes et al., 2012). This has manifested strongly in Theories of Change, where assumptions between mechanisms and outcomes need to be considered (Mayne, 2023). Assumptions are made about causal connections, events and conditions that need to be realised for the intervention to work – which map strongly onto realist conceptualisations of contextual influences. However, a real strength of behavioural science is in systematically understanding, unpicking, and developing common languages for what intervention components are, how they might work, and what outcomes they may have – which are deficiencies common with realist PTs (Renmans et al., 2024).
Comparing the approaches
Realist methods thus often produce rich, context-sensitive PTs, but without structured operationalisation; they can lack utility in future intervention design (Marchal et al., 2012) or can lack general comprehensibility by others (Renmans et al., 2024). Issues highlighted in realist methods include difficulties in: defining mechanism and context, distinguishing mechanism from context, settling on the level of abstraction required by the final PT, and describing the intervention components being analysed (Marchal et al., 2012). For example, in 121 definitions of the term ‘mechanism’ reviewed by Lemire et al. (2020) across realist studies, a third of these were analogous to programme components. Others also corroborate these issues with distinguishing context from intervention components in realist research (Greenhalgh and Manzano, 2022; Nielsen et al., 2022). Furthermore, Renmans et al. (2024) highlight that many realist evaluations ‘do not use scientific concepts to describe mechanisms’ and that there is a ‘lack of explicit intervention families’, which limits transferability of PTs (Renmans et al., 2024). This can also result in PTs that are themselves not testable or falsifiable, failing a core principle of the philosophy of science (Lemire et al., 2020). Table 1 highlights further differences – and areas of complementarity – between these approaches.
Realist versus behavioural approaches.
To date, both realist methods and behavioural science have been used largely independently. However, in recent years, there have been a few initial forays into combining realist methods with behavioural or implementation science frameworks. For example, one realist synthesis seeking to understand the behavioural mechanisms supporting practitioners to taper opioid use sought to make use of the Theoretical Domains framework to elucidate mechanisms of action (MoAs; Bhattacharya et al., 2022). This was found to complement the realist conceptualisation of mechanism. Another realist-informed ethnographic study of exercise referral schemes drew on the behavioural science theory of COM-B (Capability, Opportunity, Motivation – Behaviour) to categorise and better understand mechanisms (Downey et al., 2021), aligning with Mayne’s general approach of using behavioural science (particularly COM-B) to ground a PT in the form of a theory of change (Mayne, 2015). Frameworks from implementation science such as normalisation process theory have also been explored for their utility in complementing realist PT development (Dalkin et al., 2021; McConnell et al., 2025).
As these authors have identified, the strengths of both approaches are highly complementary. Furthermore, with respect to the fundamental notion of causation, both RE and behavioural science consider multiple, interacting causes in their explanatory account, a consistency upon which it is possible to build. It may then follow that integrating them can enhance the ability to understand what an intervention is doing and how and why it works (via behavioural science) as well as for whom, and in what contexts it is working (realist methods). However, to date, no one has set out a definitive framework for doing so. In the present article, we draw upon concepts from behavioural science to explain how realist PTs could be augmented. This includes introducing the concept of behavioural settings to further explicate context, as well as the concepts of behaviour change techniques (BCTs – the resources introduced by an intervention) and psychological MoAs to elucidate mechanisms (the psychological response to the resources). We then propose a recipe for a behaviourally informed PT development process. This article is intended to provide a resource for use in realist studies, as well as to stimulate further debate in the realist literature regarding the benefits and drawbacks of moving towards standardisation of PTs and a common language for describing them.
Methods
Operationalising context
Both context and mechanism have been subject to debate in the wider academic literature (Dalkin et al., 2015; Greenhalgh and Manzano, 2022; Nielsen et al., 2022). Others have highlighted how context in realist research has concentrated around two main ‘narratives’: (1) context as ‘“observable features” or “things”’ and (2) context as ‘“relational and dynamic” features or “forces/interactions”’ (Greenhalgh and Manzano, 2022). Greenhalgh and Manzano (2022) recommend that the latter conceptualisation of context as a ‘force’ can enhance the explanatory power of PTs, and that context is a constituent and interconnected element of a PT (Greenhalgh and Manzano, 2022). We believe that behavioural science offers concepts that complement both the conceptualisation of context as ‘things’, and as ‘forces’, through behavioural settings, as well as the wider context (Aunger, 2020b).
A particularly powerful concept which usefully concretises aspects of context via a limited number of specific factors is the notion of a ‘behaviour setting’ from mid-20th century ecological psychology (Aunger, 2020b; Barker, 1968). Behaviour settings are defined as ‘a specific and bounded type of situation in which objects, places and people interact to achieve a common purpose’ (Aunger and Curtis, 2016). Based on generalisation from a huge corpus of real-world behaviour, the behaviour setting concept draws attention to the strong interactions between individual role-playing and environmental supports (called ‘synomorphies’), such as tools which facilitate role performance in directing behaviour within a specific situation (Barker, 1968). Relevant environmental factors can be divided into physical infrastructure such as buildings and manipulable props like tools, as well as social factors such as role-associated norms. Roles themselves are related to particular motives for action, as well as requiring capabilities (which might be specialised forms of knowledge or physical skills) for successful performance. The behaviour setting concept can thus help clarify differences between aspects of context as ‘things’ or as ‘forces’.
However, to be causally precise (as realist theory demands) requires differentiating between different points in chains of causation. Interventions can be delivered distally and/or proximally in such chains. For example, policy-based interventions can change the rules by which an institution operates (e.g. by requiring new procedures that ensure patient safety in hospitals) (Aunger et al., 2022). Another type of intervention can impact directly on the way in which some component of a setting itself operates (e.g. by requiring a nurse to log onto the electronic patient database with a password). We argue that proximal interventions operate within the behaviour setting itself, in themselves affecting behaviour (through MoAs), while interventions affecting wider context can filter down to alter the components of a behaviour setting, later influencing the way in which settings operate (Aunger, 2020b).
The general problem is that there are many kinds of factors considered by different theorists to play some role in determining behaviour; the context/setting distinction is one way of bringing this list into control in three ways: by limiting the number of total factors to be considered relevant, categorising them in a way designed to bring ontological and conceptual clarity to their causal roles, and clarifying where in the causal chain interventions are exerting their effects. This separation of setting from context is consistent with the need highlighted by Nielsen et al. (2022) to ‘catalogue CMO configurations and accumulate knowledge of how they work across a broad range of contexts’ (p. 107).
Therefore, we argue that context, in an integrated realist and behavioural science approach, should be conceptualised as a multi-layered and dynamic system of influences that shapes the operation of behavioural settings and, consequently, the activation of behavioural mechanisms and resulting outcomes. Causal chains thus encompass both:
Distal contextual forces: These are the broader institutional, policy-based, cultural, economic and social factors that operate outside the immediate behavioural setting but exert a significant influence on its structure, resources, norms and the roles of individuals within it. These forces can be understood as establishing the boundary conditions for how behavioural settings function. Examples include: • Institutional policies and regulations: As mentioned, these can mandate procedures or norms that shape behaviour within a setting (e.g. hospital safety protocols). • Cultural norms and values: Wider societal beliefs and practices can influence the acceptability and prevalence of certain behaviours within specific settings (e.g. norms around communication styles in a workplace). • Economic conditions: Resource availability, funding models and market forces can impact the resources and tools present within a setting (e.g. availability of specialised equipment in a factory). • Social structures and inequalities: Power dynamics, social hierarchies and disparities can shape the roles and interactions within a setting (e.g. influence of hierarchical structures on team communication). Wider insights from sociological theories can also be incorporated here. • Technological developments: The availability and nature of technology (e.g. electricity grids, Internet access) can fundamentally alter the components and operation of a setting. However, factors such as access or barriers to a technology can be a component of a setting.
Proximal context aka the behavioural setting: This refers to the specific, bounded situation characterised by the interaction of objects, places and people that influence activation of mechanisms for programme participants (Aunger, 2020b). It is within the setting that the immediate cues, affordances, constraints and social norms directly influence behaviour through the activation of psychological MoAs. The characteristics of the behavioural setting are, however, shaped and influenced by the distal contextual forces. Key elements of the behavioural setting include: • Physical infrastructure and props: The tangible elements of the environment (e.g. layout of a classroom, tools in a workshop). • Social factors and role-associated norms: The implicit and explicit rules and expectations governing behaviour within the setting, often tied to the roles individuals occupy. • Participants and their roles: The individuals present, and the expected behaviours and motivations associated with their positions within the setting.
When analysing this approach through the lens of the broader realist literature, this conceptualisation strongly aligns with Greenhalgh et al.’s two conceptualisations of context as both ‘observable features’ (most often found in the setting) and as a ‘force’ (most often in the distal context) with a dynamic and relational nature (Greenhalgh and Manzano, 2022). The distal context, encompassing institutional and societal factors, represents these broader ‘forces’ that are often further along the causal chain, and shape the operationalisation of the more immediate behavioural setting. The setting, in turn, comprises ‘observable features’ such as objects, props and social roles and is where mechanisms are activated within the actors. This clarifies that the context component of a realist PT should not contain intervention components. By understanding that interventions modify a context to drive behaviour change, we then start to address issues of tangling together intervention components and context in realist PTs (Renmans et al., 2024).
Operationalising mechanisms
Behavioural science traditionally has focused more on individual-level behaviour, while realist methods typically situate mechanisms within broader social and organisational contexts. Realist methods can therefore be more abstracted and generate theory at higher levels of operation (i.e. beyond the individual) (Greenhalgh and Manzano, 2022). However, while context shapes and constrains behaviour, it does not act independently in itself; rather, people respond to it – or to changes to it – and change their reasoning and then their behaviour as a consequence (Dalkin et al., 2015). This is reflected in a paper by Dalkin et al., which outlines this debate within the realist space. Dalkin et al. highlight that some argue that ‘causal mechanisms sit primarily within the structural component of the social world’ whereas others, such as Pawson and Tilley themselves, argue that ‘mechanisms are identified at the level of human reasoning’ (Dalkin et al., 2015). We would argue that, indeed, all behaviour change occurs, in the end, at an individual level – that is, in people’s brains, but often in response to more distal causes. Although this happens at an individual level, this does not preclude mechanisms from emerging or occurring across a wider group of people. Indeed, realist PTs are often reported at a certain level of social stratification depending on, for example, if a single intervention or a group of interventions is being analysed. Nonetheless, concepts often described in realist methods (e.g. power dynamics, accountability, team culture) necessarily influence individual cognition and motivation rather than directly causing action.
Behavioural science offers many evidence-based frameworks setting out what psychological processes have been evidenced to change in response to various types of interventions across fields (MoAs), as well as what intervention components have been shown to exert these effects (tactics/BCTs) (Michie et al., 2021a; Renmans et al., 2024). These frameworks help clarify what can be considered an intervention component (tactics/mechanisms as a resource), versus what changes in reasoning can reasonably occur in response (mechanisms as a change in reasoning) (Dalkin et al., 2015). These changes in reasoning, called MoA within behavioural science, then cause a change in behaviour (Figure 1).
Basic depiction of process of changing behaviour (without considering context).
Frameworks including COM-B (Michie et al., 2011), Behaviour-Centred Design (Aunger, 2020a, 2020b; Aunger and Curtis, 2016), PRECEDE-PROCEED (Green and Kreuter, 1991) and Intervention Mapping (Kok et al., 2016) offer structured approaches to identifying and systematically applying behaviour change strategies to the description of interventions (Michie et al., 2014, 2017). There are various competing taxonomies of active intervention components or ‘tactics’ including ‘behaviour change techniques’ (associated with COM-B) (Marques et al., 2024), ‘behaviour change methods’ (BCMs, associated with Intervention Mapping) (Kok et al., 2016) or ‘evolutionary learning processes’ (ELPs, associated with Behaviour-Centred Design) (Crutzen and Peters, 2018). BCTs can be defined as ‘observable, replicable, and irreducible components of an intervention designed to alter or redirect causal processes that regulate behaviour’ (Michie et al., 2013). The number of such tactics ranges from 9 for ELPs, to 284 for BCTs, reflecting different intellectual foundations for such concepts.
For example, the BCT Taxonomy v1 offers a standardised, hierarchically structured classification of 93 discrete techniques used to change behaviour (Michie et al., 2013). These techniques are organised into groupings based on their function (e.g. goals and planning, feedback and monitoring, social support) and are designed to be observable, replicable and theory-linked. These BCTs often connect to other ‘grand theories’ such as social cognitive theory and the theory of planned behaviour (Michie et al., 2021b). The BCT Taxonomy aligns closely with the COM-B model (Michie et al., 2011), which posits that behaviour is influenced by Capability, Opportunity, and Motivation. Each BCT can be mapped to one or more components of COM-B, providing a systematic way to select and report intervention content that targets specific behavioural determinants. More recently, the BCT Taxonomy is being expanded to become the BCT Ontology (BCTO) – which comprises 284 (and counting) BCTs (Marques et al., 2024). The BCTO is intended to be a continually developing resource that will be readable by future machine learning systems that may help automate processes such as systematic reviews. Figure 2 depicts an example of how the BCTO is organised. We adopt the BCTO for use in formulating behavioural PTs in this framework.
Example portion of the hierarchical nature of the BCT Ontology for the parent class ‘goal directed BCT’. Depicts 25 total BCTs out of 284. Generated at https://bciovis.hbcptools.org/.
We posit that, in interventions, these tactics can attempt to modify either the distal context (in policy-level interventions) or the behavioural setting (in more proximal interventions) to produce changes in people’s psychological processing and therefore behaviour. If successful, these tactics, depending on the context and setting, then cause a change in reasoning in intervention participants via the MoAs mentioned earlier. In this conceptualisation, the activation and effectiveness of tactics are still wholly contingent on context and behavioural setting and aspects thereof. This particularly applies to the interpretation of, and engagement with, the intervention by actors playing specific roles within the setting – as is the case in wider realist methods.
Tactics are observable phenomena in the world (analogous to mechanisms as a resource), while MoAs are the invisible, aforementioned changes in reasoning that constitute hypothesised psychological constructs or processes (Schenk et al., 2024). Certain tactics have been evidenced to trigger particular MoAs (but not others) (Schenk et al., 2024). For example, behavioural modelling (a BCT) might work via the mechanism of observational learning (see Table 2 for definitions). There is also a MoA Ontology which contains 261 unique mechanisms that correspond to BCTs in the BCTO (Schenk et al., 2024). Both the MoA and BCT Ontologies are a living resource that are expanding as more of each are identified.
Clarification of key concepts as used in this article.
Ingredients required to formulate behavioural PTs
We propose that an integrated realist-behavioural science PT development process should make use of these frameworks as a common language for describing PTs. Therefore, they should incorporate three key elements, outlined below.
First, factors influencing the intervention should be conceptualised and clarified as belonging to wider context or behavioural setting (depending on the locus of the intervention). This involves identifying the contextual and/or setting-based elements that a priori seem relevant to the behaviour change task in hand. Next, authors should make reference to ontologies of evidence-based tactics (e.g. BCTs) and MoAs to provide structure and standardisation, to establish a firm basis in psychological evidence for causal mechanisms. This involves examining taxonomies of BCTs and MoAs which offer a way to systematically define and classify mechanisms, reducing the risk of realist PTs becoming vague, inconsistent or non-scientific. Similarly, the MoA framework helps identify the underlying cognitive and affective processes that drive behaviour change (in response to the intervention tactic), ensuring that realist explanations are grounded in empirical psychological constructs. These ontologies serve as essential tools for mapping intervention components to mechanisms in a systematic and reproducible way.
Second, ‘substantive theories’ such as COM-B, Behaviour-Centred Design or Intervention Mapping could be used to offer a broader conceptual foundation with explicit appreciation of the contextual nature of behaviour – and incorporation of these should be considered in a manner consistent with existing RAMESES guidance on including substantive theory (RAMESES II Project, 2017b).
Third, realist reasoning and PTs are still needed to explain how these mechanisms operate in specific contexts. For this purpose, CMOCs are typically used. While these ontologies provide a foundation for defining mechanisms, PTs articulate how these mechanisms interact with different contextual conditions to produce outcomes. This ensures that the contingent nature of interventions is preserved, rather than assuming universal applicability. The behaviour setting concept can be used to identify specific contextual factors that play a part in the causal chain, from intervention tactic to context to MoA to behaviour. Therefore, to incorporate behavioural concepts, we are proposing a formulation termed‘ augmented causal chain configurations (CCCs)’ that take into account behavioural setting, context, BCTs and MoAs – and whether interventions seek to modify either the context or the behavioural setting or both. This typically will depend on whether they are a policy-level intervention (context) or organisational/individual-level intervention (setting). Figure 3 depicts the components of a CCC and their locus of operation. The following section outlines how CCCs can be formulated. It draws on a conceptualisation of the causal chain similar to ‘Context + Intervention → Mechanism → Outcome’ outlined by Lemire et al., merged with insights from Dalkin et al. about conceptualising mechanism as resources (e.g. BCTs/tactics) and reasoning (MoAs) (Dalkin et al., 2015; Lemire et al., 2020).
CCC components and their loci of operation.
Formulating augmented behavioural CCCs
Formulating augmented CCCs will require adoption of a more disaggregated conceptualisation of what is happening when an intervention occurs. As such, a standard formulation of an interventionist behavioural programme would require the explicit recognition of new steps in the causal chain as follows:
Note that interventions can take place at various points in the causal chain, as discussed earlier – especially when the behavioural setting is unknown or can vary. For example, an intervention which targets policy or institutional regulation would work as follows (potentially across many settings):
We would encourage specifying the actor or group of actors for whom the mechanism of action works (or not) within the MoA portion of the CCC – otherwise it may not be clear ‘for whom’ the intervention may work. For example, to consider group-level MoAs, the ‘actor’ can also be considered to be a group of people in which the mechanism is activated (e.g. ‘community health workers’ in example CCC 2 below).
Selection of the relevant BCT and its related Mechanism of Action requires comparison against the Theory and Techniques tool (https://theoryandtechniquetool.humanbehaviourchange.org) and/or the more updated MoA ontology ( https://osf.io/h4sdy/files/pkq4e ) (Schenk et al., 2024). This is because BCTs have evidence that they work through certain MoAs – and not others (Connell et al., 2019). As such, comparison against this tool ensures CCCs remain evidence-based – addressing a critical deficiency found in some realist PTs (Connell et al., 2019; Renmans et al., 2024).
For some worked examples, we have drawn upon the BCT Taxonomy to highlight a few transformed CCCs. The examples in Table 3 are taken from the authors’ existing work, as well as a recent review of CMOCs in the literature which provided ‘good’ examples of CMOCs rated by expert consensus (Renmans et al., 2024). Example 4 shows a case where a tactic triggers a different mechanism than intended, due to unforeseen contextual conditions.
Comparison of CMOCs with CCCs.
It is important to note that an important aspect of realist analysis to capture is the notion that a well-planned intervention can have unintended consequences based on unexpected mechanism-context interactions (as well as issues such as logistics and timing). CCCs can capture this occurrence by demonstrating that:
A tactic may fail to modify a context in an intended manner, or the tactic may be implemented incorrectly.
A tactic may be ‘blocked’ by an aspect of the context and/or behavioural setting that was not explicitly targeted by the intervention.
An unintended MoA may trigger, or the intended MoA may fail to trigger, due to particular aspects of context or setting. For example, see CMOC 4 above.
An outcome may occur that was not the same as that intended by intervention designers.
These possibilities are indicated by the points of failure in Figure 3.
To achieve this, an integrated realist/behaviourist PT should use a systematic process which integrates these various ingredients into a common theory development framework. We have formulated a five-step process termed Generate, Refine, Align, Formulate and Embed. We have depicted this process in Figure 4 and described it in Table 4.
Diagram to depict the methodological process of building a behavioural programme theory depicted by a CCC.
Five-step process for generating behavioural programme theories.
The final output of this process would be a behaviourally informed realist PT (CCC) that explicitly connects and incorporates specific contexts and their behavioural settings with all their components, improving their precision, comparability and transferability.
Discussion
The method outlined here provides a framework that can improve the basis in psychological evidence, and improve the standardisation of language and transferability, of realist PTs. It combines strengths of realist methods such as strong consideration of context and unintended outcomes, with the objective descriptions of ‘what interventions are comprised of’ that behavioural science affords. As efforts are still ongoing to minimise confusion between contexts, situations or settings, mechanisms and outcomes (Crawshaw et al., 2023; Dalkin et al., 2015; Nielsen et al., 2022; Renmans et al., 2024), we hope this framework presents a useful method and resource for those who seek to make their PTs more precise, and the constructs within them, more objectively defined.
To achieve this, we have clarified context as comprising both wider distal factors and proximal behavioural setting – either of which can be modified with a range of interventional tactics. How much of the causal envelope a particular programme must consider depends in large part on where in the causal chain the intervention is being placed. Interventions can target modification of both context and setting, or either on their own. If the target is a change in regulations, which is quite distal to behaviour, programme designers should probably consider both context and setting (or only context), whereas in more local interventions, modifying a piece of the setting’s infrastructure can focus less on the larger context, unless it is identified through retroductive reasoning to influence behaviour.
To clarify and enhance the concept of a realist mechanism, we have adopted frameworks to elucidate both the resources offered by the intervention in terms of the BCT (the tactic) and the change in reasoning – or cognitive process that occurs – to change behaviour (the MoA) – providing a ‘common language’ that could underpin realist PTs. While this is already complex, there are yet more concepts from behavioural science that are important to be aware of. For example, mode of delivery of the tactic can be important to consider. In particular, tactics can be delivered as a message, object/tool or modification of the environmental context, which could significantly change the effectiveness of the tactic/BCT. However, we are not advocating all behaviour change concepts in a CCC be included – unless these are found to meaningfully impact intervention success or failure – otherwise the CCC could become too unwieldy. Instead, these should be considered as an aspect of the context affecting intervention success or failure.
Applying the framework
We have put aspects of this framework to the test in a recent series of publications seeking to understand interventions to reduce unprofessional behaviours such as bullying and harassment between healthcare staff (Aunger et al., 2024; Maben et al., 2023). The first step was to produce a realist review which resulted in a range of PTs (Maben et al., 2023). The second was a separate analysis of the same identified interventions, which sought to understand the BCTs used within them (Aunger et al., 2025), while testing the suitability of the updated BCTO for use in health services research. This research proved two basic requirements necessary for integrating behavioural science into realist methods: that interventions explained previously in realist terms can have their mechanisms also described in behavioural science terms, and that behavioural science frameworks are now ready for application in areas outside of purely health-related research. Taken in combination, our realist synthesis outlined essential contexts and implementation principles necessary for intervention success, while the coding against the BCTO (Marques et al., 2024) explicated exactly what the interventions tried to do. We are now using this behaviourally aligned understanding of existing unprofessional behaviour interventions in follow-up grants to both inform the design of our own intervention, and to understand what contextual conditions made existing interventions fail or produce undesired outcomes. The framework requires yet more testing in novel realist evaluations and syntheses.
Prior attempts to improve standardisation and build a common language for PTs
Others in the realist space have sought to minimise some of the limitations of the way realist methods have been generally applied. For example, there have been efforts to reverse-engineer a database of PTs. This effort attempted to address a range of issues in realist research that we also highlight here, including the notion that ‘mechanisms should be distinguished from actions or programme components’, that ‘many REs [realist evaluations] do not use scientific concepts to describe mechanisms’ and that mechanisms should ‘include stakeholder reasoning’ (Renmans et al., 2024). However, due to workload and other reasons, this effort was abandoned (Renmans et al., 2024). They also acknowledged that a lack of standardisation, frameworks and explicit ‘intervention families’ held back the efforts to formulate this database (Renmans et al., 2024). However, these authors went on to highlight that behaviour change taxonomies do exist in behavioural science (Kok et al., 2016), alluding to their potential use in realist methods.
Rather than trying to resolve issues with realist PTs such as non-scientific concepts in realist evaluations by retrospectively developing a database, we argue that it may be better to propose a solution others can adopt on a prospective basis. A ‘bottom-up’ approach as outlined here could work better. For example, if realist methods users were to prospectively adopt this pre-existing common language with which to describe intervention components and mechanisms, we could begin to move realist research towards more evidence-based theories, with greater objectivity, and cross-over with intervention design language.
This would then help transferability of insights from realist syntheses and evaluations directly into enabling intervention adaptation elsewhere. Indeed, many realist syntheses and evaluations are conducted with the intention of informing future intervention design (Pearson et al., 2015). By identifying specific behaviour change tactics linked to MoAs and contextual conditions from the outset, our approach makes intervention components explicit in realist evaluation or synthesis work, which could inform later intervention design efforts. Aligning language used across both evaluation and intervention design could also potentially reduce duplication of effort and streamline the research pipeline.
Comparison with RAMESES guidelines and publication standards
This framework aligns with existing realist evaluation and synthesis guidelines, particularly the RAMESES standards, while offering enhancements. The RAMESES guidelines remain the benchmark for conducting and reporting realist work, and we encourage their continued use (Wong et al., 2013, 2016). However, they intentionally leave open how mechanisms operate at the level of individual behaviour. Our integrated approach addresses this gap by incorporating evidence-based taxonomies of BCTs and MoAs, enabling systematic mapping of interventions to measurable psychological processes.
While the RAMESES reporting and quality standards remain applicable here (RAMESES Project, 2014; RAMESES II Project, 2017a), our approach strengthens PT development by enhancing clarity, testability and transferability. For example, the RAMESES quality standards highlight that an excellent-quality PT should comprise ‘one or more context-mechanism-outcome configurations, describing how and why different mechanisms are triggered (or not) in different contexts to generate different outcomes’ (RAMESES II Project, 2017a: 4). We believe that our approach and the CCCs we developed retain this nuance, while making enhancements in language describing mechanisms with BCT and MoAs, to ensure theories are both context-sensitive and empirically grounded. A CCC is theoretically robust because it is logically coherent, makes reference to behavioural scientific theory for causal links and includes only well-defined factors (Mayne, 2017).
This also brings CCCs in line with John Mayne’s Contribution Analysis, a theory-based approach to evaluation designed for complex settings, which recognises that an observed result (outcome or impact) is typically brought about by a combination of factors working together, of which an intervention is only one part. Identifying the ‘causal package’ of factors, including the intervention, becomes the primary focus of programme evaluation (Mayne, 2015, 2023). Importantly, this approach also retains the core strengths of realist methodology – its iterative nature, contextual depth and methodological flexibility. Future work will assess its practical value by testing the framework in various realist projects.
Limitations
Realist methods aim to develop middle-range theories that are empirically grounded, yet transferable across contexts. However, some complex interventions, especially those at organisational or system levels, may not align easily with the lower level of abstraction needed to apply behavioural settings or BCTs. To address this issue, we offer flexibility for users to adopt either distal context or behavioural setting levels (or both), though this should be made explicit in the PT/CCC. In some cases, BCTs may not be the most appropriate way to conceptualise mechanisms as resources, such as in projects where organisational or team behaviour is being analysed, wherein ‘social mechanisms of change’ could be more appropriate. For example, we have delivered prior realist projects to understand how and why inter-organisational collaborations in healthcare may work, and this required an organisational level of analysis where BCTs would have had limited utility (Aunger et al., 2020, 2022). In these cases, the tactic and/or MoA may need to be formulated in a bespoke manner, as is currently done in realist practice.
Using BCT/MoA Ontologies is a trade-off made in the interest of enabling a common scientific concept-based language for PTs, at the relative cost of flexibility in PT formulation. It is possible that, in the future, ontologies enabling a common language for social mechanisms may be developed and these could be also incorporated within the CCC in place of current MoAs. However, it may also be the case that as the behavioural ontologies expand there may be more group-level BCTs and MoAs that are added; for example, the latest MoA Ontology contains ‘group identity’ (Schenk et al., 2024). Another important development could be broadening the scope of evaluation to group-level analysis, based on a CCC involving organisations as the fundamental units. For example, institutions can be considered to have quasi-motivation in terms of a ‘mission’ to achieve some goal – an approach taken by some schools of organisational behaviour and learning (e.g. Aldrich, 1999; Chadwick and Raver, 2015). Organisation-level theories of change can then be based in causal mechanisms derived from organisational or industrial psychology, sociology or political science in place of the BCTs and MoAs we have emphasised here. Such an approach would still be behavioural in the sense of treating an identifiable, emergent unit acting within the context of some environmental constraints. However, currently, formulating CCCs with MoAs occurring in group-level actors (e.g. across staff groups) can be the best means to depict group-level change.
In realist syntheses, identifying behavioural settings and intervention components from published reports can be difficult, whereas this may be more feasible in realist evaluations. Over-reliance on behavioural taxonomies may also risk narrowing focus to individual-level mechanisms, potentially overlooking broader organisational or cultural dynamics. Mapping realist mechanisms to behavioural science frameworks requires additional expertise and may raise the barrier to entry. Without proper training, there is a risk of misclassification, and aligning CMOs with BCTs or MoAs is not always straightforward. For example, inter-rater reliability for coding study reports against the MoA Ontology has been shown to be α = 0.68 for those familiarised with the ontology, versus α = 0.47 for those unfamiliarised (Schenk et al., 2024).
It is also important to note that while realist methods support iterative theory refinement, behavioural frameworks tend to be more fixed (although they are slowly evolving). This can be an issue if mechanisms are identified that lack direct equivalents in existing taxonomies. In such cases, it may be required to deviate from the common language of ontologies, and additional conceptual work or collaboration with framework developers may be needed to ensure capture of such mechanisms.
Finally, we would like to highlight that it is becoming increasingly common to use realist approaches to understand ‘natural phenomena’ without evaluating a formal programme or intervention. An example includes our prior work around understanding the drivers of unprofessional behaviour by healthcare staff (Aunger et al., 2023). In these cases, there may be no tactic or BCT to include in the CCC, and instead, unintentional changes in the distal context or behavioural setting may cause changes in people’s reasoning, stimulating behaviour change. Then, the PT may contain only context and/or behavioural setting, MoA and outcome.
Conclusion
Realist theory argues that to learn from the experience of engaging in a programme, one must understand how contexts impact mechanisms to drive particular outcomes. Only in this way can it be determined how, for whom and under what circumstances that intervention worked. However, in practice, realist PTs have exhibited issues such as conflation of context with intervention components, mechanisms that lack scientific basis, and poor utility in informing future intervention design. We argue that behavioural science can provide a common language to address many of these issues within realist PTs, while still capturing realist methods’ strengths around consideration of context in driving outcomes. This is achieved by integrating behavioural concepts such as behavioural settings, BCTs and MoAs over a five-step behavioural PT formulation process. This makes full use of the theories available in behavioural science within a process of realist theory development, making evaluations and syntheses more robust, while providing a common language for describing mechanisms. Consequently, PTs can be more sophisticated about linkages, more evidence-based, and more transferable. Subsequently, theories can also be applied more consistently from prior realist syntheses and evaluations, into intervention design all the way through post-facto evaluation, leading to more effective interventions. Future research will further test this framework in practice in realist syntheses and evaluations.
Footnotes
Acknowledgements
Thanks to Bianca Ungureanu who helped deliver the research upon which much of this thinking was based, and to Deborah Colson and Sarah-Jane Fenton who provided feedback on earlier versions of this manuscript.
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Justin Aunger was funded by the National Institute for Health and Care Research Midlands Patient Safety Research Collaboration while writing this manuscript. The views expressed are those of the author(s) and not necessarily those of the NIHR or the Department of Health and Social Care.
Ethical considerations
Not applicable.
Consent to participate
Not applicable.
Consent for publication
Not applicable.
Data availability
Not applicable.