Developing an evidence-based online method of linking behaviour change techniques and theoretical mechanisms of action: a multiple methods study

Article history

This issue of the Health Services and Delivery Research journal series contains a project commissioned/managed by the Methodology research programme (MRP). The Medical Research Council (MRC) is working with NIHR to deliver the single joint health strategy and the MRP was launched in 2008 as part of the delivery model. MRC is lead funding partner for MRP and part of this programme is the joint MRC–NIHR funding panel ‘The Methodology Research Programme Panel’.

To strengthen the evidence base for health research, the MRP oversees and implements the evolving strategy for high-quality methodological research. In addition to the MRC and NIHR funding partners, the MRP takes into account the needs of other stakeholders including the devolved administrations, industry R&D, and regulatory/advisory agencies and other public bodies. The MRP funds investigator-led and needs-led research proposals from across the UK. In addition to the standard MRC and RCUK terms and conditions, projects commissioned/managed by the MRP are expected to provide a detailed report on the research findings and may publish the findings in the HS&DR journal, if supported by NIHR funds.

The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HS&DR editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Copyright statement

© Queen’s Printer and Controller of HMSO 2021. This work was produced by Michie et al. under the terms of a commissioning contract issued by the Secretary of State for Health and Social Care. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2021 Queen’s Printer and Controller of HMSO

Improving health by changing behaviour

Addressing many of the huge challenges facing the world today will require changes in behaviour at all levels, from governments to individual citizens. Whether it be climate change and environmental degradation, war and other conflicts, epidemic and pandemic diseases, or poverty and inequalities, the solution often lies in changing deeply ingrained patterns of behaviour. Many of the world’s leading causes of mortality result from diseases, such as cardiovascular disease, lung cancer, stroke and human immunodeficiency virus (HIV), where behaviour contributes significantly to exposure to risk – smoking, poor diet, excessive alcohol consumption, physical inactivity and unprotected sex being cases in point. 1,2 In Western societies, such behaviour-related diseases contribute to approximately half of premature deaths. 3–6 Population health can also suffer significantly from a lack of certain behaviours (e.g. people not accessing health care when needed, not engaging with screening or vaccination programmes and not adhering to medication or other treatments and advice). The behaviour of health-care providers is also critical to population health, which can be significantly impaired if health professionals do not implement evidence-based practice, such as ensuring that the prescription and use of antibiotics is necessary and appropriate,6–9 or fail to follow infection prevention and control guidelines. 10

As well as reducing premature mortality and disability, behaviour change interventions have the potential to reduce health-care expenditure. 11,12 The latest analysis of the global burden of disease concluded that investing in preventing and controlling these diseases offers a high return for countries at all income levels and contributes to economic growth. 13 Interventions that target health-related patterns of behaviours have the potential to transform the health of populations, often at low cost,12–14 and there is good evidence that interventions have a positive impact on both health and equity across a number of areas. 15,16 Despite this, the effects of these interventions tend to be small, there is wide variation in effectiveness across contexts and the effects are often not maintained long term. 17,18 The effects are also usually not at the scale required to bring about a population-level impact (see Cochrane database for examples19–21).

Changing health-related patterns of behaviour across diverse populations and settings is not easy; it requires complex interventions (i.e. interventions that contain several interacting componets22) that target multiple levels within systems (e.g. communities or organisations) and can sustain their impact over time. 16 This in turn requires sophisticated methods for developing and evaluating interventions, including how to draw on theory and how to report interventions to maximise replicability and potential for faithful implementation. At present, authors do not reliably provide explicit reporting on the theoretical propositions underpinning interventions, their development and their evaluation. 15,22,23 Where theory-based interventions are reported, there is often little detail about how the theory was applied in the development or evaluation phase. 24–26 This is partly due to the paucity of methods to support this process. Methods are needed to support intervention reporting to move beyond describing them simply as being theory informed27 towards reporting how and why theoretical principles are tested. 28–31 One of the areas in which progress could be made is in the methods for intervention development and for accumulating knowledge across intervention evaluations. Progress has been made in improving, harmonising and standardising the reporting of interventions aimed at changing behaviour32–36 and their theoretical underpinnings. 37–40 This includes specifying interventions in enough detail to allow replication and to do this using consistent and shared terminology. 15,17,41,42

Behaviour change techniques: essential components of behaviour change interventions

Behavioural interventions are often delivered as part of complex systems that include a number of behaviour change techniques (BCTs). BCTs are conceptualised as fine-grained intervention components that on their own have the potential, in the right circumstances, to bring about behaviour change. 17,32,43,44 The Behaviour Change Technique Taxonomy version 1 (BCTTv1) is a formal, standardised system used to characterise behaviour change interventions, developed with input from 400 experts across 12 countries. 17,42 It consists of 93 BCTs organised into 16 groupings and provides labels and definitions of intervention techniques. This extensive shared vocabulary enables researchers and others to specify and describe the components of behavioural interventions, allowing better reporting and facilitating accurate replication and implementation of interventions. The BCTTv1 incorporates a number of cross-behaviour BCT taxonomies33,34 and some behaviour-specific taxonomies for physical activity,36 alcohol use,35 smoking45 and condom use. 39 The BCTTv1 has been used to specify the potentially ‘active ingredients’ in many types of intervention, for example those aiming to change physical activity and dietary behaviours,40,46 oral hygiene behaviours,47 hazardous and harmful drinking,48 sexual health behaviours,49–51 blood pressure control/management behaviours,52 antibiotic prescribing53,54 and type 2 diabetes preventative behaviours. 55

As well as being widely used for intervention reporting, the BCTTv1 is used to support intervention design and development and to synthesise information across intervention evaluations. Systematic reviewers have used it to identify BCTs in published reports of intervention evaluations and to generate evidence of the effectiveness of not only the intervention as a whole, but also the component techniques, either individually or as theoretical combinations that work synergistically. 25,26,56–59 This approach is recommended by the National Institute for Health and Care Excellence (NICE) in its guidance that research should investigate which BCTs are effective in promoting the initiation and maintenance of behaviour change. 15 Studies of published interventions have found that interventions using BCT combinations were more effective than those that used only one BCT. 60–62 For example, interventions that combine self-monitoring with BCTs that support self-regulation of behaviour, such as goal-setting or action-planning, have been found to be associated with improved effectiveness. 60 More than 350 articles reporting interventions coded by BCTTv1 are available for researchers and others to use in a searchable database, which is updated as researchers add to it (see www.bct-taxonomy.com/interventions; accessed 1 March 2020).

Specifying intervention content by BCTs has transformed methods for reporting the content of behaviour change interventions, which facilitates greater precision and consistency in research. 42,63 However, identifying which specific BCTs or BCT combinations are likely to be effective for a particular behaviour in its context is challenging. A review of 135 studies examined methods used to do this and found that, although a range of methods existed, they all have limitations. 64 The methods were (1) experimental manipulation of BCTs (where it is important to specify BCTs in the interventions, both in the experimental and in the control groups22,37,65–67), (2) observational studies comparing outcomes in the presence or absence of BCTs, (3) meta-analyses of BCT comparisons, (4) meta-regressions evaluating effect sizes with and without specific BCTs, (5) reviews of BCTs found in effective interventions and (6) meta-classification and regression trees (CART).

From the review findings, Michie et al. 68,69 concluded that there are limitations to all of these methods for drawing conclusions about BCT efficacy and that research in this area would be strengthened by triangulation of methods. Similarly, policy-makers and practitioners, when making decisions about what combination of BCTs to use in an intervention, should draw on findings across different methods. Another approach that would provide a fuller understanding of the impact that an intervention has on behaviour and health, and strengthen work in this area, is to link BCTs to other intervention features in an ontology of behaviour change interventions. 68,69 Research is currently under way to develop a Behaviour Change Intervention Ontology to support this approach (see www.humanbehaviourchange.org; accessed 1 March 2020).

An important aim of this work is to increase our knowledge of the mechanisms of action (MoAs) through which the BCTs have their effect. The need for this has been recognised beyond behavioural science, for example, by the Cochrane Collaboration’s Effective Practice and Organisation of Care (EPOC) Group70 and by NICE. 71,72 By linking BCTs to their hypothesised MoAs as described in behavioural theory, researchers and intervention designers may target MoAs more deliberatively, leading to interventions that are more likely to be effective. In addition, knowledge of these links makes it easier for investigators to design studies that can evaluate the processes underlying effective interventions and thus test and advance theory. Furthermore, it enables a theoretical understanding of interventions that have not been, or at least have not been reported as being, based on theory.

Behaviour change theory: specifying the mechanisms of action of behaviour change

Theories of behaviour change summarise accumulated knowledge about how behaviour change occurs, as well as explaining variation across behaviours, settings and populations. They have the potential to be useful tools in designing interventions to change behaviour, if applied systematically and appropriately. 28,29,37,73,74 This is reflected, for example, in the UK Medical Research Council’s framework for designing and evaluating complex interventions22 and the Intervention Mapping framework75 for planning health promotion programmes.

Theories of behaviour change attempt to explain and predict when, why and how behaviours do or do not occur and change. This involves proposing MoAs, defined as ‘the processes through which a behaviour change technique affects behaviour’, and moderators of change along various causal pathways. MoAs are theoretical constructs from theories of behaviour and behaviour change, such as ‘self-efficacy’ or ‘knowledge’, that mediate the effect of a BCT on a behaviour. By contrast, moderators are variables such as the population or the setting that may modify the effect of a BCT on a behaviour. As part of a multidisciplinary consensus exercise, theory was defined as:

A set of concepts and/or statements with specification of how phenomena relate to each other [providing] an organising description of a system that accounts for what is known, and explains and predicts phenomena.

Davis et al. 76 © 2014 The Author(s). Published by Taylor & Francis. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons. org/Licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited

There are numerous formal theories, which are more or less generalisable across behaviours, settings and populations. They vary in complexity and range of application, and many overlap. A multidisciplinary literature review led by psychologists, sociologists, anthropologists and economists identified 83 theories of behaviour and behaviour change, containing > 1700 theoretical constructs. 30,76 Because the review had strict inclusion criteria, there are likely to be many other relevant theories not included. Despite the abundance of theories, there is a paucity of guidance for researchers and intervention designers as to how they should decide about which theory or theories to draw on and how they should apply them to intervention design or evaluation. The result is that a relatively small number of theories dominate the field, and often they are poorly applied or their application is poorly reported. 26,77

Efforts to help in this process include the theoretical domains framework (TDF), which is an integrative framework of theoretical constructs relevant to understanding and changing behaviour. The framework was developed to make theories more usable and accessible to an interdisciplinary audience. 38,78,79 The TDF specifies 14 theoretical domains, each of which includes several theoretical constructs that are similar in definition but derive from different theories. The TDF has been used in intervention development and design,31,80,81 as well as in systematic reviews. 82–84 Although there are integrative frameworks, such as the TDF, and general intervention development frameworks, such as intervention mapping,22,75,85,86 a consensus is needed about how the individual MoAs specified in these theories can be linked with particular BCTs. 30,37 This would allow interventions to be developed in a way that was more precisely informed by theory than is generally current practice and would enable the theoretical understanding of interventions that were found to be effective but were not explicitly developed based on theory. However, we do not yet have an agreed method for systematically linking BCTs to individual hypothesised MoAs.

Linking behaviour change techniques with theoretical mechanisms of action

Although the BCTTv1 provides a shared language with which to describe intervention content, it does not directly specify which MoAs these BCTs target in the intervention. For interventions to have a good chance of being effective, their active components (i.e. BCTs) should target relevant MoAs.

There are a large number of BCTs and theories and models of behaviour and behaviour change. However, research investigating links between BCTs and MoAs has been sparse, despite a range of frameworks for developing behaviour change interventions [e.g. intervention mapping,87 precede–proceed model,88 the behaviour change wheel (BCW)89]. The importance of understanding the processes through which behaviour change may occur (a theory of change) has been recognised internationally, for example by the Cochrane Collaboration’s EPOC Group,70 NICE’s Public Health guidelines in the UK71,72 and the Science of Behavior Change programme of research in the USA. The Science of Behavior Change programme of research is building evidence about a discrete set of interventions by experimentally testing methods for changing specified MoAs (see https://commonfund.nih.gov/behaviorchange/index; accessed 1 June 2019). There is a need for many programmes of research to accumulate evidence about the processes by which the full range of BCTs have their effects; this will maximise our potential to develop effective interventions.

Preliminary work identifying links between BCTs and MoAs has been conducted in both primary research and evidence syntheses. 15,34,68,75,88,90 For example, a set of 35 BCTs has been mapped to theoretically derived domains in the TDF. 34 This approach has been used in developing behaviour change interventions31,34 and to identify theoretical mediators of change within process evaluations. 91–93 Associations between BCTs and theory have also been investigated in evidence syntheses such as systematic literature reviews and meta-analyses,26,94 and in primary research in relation to changing particular mechanisms of action, such as self-efficacy95–98 and behavioural intention. 99 Guidance about selecting BCTs to target processes of change have been developed in the form of intervention development frameworks (e.g. the BCW,85 Intervention Mapping75 and the TDF78,79). However, these are at a general level and do not include guidance or evidence about links between specific BCTs and MoAs.

This work shows the interest in and need for developing links between BCTs and their MoAs. However, there is no transparent, agreed method for identifying these links. This is needed to provide the tools for those (1) developing ‘theory-based’ interventions and (2) attempting to make theoretical sense of interventions that specify the BCTs used in interventions but without referring to theory and/or that specify MoAs but without identifying specific BCTs in interventions. Achieving this would enable evidence accumulation to continue to advance more systematically and efficiently.

In this report, we present a programme of research using a range of methods to identify and operationalise links between behaviour change techniques and (1) specific mechanisms of action and (2) behavioural theories. 100 These are links as reported by behaviour change researchers in developing and/or evaluating behaviour change interventions, and as judged by experts in a consensus exercise. These links are therefore hypothesised rather than necessarily tested and shown to be robustly supported by evidence. It thus provides a valuable resource for researchers as to which links are most and least likely to be effective in changing behaviour through changing a particular MoA. The links are presented in the form of matrices and interactive heatmaps that allow data from three studies to be easily extracted for each link and additional data or comments to be added by users (https://theoryandtechniquetool.humanbehaviourchange.org/tool). The data come from two complementary sources. In study 1, the data are derived from a synthesis of published literature, which encapsulates thinking in past peer-reviewed work. Study 2 is of expert consensus, which encapsulates current thinking. Study 3 triangulates the data from the previous two studies, providing a reconciliation of findings about the hypothesised links between behaviour change techniques and their MoAs. Study 4 examines links between groups of BCTs and theories as a whole. The sequence of studies in this research programme is shown in the flow diagram in Figure 1. The matrices of links resulting from these studies will provide a methodological online resource available to behaviour change scientists and intervention designers, providing a more efficient and systematic way to identify and evaluate the theoretical processes hypothesised to underlie BCTs.

FIGURE 1.

Flow diagram outlining the sequence of studies.

Aims

• To identify and integrate hypothesised links between (1) BCTs and MoAs and (2) BCTs and behavioural theories.

• To generate a freely available, searchable online resource to support theory-based intervention development and evaluation.

Objectives

• Intervention development and evaluation:

  • To triangulate literature analysis and expert consensus to identify links between BCTs and both MoAs and behaviour change theories.

  • To graphically represent the strength of BCT and MoA links through searchable matrices of the findings from literature synthesis, expert consensus and triangulation.

• Efficient evidence accumulation:

  • To create an accessible and freely available online resource that allows researchers to share data, publications and conference reports relevant to individual links.

  • To seek engagement and collaborations with international scientific and intervention development communities.

The objectives of each study are listed below.

Abstract

Introduction

Cumulative progress in behavioural science could be improved by advancing our understanding of the processes through which behaviour change interventions have their effects. As outlined in Chapter 1, behaviour change interventions consist of BCTs that are the smallest, irreducible components, or the potentially ‘active ingredients’ that can bring about change in behaviour. The theoretical constructs that represent the processes by which these BCTs change behaviour are termed MoAs. Identifying links between individual BCTs and MoAs could aid intervention design (e.g. by enabling us to identify relevant intervention strategies that influence a particular MoA) and evaluation (e.g. by allowing us to better understand the processes of change in successful interventions).

A number of sources of evidence can inform our understanding of these links. One valuable source of evidence is the published intervention literature. By examining this literature, we can examine links described or hypothesised by intervention authors and/or links that have been empirically tested (e.g. through experimental studies and meta-analyses). Although the available evidence for the latter (i.e. that resulting from empirical tests of individual links) appears to be limited,26 extracting and examining researchers’ descriptions of links may provide a useful data set. For example, where published interventions describe the rationale behind BCT selection, including the MoA(s) a specific BCT is believed to target (e.g. based on a theoretical model, previous research, clinical guidelines), this can shed light on researchers’ thinking and inform our understanding of BCT–MoA links.

Any one intervention may include a number of BCT–MoA links, as behavioural interventions tend to be delivered as part of a complex system, and can involve a number of co-occurring BCTs. BCTs can enable behaviour change by augmenting factors that facilitate change, or by mitigating factors that inhibit change. For example, in a smoking cessation intervention, researchers may hypothesise that providing an individual with emotional social support will change behaviour by increasing motivation. Similarly, persuading someone about their capabilities may be hypothesised to change behaviour by decreasing negative emotions such as shame or worry.

In the above example, motivation and negative emotions can be conceptualised as MoAs. In this context, MoAs are constructs from theories of behaviour and behaviour change that can be viewed as ‘mediating’ the effects of interventions. They may be constructs that relate to the individual (i.e. psychological processes, e.g. self or identity) or to the social or physical environment (e.g. social influences from peers).

The rationale underlying the selection of BCTs is not clearly described in every intervention article and there is also not always a clear, linear pathway for how BCT X changes behaviour by changing MoA Y. However, given the increasingly available array of tools for intervention development and reporting,17,18,41 many of which offer some form of guidance regarding which BCTs can target which MoAs, it is reasonable to assume that there is a corpus of literature from which BCT–MoA links can be identified.

This study aimed to identify the hypothesised links between BCTs and MoAs to inform (1) how researchers can target MoAs of interest (i.e. informing an understanding about which BCTs to select) and (2) explanations of intervention effects (i.e. informing an understanding of the MoAs through which BCTs are having their effects). We also aimed to understand whether or not any BCT–MoA links were described with a relatively high level of frequency in the included corpus of intervention articles. Therefore, based on the published intervention literature, this study sought to determine:

• the number of times BCT–MoA links are described by authors within studies

• which specific BCTs are most frequently described as targeting specific MoAs

• which specific MoAs are most frequently described as being influenced by specific BCTs

• whether or not specific BCT–MoA links are seen with greater frequency than anticipated based on the average number of times the BCT–MoA link is described.

Method

Data analysis

Results

Characteristics of intervention articles

A total of 974 intervention articles were identified through our search methods. Of those, 697 (72%) were excluded after screening the full text based on our coding guidelines (e.g. the paper was not reporting on an intervention, the intervention description was not detailed/clear enough to code BCTs, or there was an underpinning theory described but individual links between specific BCTs and MoAs were not clear). The most common reasons for exclusion were that papers did not provide a ‘codable’ description of a link between at least one BCT and a MoA and/or they did not clearly identify MoAs.

In total, 277 intervention articles described at least one BCT–MoA link, covering > 10 target behaviours. These included physical activity (40%), dietary behaviours (18%), alcohol intake (10%) and smoking (cessation) (6%). The years of publication ranged from 1982 to 2016, with almost half (49%) published in or after 2010. Over three-quarters (78%) of articles reported on the outcome(s) of a behavioural intervention, rather than describing their design or development. No theoretical basis was specified for the intervention in 14% of the articles. In addition, 13% mentioned theory but did not specify how theory was applied to developing or evaluating interventions. The analyses and discussions described in this chapter are based on the 277 included articles. A full summary of study characteristics are in the study information and materials on Open Science Framework (https://osf.io/7qjvn/; (accessed 1 January 2020). Figure 2 is a flow diagram of the search strategy and the selection process of the articles.

FIGURE 2.

Flow diagram of search strategy.

Do any specific behaviour change technique—mechanism of action links occur more frequently than might be expected given the average frequency of behaviour change technique—mechanism of action links? (Research question 4)

There were 87 links that occurred with a relatively high level of frequency, based on the p < 0.05 criterion. This included 51 out of 93 (55%) BCTs and 24 out of 26 (92%) MoAs. Of the MoAs that were linked to a BCT at least once (i.e. all except ‘norms’), only ‘optimism’ was not linked to any BCT at the p < 0.05 threshold. This MoA was derived from the TDF. 38 Several BCTs were coded frequently but did not meet the p < 0.05 threshold for being linked to a MoA. For example, the BCT ‘1.5 review behaviour goals’ was coded 36 times and ‘3.3 social support (emotional)’ was coded 14 times. However, the relative frequency with which these were linked to a MoA did not meet the threshold.

A heat map that displays a visual representation of link frequency, where darker colours represent p-values closer to zero, can be seen in Figure 3 and are also available online as part of an interactive online tool (https://theoryandtechniquetool.humanbehaviourchange.org/tool; see Discussion for more details).

FIGURE 3.

Heat map representing the relative frequency of BCT–MoA links.

Each cell contains a numerical value (i.e. p-value) and is coloured to reflect the relative ‘heat’ of that value (in this case, the relative frequency of a particular link). Rows and columns (i.e. BCTs and MoAs, respectively) are clustered by similarity, such that BCTs linked to similar MoAs are clustered together and MoAs linked to similar numbers of BCTs are clustered together.

The 51 BCTs and the MoAs that were most frequently linked are shown in Table 1. These summarise the descriptions provided by authors of this set of interventions regarding how these 51 BCTs change behaviour. For some BCTs, there is one MoA for which there appears to be agreement. For example, the BCT ‘8.7 graded tasks’ that was frequently coded across interventions, was linked to only one MoA (‘beliefs about capabilities’) with a relatively high level of frequency (p = < 0.001). For other BCTs, there are links to multiple MoAs. In some of these cases, there is one seemingly ‘dominant’ MoA. For example, although the BCT ‘1.2 problem-solving’ was frequently linked to three MoAs, the link to ‘beliefs about capabilities’ (p = 0.008; occurring 65 times) occurred substantially more frequently than the next highest two: ‘environmental context and resources’ (p = 0.026; occurring nine times) and ‘skills’ (p = 0.038; occurring 18 times).

TABLE 1 - BCT–MoA links with relatively high frequency in 277 intervention articles organised by BCT and presented in order of BCT frequency from this study and p-values

Note

Numbers in brackets for each BCT are as per BCTTv1.

BCT	MoA	Frequency	p-value
Instruction on how to perform the behaviour (4.1)	Knowledge	17	0.013
	Skills	20	0.024
Goal-setting (behaviour) (1.1)	Behavioural regulation	15	0.003
Problem-solving (1.2)	Beliefs about capabilities	65	0.008
	Environmental context and resources	9	0.026
	Skills	18	0.038
Social support (unspecified) (3.1)	Social influences	34	< 0.001
	Social/professional role and identity	5	0.037
Demonstration of the behaviour (6.1)	Beliefs about capabilities	60	0.003
	Skills	17	0.020
	Social learning/imitation	3	0.044
Action-planning (1.4)	Behavioural regulation	14	0.001
Feedback on behaviour (2.2)	Subjective norms	19	< 0.001
	Knowledge	13	0.013
Information about health consequences (5.1)	Knowledge	18	< 0.001
	Beliefs about consequences	26	< 0.001
	Attitude towards the behaviour	19	< 0.001
	Perceived susceptibility/vulnerability	10	< 0.001
	Intention	28	0.004
Behavioural practice/rehearsal (8.1)	Skills	24	< 0.001
	Beliefs about capabilities	47	0.013
Social comparison (6.2)	Subjective norms	31	< 0.001
	Social influences	9	0.043
Information about social and environmental consequences (5.3)	Beliefs about consequences	20	< 0.001
	Attitude towards the behaviour	16	< 0.001
	Knowledge	13	0.002
Self-monitoring of behaviour (2.3)	Behavioural regulation	18	< 0.001
Credible source (9.1)	General attitudes/beliefs	2	0.007
	Social/professional role and identity	4	0.023
Adding objects to the environment (12.5)	Environmental context/resources	8	< 0.001
Prompts/cues (7.1)	Memory, attention and decision processes	8	< 0.001
	Behavioural cueing	6	0.002
	Environmental context/resources	5	0.036
Graded tasks (8.7)	Beliefs about capabilities	28	< 0.001
Pros and cons (9.2)	Beliefs about consequences	12	< 0.001
	Attitude towards the behaviour	9	< 0.001
	Feedback processes	3	0.005
	Motivation	5	0.023
Framing/reframing (13.2)	Self-image	2	< 0.050
	Attitude towards the behaviour	7	0.014
Behaviour substitution (8.2)	Behavioural regulation	5	0.016
Social reward (10.4)	Reinforcement	3	0.020
Focus on past success (15.3)	Beliefs about capabilities	23	< 0.001
Restructuring the physical environment (12.1)	Environmental context/resources	9	< 0.001
	Behavioural cueing	3	0.020
Behavioural contract (1.8)	Goals	4	0.002
Information about others’ approval (6.3)	Subjective norms	13	< 0.001
	Intention	12	0.043
Verbal persuasion about capability (15.1)	Beliefs about capabilities	27	< 0.001
Feedback on outcomes of behaviour (2.7)	Subjective norms	5	0.020
	Feedback processes	2	0.027
Reduce negative emotions (11.2)	Beliefs about capabilities	12	0.039
Salience of consequences (5.2)	Attitude towards the behaviour	4	0.025
Commitment (1.9)	Values	1	0.039
Self-monitoring of outcomes of behaviour (2.4)	Behavioural regulation	5	0.024
Information about emotional consequences (5.6)	Beliefs about consequences	6	0.005
	Attitude towards the behaviour	5	0.006
	Emotion	2	0.024
Goal-setting (outcome) (1.3)	Goals	4	0.003
Social support (practical) (3.2)	Social influences	4	0.023
	Environmental context and resources	3	0.026
Discrepancy between current behaviour and goal (1.6)	Goals	3	0.001
	Behavioural regulation	3	0.019
Avoidance/reducing exposure to cues for the behaviour (12.3)	Needs	1	0.027
Identification of self as role model (13.1)	Self-image	2	0.011
Restructuring the social environment (12.2)	Environmental context/resources	3	0.004
	Social influences	6	< 0.001
Non-specific reward (10.3)	Reinforcement	2	0.005
Habit formation (8.3)	Behavioural cueing	3	0.001
	Behavioural regulation	3	0.024
Review outcome goals (1.7)	Goals	2	0.012
Mental rehearsal of successful performance (15.2)	Motivation	3	0.008
	Values	1	0.026
Material incentive (behaviour) (10.1)	Attitude towards the behaviour	1	0.048
Monitoring of behaviour by others without feedback (2.1)	Needs	1	0.019
	Social influences	2	0.036
Generalisation of target behaviour (8.6)	Skills	2	0.047
Comparative imagining of future outcomes (9.3)	Beliefs about consequences	3	0.017
Identity associated with changed behaviour (13.5)	Values	1	0.016
	Motivation	2	0.028
Anticipated regret (5.5)	Emotion	2	0.002
Habit reversal (8.4)	Behavioural regulation	4	0.006
	Behavioural cueing	2	0.023
	Memory, attention and decision processes	2	0.036
Associative learning (7.8)	Reinforcement	1	0.038
Self-incentive (10.7)	Motivation	1	0.036
Incompatible beliefs (13.3)	Attitude towards the behaviour	1	0.048

For those BCTs for which there are multiple MoAs linked, the data suggest that the authors may be hypothesising specific causal pathways (e.g. attitude → intention → behaviour). For instance, the BCT ‘5.1 information about health consequences’ was linked to the MoAs ‘beliefs about consequences’ (p < 0.001), ‘knowledge’ (p < 0.001), ‘perceived susceptibility/vulnerability’ (p < 0.001), ‘attitude towards the behaviour’ (p < 0.001) and ‘intention’ (p = 0.004).

Table 2 shows the 24 MoAs and the BCTs that were most frequently linked. It can be interpreted as a summary of the BCTs that the authors believe are suitable to target these 24 MoAs. In some cases, there is one clear BCT for a given MoA. For example, the MoA ‘perceived susceptibility/vulnerability’ was linked only to the BCT ‘5.1 information about health consequences’ (p < 0.001) and the MoA ‘social learning/imitation’ was linked only to the BCT ‘6.1 demonstration of the behaviour’ (p = 0.044).

TABLE 2 - The BCT–MoA links with relatively high frequency in 277 intervention articles: organised by MoA alphabetically and presented in order of p-values

Note

Numbers in brackets for each BCT are as per BCTTv1.

MoA	BCT	Frequency	p-value
Attitude towards the behaviour	Information about health consequences (5.1)	19	< 0.001
	Information about social and environmental consequences (5.3)	16	< 0.001
	Pros and cons (9.2)	9	< 0.001
	Information about emotional consequences (5.6)	5	0.006
	Framing/reframing (13.2)	7	0.014
	Salience of consequences (5.2)	4	0.025
	Material incentive (behaviour) (10.1)	1	0.048
	Incompatible beliefs (13.3)	1	0.048
Behavioural cueing	Habit formation (8.3)	3	0.001
	Prompts/cues (7.1)	6	0.002
	Restructuring the physical environment (12.1)	3	0.020
	Habit reversal (8.4)	2	0.023
Behavioural regulation	Self-monitoring of behaviour (2.3)	18	< 0.001
	Action-planning (1.4)	14	0.001
	Goal-setting (behaviour) (1.1)	15	0.003
	Habit reversal (8.4)	4	0.006
	Behaviour substitution (8.2)	5	0.016
	Discrepancy between current behaviour and goal (1.6)	3	0.019
	Self-monitoring of outcomes of behaviour (2.4)	5	0.024
	Habit formation (8.3)	3	0.024
Beliefs about capabilities	Graded tasks (8.7)	28	< 0.001
	Verbal persuasion about capability (15.1)	27	< 0.001
	Focus on past success (15.3)	23	< 0.001
	Demonstration of the behaviour (6.1)	60	0.003
	Problem-solving (1.2)	65	0.008
	Behavioural practice/rehearsal (8.1)	47	0.013
	Reduce negative emotions (11.2)	12	0.039
Beliefs about consequences	Information about health consequences (5.1)	26	< 0.001
	Information about social and environmental consequences (5.3)	20	< 0.001
	Pros and cons (9.2)	12	< 0.001
	Information about emotional consequences (5.6)	6	0.005
	Comparative imagining of future outcomes (9.3)	3	0.017
Environmental context and resources	Restructuring the physical environment (12.1)	9	< 0.001
	Adding objects to the environment (12.5)	8	< 0.001
	Restructuring the social environment (12.2)	3	0.004
	Problem-solving (1.2)	9	0.026
	Social support (practical) (3.2)	3	0.026
	Prompts and cues (7.1)	5	0.036
Emotion	Anticipated regret (5.5)	2	0.002
	Information about emotional consequences (5.6)	2	0.024
Feedback processes	Pros and cons (9.2)	3	0.005
	Feedback on outcomes of behaviour (2.7)	2	0.027
General attitudes/beliefs	Credible source (9.1)	2	0.007
Goals	Discrepancy between current behaviour and goal (1.6)	3	0.001
	Behavioural contract (1.8)	4	0.002
	Goal-setting (outcome) (1.3)	4	0.003
	Review outcome goals (1.7)	2	0.012
Intention	Information about health consequences (5.1)	28	0.004
	Information about others’ approval (6.3)	12	0.043
Knowledge	Information about health consequences (5.1)	18	< 0.001
	Information about social and environmental consequences (5.3)	13	0.002
	Instruction on how to perform the behaviour (4.1)	17	0.013
	Feedback on behaviour (2.2)	13	0.013
Memory, attention and decision processes	Prompts/cues (7.1)	8	< 0.001
	Habit reversal (8.4)	2	0.036
Motivation	Mental rehearsal of successful performance (15.2)	3	0.008
	Pros and cons (9.2)	5	0.023
	Identity associated with changed behaviour (13.5)	2	0.028
	Self-incentive (10.7)	1	0.036
Perceived susceptibility/vulnerability	Information about health consequences (5.1)	10	< 0.001
Needs	Monitoring of behaviour by others without feedback (2.1)	1	0.019
	Avoidance/reducing exposure to cues for the behaviour (12.3)	1	0.027
Reinforcement	Non-specific reward (10.3)	2	0.005
	Social reward (10.4)	3	0.020
	Associative learning (7.8)	1	0.038
Self-image	Framing/reframing (13.2)	2	< 0.050
	Identification of self as role model (13.1)	2	0.011
Skills	Behavioural practice/rehearsal (8.1)	24	< 0.001
	Demonstration of the behaviour (6.1)	17	0.020
	Instruction on how to perform the behaviour (4.1)	20	0.024
	Problem-solving (1.2)	18	0.038
	Generalisation of target behaviour (8.6)	2	0.047
Social influences	Social support (unspecified) (3.1)	34	< 0.001
	Restructuring the social environment (12.2)	6	< 0.001
	Social support (practical) (3.2)	4	0.023
	Monitoring of behaviour by others without feedback (2.1)	2	0.036
	Social comparison (6.2)	9	0.043
Subjective norms	Feedback on behaviour (2.2)	19	< 0.001
	Social comparison (6.2)	31	< 0.001
	Information about others’ approval (6.3)	13	< 0.001
	Feedback on outcomes of behaviour (2.7)	5	0.020
Social learning/imitation	Demonstration of the behaviour (6.1)	3	0.044
Social/professional role and identity	Credible source (9.1)	4	0.023
	Social support (unspecified) (3.1)	5	0.037
Values	Identity associated with changed behaviour (13.5)	1	0.016
	Mental rehearsal of successful performance (15.2)	1	0.026
	Commitment (1.9)	1	0.039

In other cases, clusters of theoretically linked BCTs were linked to one MoA. An example is the MoA ‘emotion’ being linked to the BCT ‘5.5 anticipated regret’ (p = 0.002) and ‘5.3 information about emotional consequences’ (p = 0.024).

Discussion

This study provides the first data set summarising links between BCTs and MoAs, as described by authors of published intervention articles. There were 2636 links identified in this study, including 70 BCTs and 25 MoAs. Eighty-seven links were identified with a relatively high level of frequency (i.e. meeting the criterion of p-value of < 0.05). The findings of this study provide a summary of links that are described frequently (e.g. providing information about the health consequences of a behaviour is frequently hypothesised to increase knowledge), as well as those links that appear to be absent. Advancing the science of behaviour change at a theoretical and methodological level, through this and similar initiatives (see www.scienceofbehaviourchange.org), helps to provide a grounding from which researchers and practitioners can build innovative new interventions by combining BCTs, knowing where important gaps are and providing a basis for new hypotheses.

The findings indicate that, in some cases, there is one clear BCT for a given MoA and one clear MoA for a given BCT. In other cases, there are BCTs linked to more than one MoA and MoAs linked to more than one BCT. Some of the links that can be seen in our data set map closely to the theoretical literature. For example, in Bandura’s theory of self-efficacy,107 a number of intervention strategies that can be used to increase self-efficacy are described, including mastery experience, vicarious experience and verbal persuasion. These intervention strategies are comparable to the BCTs ‘8.1 behavioural practice/rehearsal’, ‘6.1 demonstration of the behaviour’ and ‘15.1 verbal persuasion about capability’, respectively, all of which were linked to the MoA ‘beliefs about capabilities’, a construct that is conceptually identical to self-efficacy in this study.

Although we identified at least one link for 91% of the coded BCTs and 92% of the MoAs, in a number of cases no clear links were identified. Two of the MoAs in our set of 26 were not linked to BCTs enough to meet the criterion of a p-value of < 0.05. These were ‘optimism’ and ‘norms’, both of which are found to be frequently occurring in behavioural theories. 30 This may reflect a lack of clarity or agreement in the behavioural intervention literature regarding the types of BCTs that should be employed to target these MoAs.

Limitations

There are a number of limitations to the current work. We did not undertake a full systematic review; the intervention articles were purposively selected to maximise the likelihood of being able to identify BCT–MoA links. This was because the aim of the study was to identify the links that were described in intervention articles, rather than to set out to examine the clarity with which they tended to be described. Although we attempted to collate articles from the international research community (e.g. by contacting the Society of Behavioral Medicine in the USA and the European Health Psychology Society), there is a wide range of published intervention evaluations that our data set of articles may not represent.

The extracted BCT–MoA links were based on authors’ descriptions and hypotheses. Very few had been tested empirically in the articles. Therefore, we were unable to synthesise the results of empirical tests of these links. This indicates that there is a need for empirical research that systematically tests the links that have been frequently described. With this aim in mind, work is ongoing in the USA to identify, measure and manipulate MoAs using experimental methods (see www.scienceofbehaviourchange.org).

We cannot use the findings of this study to draw conclusions about the links that did not appear in the included intervention articles. The links that appeared to be absent may reflect a belief among authors that they do not exist, or that they may be links that do not tend to be considered by intervention designers (despite being potentially useful). These links may also include MoAs that authors find difficult to operationalise, or those that authors tend to use implicitly and are therefore less likely to explicitly describe in detail.

Next steps

This was the first of three related studies that aim to examine the links between BCTs and (1) MoAs and (2) behavioural theories. 100 In the next stage, links between BCTs and MoAs were identified through an expert consensus study.

Abstract

Introduction

Identifying specific links between BCTs and MoAs is important for both developing interventions and understanding the processes through which interventions affect behaviour change. One method for generating evidence on the links between BCTs and MoAs would be to conduct experimental studies of BCT–MoA links by manipulating BCTs and measuring MoAs and subsequently conducting meta-analyses of their findings. 108 However, the sheer number of potential BCTs and MoAs and, consequently, BCT–MoA links render these methods relatively infeasible.

Indirect approaches, such as the examination of hypothesised links between BCTs and MoAs in published intervention literature (see Chapter 2), provide a valuable alternative source of evidence. However, the published intervention literature is limited by various publication biases. These include the selection of research projects (at least in part) being driven by funding requirements, reporting of projects by researchers and the selection of which findings to present for publication. Examining the current thinking of international experts in behaviour change provides a complementary source of evidence, capable of generating useful information on a large set of BCT–MoA links. Using an expert consensus approach permits a synthesis of the current hypotheses of experts in the field, limits the impact that publication or funding-related biases have and is informed by existing theory and evidence.

Expert consensus methods can be used to facilitate the development of research questions, solutions to existing problems and priorities for action. 109 Through expert consensus methodologies, differing ideas can be discussed, reported and organised to establish areas of consensus and priorities for further investigation. Furthermore, these methods cultivate the experts’ ownership of the resulting research, thereby increasing the likelihood that future research and practice will be influenced by the outcomes of the study. 110

This chapter reports the second in a series of four studies to develop and test a methodology for generating links between BCTs and MoAs. This study aimed to build consensus around the mechanisms through which BCTs may alter behaviour. The primary research questions were:

1. Which BCTs do experts agree influence behaviour through specific MoAs?

2. Which BCTs do experts agree do not influence behaviour through specific MoAs?

3. Do experts agree that BCTs influence behaviour through one or multiple MoAs?

The secondary research questions were:

1. Which links between BCTs and MoAs do experts disagree on?

2. Do experts agree on at least one link with a MoA for all BCTs?

3. Do experts agree on at least one link with a BCT for all MoAs?

Methods

Conduct of the consensus exercise

Materials

Round 1

Experts were e-mailed the definitions of their assigned 13 or 14 BCTs and 26 MoAs, along with the guidelines for the task (see Appendix 5). For each question, BCT and MoA definitions were provided alongside a diagram depicting the causal behaviour change flow from each BCT to MoA link as shown in Figure 4.

FIGURE 4.

Diagram depicting the causal behaviour change flow from each BCT to MoA link.

As shown in Figure 4, experts were provided with this diagram during round 1 of the consensus exercise, for each question, which depicts that a BCT influences a MoA, which in turn influences behaviour change.

Data analysis

Data from round 3 were analysed to answer the three primary and three secondary research questions (see Background). To answer all research questions, the proportion of experts who rated the same response option (i.e. ‘definitely yes’, or ‘definitely no’) was calculated using Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA) to identify which links reached consensus. For a link to reach consensus, > 80% of experts needed to agree a BCT was either ‘definitely linked’ or ‘definitely not linked’ to a MoA. In addition, to answer secondary research question 1, the distribution of proportions for each of the response options were examined (i.e. the proportion of experts rating ‘definitely yes’ vs. ‘definitely no’).

To facilitate a visual analysis of the descriptive data, four heat maps were produced in R for the final round 3 ratings. A heat map is a visual depiction of a data matrix where the values in the cells are represented by colours shaded to indicate the strength or ‘heat’ of a particular value. In the heat map, BCTs represent rows that are linked to MoA columns. In each of the heat maps, the cell values presented are the proportion of experts who agree that a BCT and MoA are (1) ‘definitely’ linked (Figure 5), (2) ‘definitely not’ linked (Figure 6), (3) ‘possibly’ linked (see Appendix 5) and (4) ‘uncertain’ whether or not they are linked (see Appendix 5).

FIGURE 5.

A heat map of the percentage of experts rating a BCT as ‘definitely’ linked to a MoA. ATB, attitude towards the behaviour; BCap., beliefs about capabilities; BCon., beliefs about consequences; BR, behavioural regulation; GAB, general attitudes and beliefs; MADP, memory, attention, and decision processes; PSV, perceived susceptibility and vulnerability; SPRI, social/professional role and identity.

FIGURE 6.

A heat map of the percentage of experts rating a BCT as ‘definitely not’ linked to a MoA. ATB, attitude towards the behaviour; BCap., beliefs about capabilities; BCon., beliefs about consequences; BR, behavioural regulation; GAB, general attitudes and beliefs; MADP, memory, attention, and decision processes; PSV, perceived susceptibility and vulnerability; SPRI, social/professional role and identity.

To facilitate this visual analysis, the heat map clusters rows (BCTs) and columns (MoAs) by similarity. For example, in the ‘definitely’ linked case, BCTs linked to similar MoAs (such as BCT X and Y both being linked to MoAs 1, 5 and 7) are clustered closer together vertically. In addition, MoAs linked to similar numbers of BCTs are clustered together horizontally (e.g. MoA 1 and 2 are both linked to five BCTs). This clustering presents patterns in the data but does not imply any statistical inference.

To examine any possible effects of group membership on experts’ ratings, intraclass correlation coefficients (ICCs) were calculated. The ICCs were calculated using the ratings data from the 52 BCT–MoA links rated by all experts. The round 1 ICCs indicate whether or not group allocation had any influence on the ratings. Round 3 ICCs indicate the extent to which the within-group discussion (round 2) influenced the ratings. To understand the extent to which variance in the experts’ ratings can be attributed to group membership, the ICC value can be translated into a percentage. For example, if the ICC value is 0.03, this means that 3% of the variance in ratings can be attributed to group membership properties. Without standard criteria to assess ICC values for group ratings, the results describe, rather than evaluate, the influence of group membership on experts’ ratings. We predicted small ICCs for round 1 due to the block randomisation procedures and larger round 3 ICCs because of the likely influence of group-specific discussion about the links.

Results

Round 3

Nearly all experts (n = 100, 95%) participated in round 3, rating 1586 links between 61 BCTs and 26 MoAs. Consensus was reached for 90 ‘definite’ (research question 1) and 464 ‘definitely not’ (research question 2; see Appendix 5) BCT–MoA links. Of the 102 links discussed in round 2 with either high disagreement and/or uncertainty, agreement was reached for eight links.

Out of the 61 BCTs rated, 51 were ‘definitely’ linked to at least one MoA. Of the 26 MoAs, 21 were ‘definitely’ linked to at least one BCT (research questions 5 and 6). The specificity with which BCTs linked to MoAs is illustrated in Figure 7, the frequency with which BCTs were definitely linked to MoAs by > 80% of experts is depicted in the dark blue bars, and the frequency with which MoAs were linked to BCTs by > 80% of experts is depicted in the light blue bars.

FIGURE 7.

The frequency of BCT–MoA links as rated by experts.

Twenty-three BCTs were linked to only one MoA, and 20 BCTs were linked to only two MoAs (research question 3). MoAs were linked to between one and nine BCTs. The MoA ‘motivation’ was linked to nine different BCTs. The 10 BCTs with no ‘definite’ links to MoAs included:

• 1.4 action-planning

• 2.1 monitoring of behaviour by others without feedback

• 2.5 monitoring of outcomes of behaviour without feedback

• 8.2 behavioural substitution

• 8.6 generalisation of the target behaviour

• 9.1 credible source

• 10.6 non-specific incentive

• 11.1 pharmacological support

• 12.6 body changes

• 15.4 self-talk.

The five MoAs with no definite links to BCTs included (1) general attitudes and beliefs, (2) needs, (3) optimism, (4) social professional role and identity, and (5) values.

The visual representations of the expert consensus on BCT–MoA links are presented in two heat maps (see Figures 5 and 6). The BCTs that experts agreed were linked to similar MoAs are vertically closer together in the heat map, and MoAs that are linked to a similar frequency of BCTs are closer horizontally.

A larger number of links did not meet the consensus criterion (n = 1032, 65%). This was due to strong disagreement among the experts for 163 links. ‘Strong disagreement’ was recorded when at least one-third of experts responded ‘definitely yes’, one-third responded ‘definitely no’ and one-third responded either ‘possibly’ or ‘don’t know/uncertain’ (research question 4). For 340 links, between 50% and 80% of experts agreed on 255 and 85 links as ‘definitely not’ and ‘definitely’ links (respectively). The remaining experts provided ratings of ‘possibly’ and ‘don’t know/uncertain’. No meaningful trends emerged among the remaining 529 links.

Analysis of between-group differences in rating patterns (intraclass correlation coefficients)

To assess whether or not there were differences in expert ratings because of group membership, we examined the ICCs for the BCTs shared across groups [i.e. instruction about how to perform the behaviour and social support (unspecified)]. The ICCs for round 1 were small for all BCT–MoA links (range 0.00–0.10), suggesting that group allocation did not affect ratings (Table 4).

TABLE 4 - Intraclass correlation coefficients for the BCT–MoA links rated by all experts

Note

Δ = Change in the ICC from round 1 to round 3. ‘3.1 social support (unspecified)’, and ‘4.1 instruction on how to perform the behaviour’ are the BCTs rated by all experts for links with the 26 MoAs. Values in the table indicate the ICC for each link rated by all experts who had been randomly allocated to one of five groups.

MoA	Social support (unspecified)			Instruction on how to perform a behaviour
	Round 1	Round 3	Δ	Round 1	Round 3	Δ
Knowledge	0.05	–0.01	–0.06	–0.03	–0.04	–0.01
Skills	0.05	0.07	0.02	0.00	0.02	0.02
Behavioural regulation	0.07	0.12	0.05	0.09	0.11	0.02
Social influences	0.00	–0.01	–0.01	–0.01	0.01	0.02
Memory, attention, decision processes	0.01	0.02	0.01	0.04	0.02	–0.02
Social professional role/identity	0.02	0.10	0.08	0.01	0.02	0.01
Beliefs about capabilities	0.09	0.18	0.09	0.02	0.01	–0.01
Beliefs about consequences	0.03	0.05	0.02	–0.03	0.01	0.04
Optimism	–0.04	0.01	0.05	–0.02	0.01	0.03
Intention	0.00	0.10	0.1	–0.02	0.08	0.1
Goals	0.04	0.07	0.03	0.07	0.06	–0.01
Reinforcement	0.10	0.13	0.03	0.03	0.04	0.01
Emotion	0.03	0.08	0.05	0.00	0.03	0.03
Environment	0.01	0.01	0	0.03	0.00	–0.03
Norms	–0.01	–0.02	–0.01	0.00	0.01	0.01
Subjective norms	0.00	–0.03	–0.03	–0.02	0.03	0.05
Attitude towards the behaviour	0.01	0.24	0.23	–0.01	0.23	0.24
Motivation	0.03	0.07	0.04	0.00	0.00	0
Self-image	–0.03	0.00	0.03	0.04	0.07	0.03
Needs	–0.02	0.09	0.11	–0.03	0.05	0.08
Values	–0.02	0.17	0.19	0.00	0.06	0.06
Feedback processes	–0.01	0.02	0.03	0.00	0.02	0.02
General attitudes and beliefs	0.08	0.26	0.18	0.03	0.10	0.07
Social learning	0.04	0.21	0.17	0.07	0.05	–0.02
Cueing	0.07	0.06	–0.01	–0.02	0.05	0.07
Perceived susceptibility/vulnerability	0.03	0.03	0	0.05	0.05	0

As predicted, ICCs increased from round 1 to round 3 for 37 BCT–MoA links, reflecting an increase in within-group agreement after the discussion round. In round 3, 1 out of 52 (1.9%) ICCs was large [for the MoA ‘general attitudes and beliefs’ with the BCT ‘3.1 social support (unspecified)’], and 13 out of 52 (25%) were moderately-sized. Ten of the moderately sized ICCs were for the BCT ‘3.1 social support (unspecified)’, suggesting that round 2 discussions engendered more agreement on the meaning of this BCT in some groups than others.

Discussion

In the current expert consensus study, 90 BCT–MoA links between 51 (of 61 commonly used) BCTs and 20 (of 26 frequently used) MoAs were identified, which may provide promise for intervention development and evidence synthesis. Experts agreed that 464 links definitely did not exist and, as such, should not be targeted by interventions, whereas no consensus was reached for 1032 links; therefore, further research is required. No links were identified in either direction for 10 frequently used BCTs and six MoAs, which may be of particular interest for future research on behaviour change.

The pattern of results indicates that, for most of the BCTs evaluated in this task, the key MoAs for behaviour change are ‘motivation’ and ‘intention’. A large number of BCTs were also hypothesised to operate through the MoAs: (1) ‘beliefs about capabilities’, (2) ‘beliefs about consequences’ and (3) ‘behavioural regulation’. Nearly all experts agreed on which BCTs are linked to ‘reinforcement’, ‘cueing’ and ‘environmental context and resources’. The extent of agreement for these links could possibly be explained by the specificity with which the theoretical literature describes techniques for changing these MoAs. In addition, several of the groupings that occur in BCTTv1 (which is structured hierarchically and groups BCTs that are similar in function within 16 clusters)17 also appear within the heat map (in terms of vertical proximity). This could indicate shared hypotheses across different consensus studies about how BCTs change behaviour.

There was no consensus for some of the most frequently used BCTs (e.g. ‘1.4 action-planning’, ‘9.1 credible source’, ‘8.2 behavioural substitution’) on the MoAs they target. Experts actively participated in this consensus exercise, so it is possible that through additional rating rounds, a greater number of participants, and/or an additional consensus exercise with different experts, the number of agreed BCT–MoA links could increase. An alternative explanation is that in several different theories these BCTs are linked to different MoAs and possibly only work in combination with other BCTs. This alternative explanation supports feedback from the experts that it was often difficult to make BCT–MoA judgements in isolation. For example, several MoAs that occur frequently within theories of behaviour change (e.g. ‘needs’, ‘values’, ‘optimism’) were not linked to any BCTs. It is possible that these MoAs do not occur in isolation and rather need to be targeted by multiple BCTs, as noted by experts during the discussion round.

The BCT–MoA-behaviour change effect depends on both the path from BCT to MoA and the path from MoA to behaviour change. The extent to which experts agreed on a link could have been influenced by experts’ beliefs about either of these effects. For example, if the experts had confidence in the BCT’s ability to alter the MoA, but limited confidence in the link between the MoA and behaviour change, experts may have rated that the BCT was not linked to the MoA. Experts echoed this concern during the discussion round. The present study cannot ascertain whether experts’ ratings were influenced more by their hypotheses about the effect of the BCT on the MoA, or the effect of the MoA on behaviour change. The modified question wording in round 3, which was altered in response to comments from experts, may have limited this to a degree by asking experts to consider their judgements in round 3 in response to the hypothetical, ‘When this BCT works . . .’. Further research is needed to explore the rationale behind experts’ ratings. For example, 95% of experts agreed that ‘1.1 goal-setting (behaviour)’ changes behaviour by eliciting changes in intentions; despite comments during the discussion round on experts’ difficulty linking BCTs to intention. This finding may be a result of experts’ knowledge of previous research findings that changes in intentions are associated with small to medium changes in behaviour.

In this study, we chose a stringent criterion to evaluate expert consensus. In a systematic review of expert consensus studies, the percentage of experts in agreement was the most frequently used method to determine consensus. Studies using percentage of experts in agreement applied a range of criteria to establish consensus (50–97%), with the median threshold value at 75%. 113 The relatively small number of experts rating each link compared with the number of links rated by each expert was a major factor in the decision to use such a stringent criterion. Future research should generate empirical evidence for the links where more than two-thirds of the experts agreed (i.e. the more moderately shaded links in Figure 5).

At present, when selecting BCTs for use in an intervention, intervention developers tend to consult theory, published literature and their own hypotheses to determine what may be most effective for modifying the relevant MoAs and consequently changing the behaviour of interest. The current data provide evidence about the shared judgements of experts and an additional resource for both the development of interventions and the generation of data-driven hypotheses. Similarly, these data can also be used for evaluation of how an intervention has its effect. Furthermore, the data indicating which BCTs are not linked to specific MoAs, and which MoAs are not linked to specific BCTs, may also be useful for understanding what mechanisms do not provide promise. Therefore, these results generate hypotheses about effective links, which can foster empirical research to test the most promising BCT–MoA links and increase our understanding of how intervention components have their effects.

Limitations

The range of links identified in this study was restricted to the subset of BCTs and MoAs evaluated (i.e. 61 of 93 BCTs in the BCTTv1). We also chose a specific set of experts based on our inclusion/exclusion criteria. A different recruitment strategy or panel of experts may have led to different results. However, several measures were taken in this study to limit this possibility, including the large sample size and variety of recruitment procedures used. Some of the expert groups were able to agree on certain BCT–MoA links more than other groups; however, most of the links with better agreement in some groups were for the BCT ‘3.1 social support (unspecified)’. In previous studies this BCT has demonstrated lower than average reliability. 17,32 We changed the question prompt and the rating scale used from round 1 to round 3, which could have limited the consensus among experts. The difficulty experts had with the original question and rating scale may have limited the total number of links agreed on at the end of the consensus exercise. Lastly, we restricted the set of MoAs in the study based on the frequency with which they appear in behaviour change theories. Therefore, it is possible that there are other important MoAs that were not considered in this study.

Future directions

This study focused on theoretically based MoAs that have not necessarily been explicitly linked in theories or theoretical frameworks. A given theory may propose how MoAs interact to change behaviour. Thus, there is reason to explore whether the BCT–MoA links have additive (independent), synergistic, or antagonistic effects, as such effects have been explored in previous research. 60–62 Therefore, the next study (see Chapter 4) presents triangulated data on the two previous studies (see Chapters 2 and 3) to increase confidence in the links identified through both the literature synthesis and the present expert consensus study. Future analyses of the collected data could also include a qualitative analysis of the expert discussions conducted throughout the project.

Conclusions

This study provides a systematically derived summary of experts’ judgements about how BCTs change behaviour. The results can be used to inform intervention development, evaluation and synthesis of effectiveness evidence. The majority of BCT–MoA links for which there was uncertainty among the experts could be of particular interest for future studies. These results offer a first level of evidence-based hypotheses, which can be confirmed or refuted through further empirical research to delineate how BCTs have their effects on behaviour.

Abstract

Introduction

Interventions to change behaviour are necessary to address many current global challenges, including health, climate and environmental issues. 1,2 Interventions depend on having a sound theoretical and evidence base for optimal development. Considerable progress has been made in specifying BCTs, which are the ‘active ingredients’ of an intervention aimed at changing behaviour. The BCTTv1 specifies 93 BCTs with labels and definitions. 17,42,114 However, it is still unclear how BCTs should be selected for inclusion in an intervention. For example, how would a physiotherapist aiming to increase physical activity in a patient following stroke choose among BCTs as diverse as ‘demonstration of the behaviour’ and ‘giving rewards if and only if the activity is performed’? In addition, finding that an intervention successfully changed behaviour, one might wish to understand the process by which BCTs were enacted. However, this process is understudied and underspecified and lacks evidence.

A scientist or practitioner developing an intervention typically has an idea or theory of how their intervention might change behaviour, which guides their selection of techniques to incorporate in the intervention. For example, a physiotherapist may consider that the intervention may need to increase the patient’s motivation or skills. These ideas about intervention development can benefit from progress in behavioural science regarding influences on behaviour, which mediate the effect of behaviour change interventions (i.e. the MoAs of the interventions). Deciding which BCTs may affect change in these MoAs is a key part of intervention development, but one where there has been little evidence-based guidance. For example, if a physiotherapist believed that motivation was the MoA, it is unclear what evidence could guide their BCT choice.

To implement a successful intervention or interpret the findings from evidence synthesis, it is also important to learn how and why the intervention worked. Understanding the MoA through which the BCTs worked may provide insight into further relevant BCTs, or methods of enhancing the effectiveness of the included BCTs. Thus, gaining evidence about the links between BCTs and MoAs may facilitate both the selection of BCTs to target an MoA in designing an intervention, and the interpretation of the MoA of BCTs delivered in an effective intervention.

There is a clear scientific need to identify links between BCTs and MoAs. In the USA, the Science of Behavior Change programme of the National Institutes of Health is testing the effect of experimentally changing MoAs. 115–117 Furthermore, the experimental medicine approaches to behaviour change advocate testing mediation of intervention effects via MoAs to identify routes to behaviour change. 118 In the intervention mapping framework, linking the hypothesised determinants of behaviour to behaviour change methods is a key step in their systematic approach to intervention design. 119 Given the enormous number of possible combinations of MoAs and BCTs, it would be unrealistic and potentially a waste of resources to test the mediating effect of each possible link empirically. Therefore, guidance on those BCT–MoA combinations that are most likely to be effective (‘links’) and ineffective (‘non-links’) would be a useful and practical scientific advance.

The previous chapters provided preliminary evidence on BCT–MoA links, with Chapter 2101 involving a synthesis of literature reporting behaviour change interventions and Chapter 3102 involving an expert consensus study. The findings have similarities but are not identical. What we need now is a triangulation study to establish the degree of agreement between these two sources of evidence, to identify points of disagreement and to reconcile any discrepancies. Otherwise, intervention designers or those interpreting and evaluating interventions may be faced with conflicting, diverging or confusing evidence.

Triangulation of evidence using different methods to answer the same question is good scientific practice. Compared with simple replication of the same methods, one can have more confidence in the validity of the findings if biases in the design, techniques and analytic methods are not repeated. 120 The literature synthesis study101 (see Chapter 2) provides findings that reflect the previous thinking and methods of intervention designers addressing a variety of different behaviour change targets across disciplines and countries. However, findings are limited by reporting methods and publication constraints. Conversely, the experts involved in the consensus study102 (see Chapter 3) used their knowledge of current evidence and theory in reaching their decisions. However, for practical reasons they could consider only a restricted set of BCTs and MoAs. Therefore, triangulating findings derived from these two methods would strengthen confidence in the results.

Both previous studies produced a matrix of links between BCTs and MoAs. Although some concordance is expected between matrices, there are important differences between the two sets of data that limit the extent of possible concordance. First, experts in the consensus study could agree that a possible ‘link’ was improbable and therefore probably a ‘non-link’, whereas in the literature study it was only possible to find that a link had not occurred and there was no way of separating improbable links from those that simply had not been used in the 277 interventions examined. Second, the literature study could examine any and every BCT and MoA that authors had investigated. However, for practical reasons the consensus study included a set restricted to 61 BCTs and 26 MoAs. Thus, concordance could only be examined for the possible links included in both studies.

It is important first to examine concordance to identify points of agreement between Chapters 2 and 3, and then to examine whether or not the two methods achieve agreement that goes beyond chance coincidence. Otherwise any guidance to researchers, policy-makers and practitioners would be built on misleading evidence. Therefore, this chapter aims to:

1. investigate the degree of agreement between the findings of the two studies reported in Chapters 2 and 3 using statistical methods to evaluate concordance between them

2. reconcile remaining uncertainties and discrepancies between the literature synthesis and expert consensus studies (Chapters 2 and 3) using a new group of experts.

Based on these findings, we aimed to make the results more accessible to potential users by developing an online resource that provides researchers and intervention designers with information on the ‘links’ and ‘non-links’ between BCTs and MoAs.

Methods

Results

The flow chart in Figure 8 summarises the stages of the research and the data at each stage.

FIGURE 8.

Flow chart showing stages in the research and the links between BCTs and MoAs.

Results of multilevel modelling to quantify agreement between the results of the literature synthesis and expert consensus studies

The overall evidence of agreement is first presented visually, as scatterplots with regression lines, in Figure 9 for expert consensus on a ‘link’. The p-values from the literature study are plotted against the proportion of ‘definitely yes’ and ‘definitely no’ responses in the expert consensus study, with each line representing one MoA for all MoAs except ‘norms’, which was not found. In each case, lower p-values indicate higher frequency of links. The negative slope of the regression lines indicates that lower p-values were associated with a higher frequency of ‘yes’ responses by experts. Thus, the slopes indicate agreement between the two sources of evidence and this pattern is apparent for each MoA. Each line represents the prediction from the multilevel model. The thick line is the overall regression line. Other lines represent each MoA. The dots indicate the predicted values for actual data points. The occasions when no link was proposed in the literature synthesis study (coded as p = 1) have been omitted.

FIGURE 9.

Relationship between findings from the literature synthesis and expert consensus studies when links between BCTs and MoAs have been proposed (i.e. ‘yes’, BCT and MoA linked). Note that each dot represents one BCT (with BCTTv1 label). The line represents the prediction from the multilevel model omitting BCTs for which p = 1 in the literature study (but the expert consensus values for such BCTs are shown).

Figure 10 illustrates the findings for a single MoA with the p-values from the literature study plotted against the proportion of experts responding ‘yes’ (or ‘no’) showing the regression line and with each data point being one BCT labelled as in BCTTv1. 17 The MoA ‘behavioural cueing’ is strongly associated in both studies with the BCT ‘7.1 prompts and cues’ and BCT ‘12.5 adding objects to the environment’ for ‘yes’ responses by experts, indicating agreement that these BCTs may act via this MoA in changing behaviour.

FIGURE 10.

Relationship between literature and expert consensus when links between BCTs and MoA ‘behavioural cueing’ have been proposed. Note that each dot represents one BCT (with BCTTv1 label). The line represents the prediction from the multilevel model omitting BCTs for which p = 1 in the literature study (but the expert consensus values for such BCTs are shown).

Detailed scatterplots for each of the 25 MoAs (not for ‘norms’ as this was not found in the literature study), showing which BCTs are linked to which MoAs in the two studies along with the labels for each BCT, can be accessed in Report Supplementary Material 2.

The multilevel models are reported in Table 6 and details of unstandardised regression intercepts and slopes for each MoA are shown in Table 7. Because the results are broadly similar whether p-values of 1 (which occur when a BCT is not linked to the target MoA) are included or removed, the latter are reported here, but both are presented in Tables 6 and 7. The total variance explained by the ‘yes’ model was 42.8%. The comparable value for the ‘no’ responses was 45.0%. As there was little difference in slopes between MoAs, larger intercepts for ‘yes’ responses indicate greater agreement across studies that a link exists: the slopes indicate the relationship between the studies for each MoA. Apart from the MoA ‘norms’, which was not found in the literature synthesis study, each MoA shows negative slopes for ‘yes’ responses and positive slopes for ‘no’ responses, indicating moderate concordance between the results of the two studies for virtually all MoAs.

TABLE 6 - Multilevel model predicting judgements of expert consensus from the literature synthesis

*p < 0.01; **p < 0.01; ***p < 0.001.

	‘Yes’ expert consensus		‘No’ expert consensus
	All data	p = 1 omitted	All data	p = 1 omitted
Fixed effects, estimated beta weights (standard error)
Intercept	0.474 (0.037)***	0.510 (0.036)***	0.295 (0.030)***	0.270 (0.029)***
Literature	–0.325 (0.031)***	–0.464 (0.042)***	0.327 (0.240)***	0.439 (0.035)***
Random effects, variances (standard error)
Level 2: MoA
Intercept	0.029 (0.01)***	0.025(0.009)*	0.017 (0.007)*	0.014 (0.006)
Literature	0.016 (0.007)*	0.014 (0.011)	0.006 (0.004)	0.005 (0.007)
Level 1: BCT
Intercept	0.045 (0.002)***	0.059 (0.004)***	0.056 (0.002)***	0.058 (0.003)***

TABLE 7 - Predicting expert consensus from literature study for each MoA for ‘yes’ and ‘no’ responses

Int, intercept; Slp, slope.

Note

Multilevel modelling with and without p = 1 BCTs. For each MoA, the intercept indicates the degree of agreement between experts and literature when the probability of a link in the literature is greatest. The slopes indicate the agreement over reducing levels of the probability in the literature.

MoA	‘Yes’ all			‘Yes’, p = 1 BCTs removed			‘No’ all			‘No’, p = 1 BCTs removed
	n	Int	Slp	n	Int	Slp	n	Int	Slp	n	Int	Slp
Knowledge	56	0.48	–0.37	29	0.52	–0.50	56	0.30	0.36	29	0.26	0.43
Skills	56	0.35	–0.29	36	0.40	–0.41	56	0.47	0.35	36	0.39	0.54
Social/professional role and identity	56	0.24	–0.19	19	0.32	–0.40	56	0.46	0.35	19	0.39	0.52
Beliefs about capabilities	56	0.58	–0.36	51	0.60	–0.43	56	0.18	0.29	51	0.14	0.38
Optimism	56	0.22	–0.15	19	0.29	–0.38	56	0.38	0.29	19	0.36	0.49
Beliefs about consequences	56	0.63	–0.44	37	0.66	–0.55	56	0.19	0.35	37	0.20	0.37
Reinforcement	56	0.54	–0.33	20	0.56	–0.51	56	0.26	0.31	20	0.26	0.44
Intentions	56	0.59	–0.25	42	0.62	–0.33	56	0.14	0.22	42	0.08	0.34
Goals	56	0.61	–0.38	28	0.63	–0.53	56	0.19	0.29	28	0.18	0.37
Memory, attention and decision processes	56	0.43	–0.33	28	0.48	–0.47	56	0.29	0.36	28	0.27	0.43
Environmental context and resources	56	0.71	–0.54	22	0.71	–0.62	56	0.20	0.40	22	0.22	0.37
Social influences	56	0.60	–0.50	28	0.62	–0.53	56	0.27	0.43	28	0.26	0.40
Emotion	56	0.42	–0.29	20	0.46	–0.47	56	0.27	0.30	20	0.29	0.43
Behavioural regulation	56	0.59	–0.36	36	0.61	–0.41	56	0.20	0.32	33	0.15	0.38
Norms	–	–	–	–	–	–	–	–	–	–	–	–
Subjective norms	56	0.41	–0.34	25	0.46	–0.44	56	0.36	0.40	25	0.30	0.46
Attitude towards the behaviour	56	–0.57	–0.43	27	0.59	–0.54	56	0.20	0.35	27	0.21	0.37
Motivation	56	0.76	–0.40	36	0.77	–0.43	56	0.04	0.27	36	0.03	0.29
Self-image	56	0.34	–0.17	15	0.39	–0.41	56	0.30	0.27	15	0.30	0.47
Needs	56	0.15	–0.12	6	0.21	–0.36	56	0.57	0.21	6	0.55	0.66
Values	56	0.24	–0.13	4	0.27	–0.38	56	0.39	0.29	4	0.42	0.55
Feedback processes	56	0.60	–0.42	12	0.61	–0.52	56	0.28	0.36	12	0.24	0.42
Social learning/imitation	56	0.38	–0.33	10	0.43	–0.46	56	0.47	0.36	10	0.41	0.54
Behavioural cueing	56	0.60	–0.37	21	0.61	–0.54	56	0.23	0.29	21	0.26	0.42
General attitudes/beliefs	56	0.46	–0.34	3	0.49	–0.47	56	0.28	0.34	3	0.27	0.43
Perceived susceptibility/vulnerability	56	0.37	–0.31	11	0.43	–0.49	56	0.41	0.36	11	0.38	0.50

Development of an online tool to make results accessible to users: the Theory and Technique Tool

An online interactive tool was developed to include each of the 1456 BCT–MoA links in an interactive format incorporating the data from all studies. Each BCT–MoA combination was classified as a ‘link’, a ‘non-link’, ‘inconclusive’ or ‘no evidence’ and presented in a colour-coded matrix of BCTs by MoAs. Clicking on any BCT–MoA cell of the matrix reveals the evidence for that combination from the three studies. Figure 11 shows a section of the matrix in a screenshot of the homepage of the tool, which can be accessed at https://theoryandtechniquetool.humanbehaviourchange.org/ (accessed 1 January 2020). Figures 12 and 13 show the screenshot following clicking on a ‘link’ and a ‘non-link’ and Figures 14 and 15 show the Resources and Collaboration portals, respectively.

FIGURE 11.

Theory and Technique Tool (TaTT) illustration showing home page.

FIGURE 12.

Theory and Technique Tool (TaTT) illustration of results of clicking on a cell for a ‘link’.

FIGURE 13.

Theory and Technique Tool (TaTT) illustration of results of clicking on a ‘non-link’.

FIGURE 14.

Theory and Technique Tool (TaTT) illustration of screens for Resources.

FIGURE 15.

Theory and Technique Tool (TaTT) illustration of screens for Connect & Discuss.

Discussion

This study aimed to triangulate evidence between a literature synthesis study of 277 intervention reports and a consensus study of 100 experts. Both studies provided evidence of potential links between BCTs and their MoA. However, by examining the degree of agreement between findings based on two quite different methods, the results provide more reliable evidence of hypothesised links. As a consequence, results could be used to provide a useful resource for evidence synthesis, the design and evaluation of interventions, and for identifying priorities for empirical research on BCT–MoA relationships.

Overall, there was evidence of agreement between the results of the two previous studies for 25 of the 26 MoAs investigated (as ‘norms’ was not found in the literature). There was also agreement for both links (where experts responded ‘definitely yes’) and non-links (where experts responded ‘definitely no’). Comparison of the matrices of BCT–MoA combinations reaching the threshold for a ‘link’ or a ‘non-link’ identified 37 links and 460 non-links. Following resolution by a further 16 ‘reconciliation’ experts, this increased to 92 hypothesised links that could be used to guide the design of interventions. In addition, 465 ‘non-links’ were agreed on, indicating the specific BCTs in BCT–MoA combinations that are unlikely to be effective in changing specific MoAs. The results also identified BCT–MoA combinations where there is insufficient evidence to hypothesise that specific BCTs would or would not effect change via specific MoAs.

Triangulation of the data obtained by two different methods resulted in evidence of substantial agreement. This provides evidence of replicability and thus greater confidence in the shared findings and a more secure basis for both interpreting intervention results theoretically and designing interventions. Although the methods used are quite different, they are not completely independent. On the one hand, experts’ judgements are likely to have been influenced by their knowledge of the published literature and, on the other hand, experts are likely to have designed and published some of the studies reviewed in the literature synthesis. Nevertheless, the results go beyond simple confirmation by replication of quantitative findings by identical methods, and instead used a reconciliation study to clarify discrepant and marginal findings.

The reconciliation experts resolved 60 of the 179 inconclusive BCT–MoA combinations, most often by agreeing with the previous expert consensus rather than the literature synthesis. This is likely to have occurred because of experts in both the previous expert consensus and the reconciliation studies being likely to have an awareness of current literature, whereas the literature synthesis may have provided more of a historical record of studies of behaviour change interventions. In addition, the experts in both studies could consider possible combinations, whereas the literature survey was restricted to what had actually been done and reported. Furthermore, reconciliation experts had much greater information than the original experts and comments in the discussion round of the reconciliation study indicated that they took into account all the previous evidence as well as their own opinions. However, the majority of reconciliation experts were European, perhaps due to the origin of BCT research being in the UK. Therefore, it is possible that different results may be obtained with experts from other continents but, given the spread of BCT research, this seems unlikely.

The links are relevant to a wide range of theoretical frameworks as they address the main current theoretical domains and commonly investigated theoretical constructs. The results provide a resource that may guide the selection of BCTs when an intervention designer has a hypothesis about the theoretical construct that influences the target behaviour. For example, a physiotherapist aiming to increase the activity of the stroke patient who was motivated but forgetful may find, from Figure 10, that the inclusion of the BCT ‘7.1 prompts and cues’ may be helpful.

Findings may also be used to support future research. For example, a researcher conducting a theory-based literature synthesis may use the hypothesised BCT–MoA links to suggest alternative theoretical explanations to those proposed or illustrated as being implicit within the original studies. A similar method was used in Gardner et al. ’s124 theoretical synthesis of audit and feedback interventions. The authors started by identifying techniques used in interventions and then mapped them to ‘. . . the most relevant behaviour change theories . . .’. 124 For example, from the BCTs ‘2.3 self-monitoring of behaviour’ and ‘1.4 action-planning’ they worked via the MoAs in control theory to reach a more coherent interpretation of the intervention trial results. 62 Therefore, the links identified in the current study may be helpful in clarifying other diverse intervention results. Nevertheless, it is important to remember that the current data do not provide empirical evidence of links but only indicate hypothesised links. More work is required to test whether or not the BCTs actually work by the hypothesised MoAs. The online TaTT may assist in selecting BCT–MoA combinations for further investigation and in integrating the resulting evidence.

The expert consensus and literature synthesis methods may have produced similar results as both may be informed by dominant theories of behaviour change. 125 For example, if experts had social cognitive theory126 in mind, it is plausible that a link between MoA ‘beliefs about capabilities’ might be linked to BCTs designed to enhance self-efficacy. Interventions reported in the literature are more likely to have been driven by theories involving rational, reflective, deliberative behaviour change MoAs than more recent studies involving more impulsive, associative, automatic processes, which have gained greater prominence in more recent times with the increasing emphasis on dual-processing theories. 127 For example, a greater emphasis on non-deliberative processes may lead to more intervention studies changing the physical environment. 127 In either case, identifying concordant links may enable intervention designers to make stronger connections between their theoretical framework, and resulting interventions may be more likely to overcome the criticism that interventions are more often theory-inspired than theory-based. 26,77 Even so, because of the complexity of behaviour change, the strength of BCT–MoA links may differ in different populations, settings and behavioural domains. Therefore, it would be valuable to investigate the moderating effects of context and behaviour in assessing the effects of interventions. 69

Although hypothesised links and non-links have been identified, results were inconclusive for a large number of BCT–MoA combinations. This may simply be reflective of the recency of the field. However, it may also reflect the lack of a shared terminology to enable aggregation of evidence as is a current focus of the Human Behaviour Change Project. 69 Furthermore, the current study did not investigate the co-occurrence of BCTs and it is possible that some BCTs may have a synergistic or antagonistic effect on the action of another BCT on the specified MoA. In most theories of behaviour change, theoretical constructs work together to change behaviour, and intervention mapping87 identifies parameters that add to the effectiveness of behaviour change methods. If interventions were designed with theoretical clusters in mind, then one would expect to see recurring clusters as similar theories are used and, consequently, this is investigated in Chapter 5. However, it was clear in the literature synthesis study that a single BCT could be used to target several MoAs, and that MoAs were frequently targeted by more than one BCT. BCTs working via several MoAs may be particularly useful in practical situations but less informative in clarifying theoretical questions.

The online tool provides detailed evidence for each BCT–MoA combination, giving users easy access to the information that they need without having to base an intervention plan on common sense or implicit theorising. It provides users with the results of the current, and two previous, studies and additionally allows them to interact with the tool by adding information, posting comments and collaborating with other users.

Conclusion

The concordance and reconciliation studies have triangulated two distinct sources of evidence: literature synthesis and expert consensus. The resulting set of hypothesised links between BCTs and MoAs provide a resource for researchers and practitioners (see Appendices 5 and 6 and in the interactive online TaTT). Evidence of the ‘links’ and ‘non-links’ could be used to guide intervention development and the theoretical interpretation of results of behaviour change interventions. Moreover, this study has revealed key areas where we lack evidence on potential BCT–MoA links, offering many opportunities for further research to strengthen the TaTT. Thus, the work reported could provide the basis for more systematic, co-ordinated research studies to examine and strengthen the evidence base underlying behaviour change interventions.

Abstract

Introduction

There has been increasing emphasis on the desirability of drawing on theory when developing interventions to change behaviour or when interpreting the findings of evaluations of interventions. 15,41,71,72,128,129 Theories help to make sense of the complexity of behaviour change processes by summarising, in concise form, what is known about key entities and the relationships between these entities. Systematic application of theory can guide action by identifying MoAs to target with an intervention. Having an explicit, theory-based rationale for an intervention also enables that theory to be tested and refined on the basis of evaluation results. Because theory provides an organising framework, it can also be very helpful in synthesising evidence across studies about processes of change, thus building our knowledge about not only what interventions are effective in specific circumstances, but also how interventions have their effects. However, applying theory to interventions is not just a question of selecting a theory. The process also requires one to understand which combinations of BCTs are most likely to be effective in changing the MoAs specified by the theory, and hence the behaviour(s) of interest.

Despite the increasingly recognised importance of applying theory to developing interventions,15,41,71,72,128,129 the reporting of the use of theory is variable, often incomplete and the underpinning theory may not be explicitly stated. Rather, it may have to be inferred from the intervention report. 28,101 In these cases, it may be possible to identify clusters of BCTs associated with specific theories, thus giving an indication of the likely theoretical basis of the intervention.

In many cases, theories do not directly point to which BCTs, or BCT combinations, link to their component MoAs and thus they do not provide direct guidance to intervention developers. This is illustrated in relation to a frequently cited theory of behaviour change, the theory of planned behaviour. According to one of its authors, ‘Once it has been decided which beliefs the intervention will attempt to change, an effective intervention method must be developed. This is where the investigator’s experience and creativity come into play. The theory of planned behaviour can provide general guidelines, described below, but it does not tell us what kind of intervention will be most effective’. 130 Other theories provide more guidance about principles to draw on in developing interventions. For example, Bandura’s107 social-cognitive theory proposes four methods of intervention to increase self-efficacy: graded mastery, social modelling, social persuasion and re-interpretation of physiological states. Theories of risk perception posit that interventions will be more effective if they use techniques that increase perceived self- and response-efficacy alongside increasing fear arousal, as increasing fear arousal without increasing people’s beliefs that they can do something to reduce the threat is likely to be ineffective or even counter-productive. 131

When an intervention’s theoretical underpinning is not explicitly stated, it may be inferred from the BCTs included in the intervention. For example, the intervention ‘audit and feedback’ has considerable evidence of its effectiveness in changing clinician behaviour but there is less knowledge about how the intervention brings about its effects. 132 The combination of BCTs included in audit and feedback was analysed to investigate possible theoretical underpinnings. Three key BCTs were identified: ‘1.1 goal-setting’, ‘1.4 action-planning’ and ‘2.2 feedback on behaviour. 124 These were all linked to a self-regulation theory of behaviour change: control theory. 133 The theory posits that these BCTs work synergistically rather than additively, increasing the power of using them together. The authors of the updated Cochrane review62 were able to use this knowledge in a theory-based analysis of audit and feedback, which found that interventions including all three BCTs were more effective than those using feedback alone. 62 This is a good example of ‘retro-fitting’ a theoretical base from frequently co-occurring groups of BCTs that makes explicit what may have been an implicit theoretical understanding of those who developed the intervention. When implicit theories are made explicit, they become more useable and useful to others and therefore help to advance understanding and practice. 26,134,135

This chapter reports the final study in the programme of research, which investigated whether or not frequently occurring combinations of BCTs used in behaviour change interventions could be linked to behaviour change theories. We addressed this question with three main aims:

1. To what extent do BCTs co-occur across published reports of behaviour change interventions?

2. To what extent do behaviour change experts identify links between identified groups of BCTs and behaviour change theories?

3. For explicitly theory-based interventions, to what extent do the experts’ identification of theories linked to groups of BCTs match the theories stated by authors of interventions using similar groups of BCTs?

Methods

Results

Discussion

This study identified five distinct co-occurring groups of BCTs in an analysis of 277 published reports of behaviour change interventions, with few BCTs statistically loading on more than one factor (BCT group). The results suggest that there is a shared explanatory model among authors concerning which BCTs are likely to have a synergistic effect when combined. This assumption may have been implicit rather than explicitly stated. There was consensus among experts on three of these groups of co-occurring BCTs linking to one or more of five theories. Although several theories were mentioned below criterion level, experts did not reach the criterion for a link of the same theory to multiple BCT groups, suggesting that experts were able to implicitly identify a theory based on sets of frequently co-occurring BCTs.

Examination of the component BCTs suggested that group 1 focused on intention, goals and planning; group 3 focused on changing beliefs about norms and behaviour, and antecedents and consequences of behaviour; and group 4 focused on instruction, demonstration, practice and modelling. The links with theory were coherent, with group 1 being linked to the HAPA and self-regulation theory, group 3 being linked to the theory of planned behaviour, and group 4 being linked to social cognitive theory and self-efficacy theory.

These results demonstrate that it is possible to infer theories underpinning behaviour change interventions by examining groups of BCTs used together in interventions. This enables the theoretical understanding of interventions that have not been explicitly developed on the basis of theory. However, two groups of BCTs found to co-occur in intervention reports were not consistently linked to a theory, despite experts being provided with descriptions of 83 theories and having knowledge of others. It may be that intervention developers were not basing their interventions on likely MoAs, or that they were and the absence of a linked theory reflected a gap in theoretical knowledge in the field.

Expert discussion in the consensus study suggested that there was confusion about how to apply theories to intervention development resulting from the large number of theories and lack of guidance about their application. This is likely to explain the substantial number of interventions that were not reported as theory-based, as shown in Table 11. There appears to be much greater clarity about links between individual BCTs and individual MoAs than about links between BCT groups and theories (groups of mechanisms), as evidenced by the triangulation study reported in Chapter 4. 103 This study found 92 links identified in both the literature analysis and the expert consensus. 101,102 Further research comparing these links with the BCT group–theory links would provide another layer of validation of both sets of links.

In the comparison of the literature analysis and the expert consensus exercise, four out of the five links were also made by study authors, demonstrating consistency between explicit theorising in published reports and implicit theorising by working in reverse order, starting with co-occurring BCT groups. In considering the likelihood of a match between the theory associated with a group of BCTs as agreed by the experts and the theory reported by the study authors, it is important to bear in mind that in each case a choice is made from a very large number of possible theories. The experts were offered a source with 83 theories but were free to name any additional theories. The authors had no restrictions and one can assume that they could have chosen from at least 83 theories. However, there may be other constraints on the number of possible theories that experts or authors were choosing from. Nevertheless, given the current level of understanding in theory and intervention development, this methodology seems unlikely to enable readers to pinpoint a precise theoretical underpinning for a published intervention based on the combination of BCTs used. The strongest match was for group 4 and social cognitive theory (and self-efficacy theory), where 40 (+ 8) of 152 experts’ responses matched the authors’ theory. This may have been more successful as Bandura107 gives clear guidance on the methods that can be used to enhance a key construct in the theory, namely self-efficacy, whereas for other popular theories, such as the theory of planned behaviour, no such guidance is offered. If so, then developing a shared approach to working from theory to MoAs to the choice of BCTs could advance this field. The evidence and the TaTT described here may contribute to clearer delineation of the steps between theory and intervention.

Limitations

The results of the current study should be interpreted in the light of the size of the database: 277 interventions of evaluations with a health focus. Further investigation with more, or more diverse, interventions may produce different findings. Another further limitation is that factor analysis requires judgement when, as in this case, there are several possible numbers of factors with an acceptable model fit. We selected the factor solution with the fewest number of cross-loading BCTs; however, a different solution may have yielded different results. Another word of caution is that the 25 experts were mainly from health and psychology backgrounds. A larger and more diverse group may have provided different judgements. However, the congruence between the expert consensus and the literature analysis (see Chapters 2–4) provides a degree of confidence in the results.

A drawback of intervention designers and researchers not using theory routinely in their work is that it precludes the use of organising summary frameworks that enable the accumulation of knowledge over studies and domains about processes of change. 128 However, parallel programmes of research are addressing this problem. The first is the application of an ontology-based modelling system to represent 77 theories both diagrammatically and in computer-readable format in a searchable database. 140 The second is the development of a behaviour change intervention ontology, an organising knowledge framework, produced as part of the Human Behaviour-Change Project69 [www.humanbehaviourchange.org (accessed 1 January 2020)].

Conclusions

The five BCT group–theory links identified across literature analysis and expert consensus can inform intervention development, evaluation and evidence synthesis. This supports clear and systematic thinking about how theories may be applied by identifying appropriate groups of BCTs. It also enables evaluation of the processes of change of interventions that have not been explicitly developed on the basis of theory. The comparison between the expert consensus and (where available) the authors’ explicit theories suggests that it may also be possible to retro-fit theory to interventions when authors have not made an explicit theory link, but this is likely to be most successful when theories are specific about intervention methods. The evidence from this and related work can lead to advances in intervention design, theoretical thinking about behaviour change and creating effective interventions.

Summary of purpose, methods and results

Many of the health challenges worldwide may be addressed by changing people’s behaviour. Behaviours such as physical inactivity, alcohol misuse, smoking and some dietary behaviours contribute to deaths and disability from cancers, cardiovascular diseases, type 2 diabetes, and so on, and interventions have been designed to change behaviours with a view to reducing exposures to these health risks. However, although there have been many interventions, the effects of these are variable and further information is needed to enhance their success. More information is needed about how interventions work to generate more useful theories of behaviour change, to guide research and to achieve more successful implementable interventions. Recent research has developed a taxonomy of 93 well-defined BCTs but we need more information on their MoAs (i.e. how these techniques might work to change behaviour). We therefore aimed to examine links between BCTs and MoAs and to make the results available in a format that is usable by researchers, practitioners and policy-makers.

First, we conducted a literature synthesis study in which the links made by authors of 277 behaviour change intervention studies were extracted and links occurring with significant frequency were identified. Second, behaviour change experts were engaged in a three-round consensus study in which they discussed and rated their confidence that each of 61 BCTs worked through each of 26 MoAs: 90 links and 460 ‘non-links’ reached the pre-set threshold of 80% agreement. The third study aimed to triangulate the findings of the first two and to reconcile differences. We viewed triangulation as important both as good scientific practice, and to be able to discuss discrepancies and identify new agreed links. We found agreement between the frequency of links in the first study and links or non-links in the second. Discrepant or uncertain BCT–MoA combinations were brought to a reconciliation consensus study with a new group of experts. Following their discussions and ratings, the final total was 92 links and 465 non-links. In the fourth study, we explored whether or not it was possible to deduce any shared thinking about how BCTs work together from the combinations of BCTs that authors frequently use together in interventions, particularly as the theories underlying published behaviour change interventions are often implicit or under-reported. We examined whether or not groups of BCTs identified through factor analysis revealed shared theoretical underpinnings. We found that experts agreed on underlying theory for three groups of BCTs and that the theories were also identified by authors using similar sets of BCTs.

Our results provide a framework that policy-makers and practitioners can use to guide the selection of BCTs to include in interventions. Table 14 suggests how the TaTT might be used as a guide to intervention by a policy-maker faced with a problem of spreading infection or by a general practitioner managing a diabetic patient who is having difficulty with weight loss.

Given the state of the field, the findings from our programme of research cannot be used to infer that the BCTs are effective in targeting the MoA, only that the links identified are ‘best bets’ for interventions to be effective in changing behaviour by changing the MoA, and that non-links are unlikely to be effective. However, as researchers become more mindful of the hypothesised links between BCTs and MoAs, they will be in a better position to generate the evidence needed to test the effectiveness of the proposed linkages. The creation of the TaTT based on the findings of studies 1, 2 and 3 provides easy identification of links for which effectiveness needs to be tested [https://theoryandtechniquetool.humanbehaviourchange.org/ (accessed 1 January 2020)]. The tool makes our results usable by researchers, practitioners and policy-makers. We believe that by enabling the pooling of resources and facilitating online interactive discussion and collaboration, the TaTT will be a key tool for generating the empirical work that is needed to move the field forward.

Strengths and limitations

We have identified several strengths of this programme of work:

1. Need and relevance – this project is a response to the needs of the field of behavioural science and intervention development more generally. The development of the BCTTv1 was an advance in providing behavioural science with a method for specifying intervention components using a common language. Over the last decade or so, there has been increasing recognition within behavioural medicine and population health research of the desirability of informing intervention development and evaluation with theory; however, there is a lack of guidance for doing this.

2. Triangulation – the methodology for generating final results was based on an analysis that triangulated two sources of evidence. The first was an analysis of published literature (see Chapter 2) reporting how authors conceptualise links between BCTs and MoAs over 10 behavioural domains and ranging from 1982 to 2016. The second was a consensus exercise of 100 experts spanning Europe, North America, Australasia and Africa. The study demonstrated agreement about both the presence of links and also the absence of links (i.e. agreement that BCTs were not linked to MoAs).

3. Multidisciplinary – we received advice from a scientific advisory board representing a deliberately broad range of countries, disciplines and topics of study.

4. Online tool – we made the findings maximally accessible and therefore useable by developing an open access online TaTT resource (https://theoryandtechniquetool.humanbehaviourchange.org/) that enables users to quickly find whether links are present, absent, uncertain or unknown along with the data that these categories are based on.

5. BCT and MoA combinations – in recognition of the fact that BCTs are often used in combination and theories usually involve more than one MoA, we investigated combinations of BCTs and theories as a whole in addition to single BCT–MoA links.

6. Dissemination and engagement – we disseminated widely using social media [e.g. blogs, Twitter (www.twitter.com; Twitter, Inc., San Francisco, CA, USA)] as well as via scientific conferences and journal articles. 101–104 We also engaged the research and intervention designer communities via workshops that gave participants experience of using the TaTT as well as learning about the work and considering future research possibilities and applications of the TaTT.

7. Collaboration enablement – in recognition of the desirability of making research as efficient and productive as possible, we included facilities for collaborating and sharing resources within the TaTT (https://theoryandtechniquetool.humanbehaviourchange.org/). This is a facility for researchers to upload information, evidence and research about specific links and/or seek collaborators for joint research.

A limitation of the work is that a project of this scale could not investigate all possible BCTs and MoAs, given the size and complexity of this kind of data set. The BCTTv1 provides labels, definitions and evidence from previous studies, which allowed identification of frequently cited BCTs for inclusion. However, there was no similar taxonomy of MoAs and, therefore, no reliable set of labels and definitions. As a result, the MoAs investigated represent the broad theoretical domains of the theoretical domains framework38 supplemented by MoAs occurring frequently in behaviour change theories. Furthermore, BCT–MoA combinations that are not covered here could be investigated as links or non-links. In addition, we were not able to include all behavioural domains in our analyses nor behavioural experts from all continents. The literature search identified more studies addressing lifestyle behaviours and fewer dealing with health service utilisation or the behaviour of health-care professionals. The behavioural experts were selected based on self-definition by those responding to our calls and were predominantly working in the field of psychology (60%) and in academic settings (75%). However, we made efforts to cover a range of domains and the MoAs were largely generated via a review of theories guided by anthropologists, sociologists, economists and psychologists. For future work, we would recommend checking expertise by web searches and using a multidisciplinary advisory board to generate experts from wider disciplinary backgrounds.

Another important issue to bear in mind is that this work is based on expert opinion, reporting and judgement; it is not based on empirical evidence about the range of effectiveness of links in published evaluations. This work did not set out to do this, but it lays the basis for others to systematically review the evidence for the links and to empirically test the links.

Sustainability of resources developed for use by the research community is a challenge, given that there is no model for the long-term sustainability of such resources outside the commercial one. The Cochrane Collaboration and the Equator set of guidelines including Consolidated Standards of Reporting Trials (CONSORT) have transformed the quality of scientific research and evidence synthesis. However, they are maintained with difficulty on the basis of piecemeal grants. For the BCTTv1 and the TaTT, we are exploring the model of a not-for-profit company to sustain and develop the resources as the need and opportunities arise.

Recommendations for researchers and intervention designers

TaTT can be used by researchers and intervention designers in the following ways.

Existing interventions:

• Intervention reporting – enables reporting of intervention techniques and theoretical mechanisms in a standardised fashion, using a shared language. It enables the explicit reporting of theoretical processes underpinning interventions.

• Intervention improvement – enables an analysis of current interventions in terms of the number of BCT–MoA links that are consistent with the evidence summarised in the TaTT and those that are inconsistent. This information can be used to inform potential improvements.

• Intervention tailoring – offers suggestions for intervention designers regarding potential BCTs to use given the need to engage groups of individuals who differ in MoAs (e.g. knowledge, attitudes towards the behaviour).

Intervention development:

• Provides information about which BCTs could be used to target MoAs that the intervention is designed to target (e.g. increasing intention, changing emotions), and which BCTs not to use.

• Suggests theoretically coherent combinations of BCTs.

Understanding how interventions work:

• Process evaluation –selecting BCT–MoA combinations that are links in TaTT provides a systematic method for identifying the MoAs to evaluate (or not evaluate) for component BCTs.

• Intervention effects – suggests likely MoAs for interventions that have been developed without being explicitly based on theory, based on the BCTs included, providing a basis for intervention testing.

• Ineffective interventions – supports the generation of hypotheses about why ineffective interventions did not have predicted effects by identifying ‘non-links’ between BCTs and MoAs.

Theory development:

• Facilitates the specification of propositions within theories about links between interventions and theoretical mechanisms, enhancing the ability to develop studies that can test these propositions and, in turn, generate an evidence base that can inform the refinement of theories.

Future research

Contributions of authors

Susan Michie (https://orcid.org/0000-0003-0063-6378) (Professor of Health Psychology and Principal Investigator) and Marie Johnston (https://orcid.org/0000-0003-0124-4827) (Professor of Health Psychology, co-investigator) led the development of the research proposal, study design and protocol. Together with co-investigators Alexander J Rothman (https://orcid.org/0000-0003-3163-1895) (Professor of Psychology), Marijn de Bruin (https://orcid.org/0000-0003-3100-6803) (Professor of Health Psychology) and Michael P Kelly (https://orcid.org/0000-0002-2029-5841) (Professor and Senior Visiting Fellow), they provided the scientific leadership for the project.

Rachel N Carey (https://orcid.org/0000-0002-3299-9723) (Research Associate) and Lauren EC Bohlen (https://orcid.org/0000-0003-4207-9872) (Research Associate) contributed to the design of the four studies and led the conduct of the research.

Hilary NK Groarke (https://orcid.org/0000-0001-6282-3254) (Research Assistant) led the conduct of study 4.

Niall C Anderson (https://orcid.org/0000-0003-1834-9086) (Senior Research Associate) contributed to and revised the final draft of this manuscript.

Silje Zink (https://orcid.org/0000-0003-4077-8629) (Research Assistant) prepared, contributed to and revised the final draft of this manuscript.

All the authors contributed to the development of the protocol and the writing of the manuscript, including its final version.

Publications

Carey RN, Connell LE, Johnston M, Rothman AJ, de Bruin M, Kelly MP, Michie S. Behaviour change techniques and their mechanisms of action: a synthesis of links described in published intervention literature. Ann Behav Med 2018;53:693–707.

Connell LE, Carey RN, de Bruin M, Rothman AJ, Johnston M, Kelly MP, Michie S. Links between behaviour change techniques and mechanisms of action: an expert consensus study. Ann Behav Med 2018;53:708–20.

Johnston M, Carey RN, Connell LB, Johnston DW, Rothman AJ, de Bruin M, et al. Linking behaviour change techniques and mechanisms of action: triangulation of findings from literature synthesis and expert consensus [published online ahead of print August 4 2020]. Transl Behav Med 2020.

Bohlen LC, Michie S, de Bruin M, Rothman AJ, Kelly MP, Groarke HN, et al. Do combinations of behaviour change techniques that occur frequently in interventions reflect underlying theory? Ann Behav Med 2020;54:1–16.

Data-sharing statement

We have published our data set online through Open Science Framework (https://osf.io/7qjvn/) and through our interactive online tool (https://theoryandtechniquetool.humanbehaviourchange.org/tool), with the aim of the data being transparent and usable by the research community. We provide the full database of links and not just those links that met the p < 0.05 criterion, such that researchers can draw on all of the data as needed (e.g. using criteria of their choosing). We welcome further data extraction and/or additional analyses by researchers who are interested in examining these characteristics.

Data in the form of BCT coding of the 277 interventions identified through this study have been added to an open access online database that collates intervention articles in which BCTs have been identified (www.bct-taxonomy.com/interventions). Researchers can search this database using a number of criteria, including author, year, BCT and target behaviour. It also includes an option for researchers to upload their or others’ articles where BCTs have been coded using BCTTv1. The study findings have been incorporated into the interactive, online BCT–MoA tool, which allows users to select a particular BCT–MoA link and view the associated data. There is also a facility for researchers to add data and other information (e.g. ongoing research projects) relating to a particular link. We hope that this resource will facilitate collaborative research efforts to accumulate empirical evidence about each of the links in the matrix. Any other queries should be addressed to the corresponding author for consideration.

Disclaimers

This report presents independent research funded under a MRC–NIHR partnership. The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, the MRC, NETSCC, the HS&DR programme or the Department of Health and Social Care. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, the MRC, NETSCC, the HS&DR programme or the Department of Health and Social Care.

Content of the guidelines for round 1 of the consensus exercise

What is a mechanism of action?

These are defined as ‘the processes through which a BCT affects behaviour’. The following diagram may be helpful in conceptualising the processes through which a particular BCT might affect behaviour.

Sample question

Does the behaviour change technique goal-setting (behaviour) change behaviour through the mechanism: an awareness of the existence of something (knowledge)?

□ Definitely no

□ Probably no

□ Uncertain/don’t know

□ Probably yes

□ Definitely yes

Note: Additional information provided in the guidelines document was excluded from this paper, this information included a project summary, BCT definitions and MoA definitions. For more information about the project see the project protocol, published in Annals of Behavioral Medicine,100 for BCT definitions see the BCT taxonomy app or website: www.bcts.23.co.uk, (accessed 1 January 2020), and for more information about the MoAs included in this study, see Table 19.

FIGURE 18.

Sample statistical summary sent to experts after the completion of round 1 for the BCT goal-setting (behaviour). Note, the above distribution depicts the frequency with which experts thought that the BCT ‘goal-setting (behaviour) was linked to any of the 26 MoAs. For round 1, yes = definitely yes and probably yes, no = definitely no and probably no.

FIGURE 19.

Proportion of experts rating ‘possibly’.

FIGURE 20.

Proportion of experts rating ‘uncertain/don’t know’.

TABLE 20 - Agreed links between BCTs and MoAs (by BCT)

The table above indicates mechanisms of action for which ≥ 80% of experts responded ‘Definitely yes’ in response to the question, ‘When [BCT X] works, does it work through changing [MoA Y]?’, or ‘Definitely no’. The proportion of experts selecting a response option is indicated, with the number of experts rating each link varying by BCT.

BCT	MoA
	‘Definitely yes’ linked	Proportion of experts	‘Definitely no’ not linked	Proportion of experts
1.1 Goal-setting (behaviour)	Goals	0.95	Social learning	0.90
	Intention	0.95	Social/professional role and identity	0.90
	Motivation	0.80	Perceived susceptibility and vulnerability	0.90
			Skills	0.85
			Social influences	0.80
			Needs	0.80
1.2 Problem-solving	Behavioural regulation	1	Normative beliefs	0.95
	Beliefs about capabilities	0.95	Norms	0.95
			Social learning	0.95
			Social/professional role and identity	0.90
			Values	0.90
			Needs	0.90
			Reinforcement	0.86
			Social influences	0.81
			General attitudes and beliefs	0.81
			Emotion	0.81
1.3 Goal-setting (outcome)	Goals	1	Perceived susceptibility/vulnerability	1.00
	Intention	0.86	Social/professional role and identity	0.95
			Norms	0.95
			Social learning/imitation	0.95
			Skills	0.90
			Social influences	0.90
			Subjective norms	0.90
			Knowledge	0.86
			Emotion	0.86
			Values	0.86
			General attitudes and beliefs	0.86
			Environmental context and resources	0.81
			Needs	0.81
1.4 Action-planning			Social/professional role and identity	0.95
			Norms	0.95
			Subjective norms	0.95
			Needs	0.95
			Social learning/imitation	0.95
			Social influences	0.89
			General attitudes and beliefs	0.89
			Perceived susceptibility/vulnerability	0.89
			Knowledge	0.84
			Reinforcement	0.84
			Values	0.84
			Feedback processes	0.84
1.5 Review behaviour goal(s)	Feedback processes	0.95	Social learning/imitation	0.90
	Goals	0.90	Norms	0.85
			Perceived susceptibility/vulnerability	0.85
			Social/professional role and identity	0.80
			Subjective norms	0.80
			Needs	0.80
1.6 Discrepancy between current behaviour and goal	Feedback processes	1	Social learning/imitation	1.00
	Goals	0.81	Social/professional role and identity	0.90
			Reinforcement	0.90
			Norms	0.90
			Subjective norms	0.90
			Environmental context and resources	0.86
			Needs	0.86
			Perceived susceptibility/vulnerability	0.86
			Skills	0.81
			Optimism	0.81
			Behavioural cueing	0.81
			General attitudes and beliefs	0.81
1.7 Review outcome goal(s)	Goals	0.89	Social/professional role and identity	0.95
			Social influences	0.95
			Norms	0.95
			Subjective norms	0.95
			Social learning/imitation	0.95
			Environmental context and resources	0.89
			Perceived susceptibility/vulnerability	0.89
			Knowledge	0.84
			Skills	0.84
			Needs	0.84
			Behavioural cueing	0.84
1.8 Behavioural contract	Intention	0.80	Social learning/imitation	0.95
			Needs	0.90
			Perceived susceptibility/vulnerability	0.90
			Skills	0.85
			Knowledge	0.80
1.9 Commitment	Intention	0.84	Skills	0.84
	Motivation	0.84	Reinforcement	0.84
			Social learning/imitation	0.84
2.1 Monitoring of behaviour by others without feedback			Needs	0.89
			Optimism	0.84
2.2 Feedback on behaviour	Feedback processes	0.95	Needs	0.84
			Social learning/imitation	0.84
2.3 Self-monitoring of behaviour	Feedback processes	0.80	Social learning/imitation	0.95
			Norms	0.90
			Subjective norms	0.90
			Social/professional role and identity	0.85
			Needs	0.85
			Values	0.85
			General attitudes and beliefs	0.85
			Perceived susceptibility/vulnerability	0.85
			Social influences	0.80
2.4 Self-monitoring of outcome(s) of behaviour	Feedback processes	0.81	Norms	1
			Social influences	0.95
			Subjective norms	0.95
			Needs	0.95
			Social learning/imitation	0.95
			Social/professional role and identity	0.90
			Perceived susceptibility/vulnerability	0.90
			Values	0.86
			General attitudes and beliefs	0.81
2.5 Monitoring of outcomes of behaviour without feedback			Needs	0.95
			Social learning/imitation	0.95
			Skills	0.86
			Social/professional role and identity	0.86
			Optimism	0.86
			Beliefs about capabilities	0.81
			Memory, attention, and decision processes	0.81
			Environmental context and resources	0.81
			Norms	0.81
			Values	0.81
			General attitudes and beliefs	0.81
2.6 Biofeedback	Feedback processes	0.89	Social/professional role and identity	0.89
			Norms	0.89
			Subjective norms	0.89
			Social learning/imitation	0.89
			Social influences	0.84
			Values	0.84
2.7 Feedback on outcome(s) of behaviour	Feedback processes	0.80	Subjective norms	0.80
			Social learning/imitation	0.95
			Environmental context and resources	0.90
			Norms	0.90
			Perceived susceptibility/vulnerability	0.85
3.1 Social support (unspecified)	Social influences	0.87	Skills	0.87
			Perceived susceptibility/vulnerability	0.82
3.2 Social support (practical)	Environmental context and resources Social Influences	0.90 0.81	Perceived susceptibility/vulnerability	0.95
			Feedback processes	0.90
			Self-image	0.86
			Needs	0.86
			Values	0.81
3.3 Social support (emotional)	Emotion	0.84	Feedback processes	0.84
4.1 Instruction on how to perform the behaviour	Skills	0.86	Needs	0.90
			Perceived susceptibility/vulnerability	0.90
			Values	0.87
			Emotion	0.84
			Social/professional role and identity	0.82
			Subjective norms	0.82
			Norms	0.81
4.2 Information about antecedents	Knowledge	0.86	Needs	0.95
	Behavioural cueing	0.81	Social/professional role and identity	0.90
			Norms	0.90
			Subjective norms	0.90
			Values	0.90
			Feedback processes	0.90
			General attitudes and beliefs	0.90
			Skills	0.86
			Self-image	0.86
			Social learning/imitation	0.86
			Optimism	0.81
			Reinforcement	0.81
			Social influences	0.81
			Attitude towards the behaviour	0.81
5.1 Information about health consequences	Beliefs about consequences	0.95	Reinforcement	0.95
	Knowledge	0.89	Social learning/imitation	0.95
	Perceived susceptibility and vulnerability	0.84	Skills	0.89
			Optimism	0.89
			Behavioural regulation	0.89
			Subjective norms	0.89
			Social/professional role and identity	0.84
			Environmental context and resources	0.84
			Self-image	0.84
			Needs	0.84
5.2 Salience of consequences	Beliefs about consequences	0.85	Skills	0.95
			Social learning/imitation	0.90
			Optimism	0.85
			Needs	0.85
			Social/professional role and identity	0.80
			Beliefs about capabilities	0.80
5.3 Information about social and environmental consequences	Beliefs about consequences	0.95	Skills	1
	Motivation	0.86	Social learning/imitation	0.95
			Behavioural regulation	0.90
			Feedback processes	0.90
			Beliefs about capabilities	0.86
			Optimism	0.86
			Needs	0.81
5.4 Monitoring of emotional consequences	Emotion	0.80	Skills	0.95
			Social/professional role and identity	0.90
			Memory, attention and decision processes	0.90
			Norms	0.90
			Subjective norms	0.90
			Social learning/imitation	0.90
			Environmental context and resources	0.85
			Values	0.85
			Perceived susceptibility/vulnerability	0.80
5.6 Information about emotional consequences	Beliefs about consequences	0.90	Social influences	0.80
			Social learning/imitation	0.95
			Behavioural cueing	0.90
			Skills	0.85
			Social/professional role and identity	0.85
			Memory, attention and decision processes	0.85
			Environmental context and resources	0.85
			Beliefs about capabilities	0.80
			Social influences	0.80
			Norms	0.80
			Subjective norms	0.80
6.1 Demonstration of behaviour	Social learning/imitation	0.84	Needs	0.95
			Perceived susceptibility/vulnerability	0.89
			Self-image	0.84
6.2 Social comparison	Social influences	1	Skills	1
	Normative beliefs	0.81	Needs	0.95
	Feedback processes	0.90	Behavioural regulation	0.86
	Norms	0.90	Perceived susceptibility/vulnerability	0.86
6.3 Information about others’ approval	Normative beliefs	0.89	Skills	0.95
	Norms	0.89	Needs	0.84
	Social influences	0.84
7.1 Prompts/cues	Behavioural cueing	1	Needs	1
	Environmental context and resources	0.90	Perceived susceptibility/vulnerability	1
			Knowledge	0.95
			Social/professional role and identity	0.95
			Self-image	0.95
			Values	0.95
			Social learning/imitation	0.95
			General attitudes and beliefs	0.95
			Beliefs about consequences	0.90
			Norms	0.90
			Subjective norms	0.90
			Feedback processes	0.90
			Beliefs about capabilities	0.86
			Optimism	0.86
			Attitude towards the behaviour	0.86
			Emotion	0.81
7.5 Remove aversive stimulus	Environmental context and resources	0.90	Knowledge	1
			Needs	1
			Skills	0.95
			Social/professional role and identity	0.95
			Values	0.90
			Subjective norms	0.86
			Self-image	0.86
			Perceived susceptibility/vulnerability	0.86
			Norms	0.81
			Feedback processes	0.81
			Social learning/imitation	0.81
8.1 Behavioural practice/rehearsal	Skills	0.95	Norms	0.95
	Beliefs about capabilities	0.86	Needs	0.95
			Values	0.95
			Social/professional role and identity	0.90
			Subjective norms	0.90
			Social influences	0.86
			Social learning/imitation	0.86
			Perceived susceptibility/vulnerability	0.86
			General attitudes and beliefs	0.81
8.2 Behaviour substitution			Needs	1
			Social/professional role and identity	0.95
			Social learning/imitation	0.95
			Social influences	0.89
			Emotion	0.89
			Norms	0.89
			Subjective norms	0.89
			General attitudes and beliefs	0.89
			Self-image	0.84
8.3 Habit formation	Behavioural cueing	0.85	Perceived susceptibility/vulnerability	0.90
			Needs	0.85
			General attitudes and beliefs	0.85
			Knowledge	0.80
			Optimism	0.80
			Norms	0.80
			Values	0.80
			Feedback processes	0.80
8.4 Habit reversal	Skills	0.84	Needs	0.84
8.6 Generalisation of the target behaviour			Needs	0.89
			Perceived susceptibility/vulnerability	0.84
8.7 Graded tasks	Beliefs about capabilities	0.90	Social influences	1
	Skills	0.81	Norms	1
			Needs	1
			Subjective norms	0.95
			Values	0.95
			Social/professional role and identity	0.86
			Social learning/imitation	0.86
			Perceived susceptibility/vulnerability	0.86
			Knowledge	0.81
			Behavioural cueing	0.81
9.1 Credible source			Skills	0.89
			Needs	0.89
			Reinforcement	0.84
9.2 Pros and cons	Beliefs about consequences	0.90	Social learning/imitation	1
	Motivation	0.86	Skills	0.90
	Attitude towards the behaviour	0.81	Behavioural cueing	0.86
			Social/professional role and identity	0.81
			Self-image	0.81
			Needs	0.81
9.3 Comparative imagining of future outcomes	Beliefs about consequences	1	Skills	1
	Intention	0.81	Social learning/imitation	0.95
	Motivation	0.86	Reinforcement	0.90
			Norms	0.90
			Environment	0.86
			Social influences	0.86
			Social/professional role and identity	0.81
10.1 Material incentive (behaviour)	Reinforcement	0.95	Social learning/imitation	0.84
	Motivation	0.84
10.2 Material reward (behaviour)	Reinforcement	0.89	Social/professional role and identity	0.95
			Social learning/imitation	0.95
			Perceived susceptibility/vulnerability	0.95
			Knowledge	0.89
			Skills	0.89
			Memory, attention and decision processes	0.89
			Social influences	0.89
			Norms	0.89
			Subjective norms	0.89
			Self-image	0.89
			Optimism	0.84
10.3 Non-specific reward	Reinforcement	0.95	Social/professional role and identity	0.95
	Motivation	0.84	Subjective norms	0.95
			Self-image	0.95
			Needs	0.95
			Values	0.95
			Social learning/imitation	0.95
			Perceived susceptibility/vulnerability	0.95
			Optimism	0.89
			Social influences	0.89
			Behavioural regulation	0.89
			Norms	0.89
			Knowledge	0.84
			Skills	0.84
			Feedback processes	0.84
			General attitudes and beliefs	0.84
10.4 Social reward	Social comparison	0.95	Perceived susceptibility/vulnerability	1
	Motivation	0.81	Skills	0.95
			Memory, attention and decision processes	0.90
			Social learning/imitation	0.90
			Knowledge	0.86
			Needs	0.86
			General attitudes and beliefs	0.86
10.6 Non-specific incentive			Skills	0.95
			Social/professional role and identity	0.95
			Social influences	0.95
			Social learning/imitation	0.95
			Perceived susceptibility/vulnerability	0.95
			Knowledge	0.89
			Optimism	0.89
			Norms	0.89
			Subjective norms	0.89
			Self-image	0.89
			Beliefs about capabilities	0.84
			Values	0.84
			Feedback processes	0.84
10.8 Incentive (outcome)	Intention	0.95	Skills	1
	Beliefs about consequences	0.90	Perceived susceptibility/vulnerability	1
	Reinforcement	0.90	Social/professional role and identity	0.95
			Self-image	0.95
			Social learning/imitation	0.95
			Subjective norms	0.90
			Social influences	0.86
			Norms	0.86
			Values	0.86
			Beliefs about capabilities	0.81
			Memory, attention and decision processes	0.81
			General attitudes and beliefs	0.81
10.10 Reward (outcome)	Reinforcement	0.90	Skills	1
	Motivation	0.90	Perceived susceptibility/vulnerability	1
	Beliefs about consequences	0.86	Social/professional role and identity	0.95
			Self-image	0.95
			Social learning/imitation	0.95
			Subjective norms	0.90
			Social influences	0.86
			Norms	0.86
			Values	0.86
			Beliefs about capabilities	0.81
			Memory, attention and decision processes	0.81
			General attitudes and beliefs	0.81
11.1 Pharmacological support			Skills	0.95
			Subjective norms	0.95
			Values	0.95
			Feedback processes	0.95
			Social learning/imitation	0.95
			Social/professional role and identity	0.89
			Norms	0.89
			Perceived susceptibility/vulnerability	0.84
11.2 Reduce negative emotions	Emotion	0.95	Subjective norms	0.95
	Behavioural regulation	0.90	Values	0.95
			Social/professional role and identity	0.90
			Norms	0.90
			Feedback processes	0.90
			Social learning/imitation	0.90
			Knowledge	0.86
			Behavioural cueing	0.86
11.3 Conserving mental resources	Memory, attention and decision processes	1	Social/professional role and identity	1
			Norms	1
			Subjective norms	1
			Perceived susceptibility/vulnerability	1
			Social influences	0.95
			Needs	0.95
			Values	0.95
			Feedback processes	0.95
			Social learning/imitation	0.95
			Self-image	0.90
			Reinforcement	0.86
			Emotion	0.86
			General attitudes and beliefs	0.86
			Beliefs about consequences	0.81
12.1 Restructuring the physical environment	Environmental context and resources	0.95	Social/professional role and identity	0.84
	Behavioural cueing	0.89	Optimism	0.84
			Self-image	0.84
			Needs	0.84
12.2 Restructuring the social environment	Environmental context and resources	0.95	Skills	0.90
	Behavioural cueing	0.80	Optimism	0.90
			Needs	0.90
			Beliefs about consequences	0.85
			Feedback processes	0.85
			Attitude towards the behaviour	0.80
12.3 Avoidance/reducing exposure to cues for the behaviour	Environmental context and resources	0.95	Needs	0.89
	Behavioural cueing	0.95	Feedback processes	0.95
			Social learning/imitation	0.95
			Social/professional role and identity	0.89
			Subjective norms	0.89
			Needs	0.89
			General attitudes and beliefs	0.89
			Knowledge	0.84
			Norms	0.84
			Self-image	0.84
12.5 Adding objects to the environment	Environmental context and resources	0.95	Social learning/imitation	1
	Behavioural cueing	0.95	Skills	0.86
			Social/professional role and identity	0.86
			Beliefs about consequences	0.86
			Norms	0.86
			Values	0.86
			Optimism	0.81
			Social influences	0.81
			Self-image	0.81
			Feedback processes	0.81
			Perceived susceptibility/vulnerability	0.81
12.6 Body changes			Norms	1
			Subjective norms	0.95
			Social learning/imitation	0.95
			Knowledge	0.90
			Values	0.90
			Social/professional role and identity	0.86
			Social influences	0.86
			Memory, attention and decision processes	0.81
			Feedback processes	0.81
13.1 Identification of self as role model	Self-image	0.90	Perceived susceptibility/vulnerability	0.90
			Skills	0.80
			Optimism	0.80
			Needs	0.80
13.2 Framing/reframing	Attitude towards the behaviour	0.81	Skills	0.95
			Needs	0.95
			Social/professional role and identity	0.90
			Social learning/imitation	0.90
			Reinforcement	0.86
			Norms	0.86
			Behavioural cueing	0.86
			Environmental context and resources	0.81
			Feedback processes	0.81
14.1 Behaviour cost	Reinforcement	0.95	Skills	0.84
			Social/professional role and identity	0.84
			Subjective norms	0.84
15.1 Verbal persuasion about capability	Beliefs about capabilities	1	Needs	0.95
			Social/professional role and identity	0.90
			Memory, attention and decision processes	0.90
			Social learning/imitation	0.90
			Norms	0.86
			Values	0.86
			General attitudes and beliefs	0.86
			Perceived susceptibility/vulnerability	0.86
			Feedback processes	0.81
15.4 Self-talk			Knowledge	0.95
			Environment	0.95
			Norms	0.95
			Perceived susceptibility/vulnerability	0.95
			Social influences	0.89
			Subjective norms	0.89
			Values	0.89
			Feedback processes	0.89
			Social learning/imitation	0.89
			Social/professional role and identity	0.84
			Reinforcement	0.84
			Needs	0.84
16.3 Vicarious consequences	Social learning/imitation	0.85	Skills	0.95
	Norms	0.80	Needs	0.90
			Perceived susceptibility/vulnerability	0.85

TABLE 21 - Agreed links between BCTs and MoAs (by MoA)

MoA	BCT
	‘Definitely yes’ linked	Proportion of experts	‘Definitely no’ not linked	Proportion of experts
Knowledge	4.2 Information about antecedents	0.86
	5.1 Information about health consequences	0.89
Skills	4.1 Instruction on how to perform the behaviour	0.86
	8.1 Behavioural practice/rehearsal	0.95
	8.4 Habit reversal	0.84
	8.7 Graded tasks	0.81
Social/professional role and identity			1.1 Goal-setting (behaviour)	0.90
			1.2 Problem-solving	0.90
Beliefs about capabilities	15.1 Verbal persuasion about capability	1	2.1 Monitoring of outcomes of behaviour without feedback	0.81
	1.2 Problem-solving	0.95	5.2 Salience of consequences	0.80
	8.7 Graded tasks	0.90	5.3 Information about social and environmental consequences	0.86
	8.1 Behavioural practice/rehearsal	0.86	5.6 Information about emotional consequences	0.80
			7.1 Prompts/cues	0.86
			10.6 Non-specific incentive	0.84
			10.8 Incentive (outcome)	0.81
Optimism
Beliefs about consequences	5.1 Information about health consequences	0.95
	5.2 Salience of consequences	0.85
	5.3 Information about social and environmental consequences	0.95
	5.6 Information about emotional consequences	0.90
	9.2 Pros and cons	0.90
	9.3 Comparative imagining of future outcomes	1
	10.8 Incentive (outcome)	0.90
	10.10 Reward (outcome)	0.86
Reinforcement	10.1 Material incentive (behaviour)	0.95	1.2 Problem-solving	0.86
	10.2 Material reward (behaviour)	0.89
	10.3 Non-specific reward	0.95
	10.8 Incentive (outcome)	0.90
	10.10 Reward (outcome)	0.90
	14.1 Behaviour cost	0.95
Intention	1.1 Goal-setting (behaviour)	0.95
	1.3 Goal-setting (outcome)	0.86
	1.8 Behavioural contract	0.80
	1.9 Commitment	0.84
	9.3 Comparative imagining of future outcomes	0.81
	10.8 Incentive (outcome)	0.95
Goals	1.1 Goal-setting (behaviour)	0.95
	1.3 Goal-setting (outcome)	1
	1.5 Review behaviour goal(s)	0.90
	1.6 Discrepancy between current behaviour and goal	0.81
	1.7 Review outcome goal(s)	0.89
Memory, attention and decision processes	11.3 Conserving mental resources	1
Environmental context and resources	3.2 Social support (practical)	0.90
	7.1 Prompts/cues	0.90
	7.5 Remove aversive stimulus	0.90
	12.1 Restructuring the physical environment	0.95
	12.2 Restructuring the social environment	0.95
	12.3 Avoidance/reducing exposure to cues for the behaviour	0.95
	12.5 Adding objects to the environment	0.95
Social influences	3.1 Social support (unspecified)	0.87	1.1 Goal-setting (behaviour)	0.80
	3.2 Social support (practical)	0.90	1.2 Problem-solving	0.81
	6.2 Social comparison	1	5.6 Information about emotional consequences	0.80
	6.3 Information about others’ approval	0.84
Emotion	3.3 Social support (emotional)	0.84	1.2 Problem-solving	0.81
	5.4 Monitoring of emotional consequences	0.80
	11.2 Reduce negative emotions	0.95
Behavioural regulation	1.3 Problem-solving	1
	11.2 Reduce negative emotions	0.90
Norms	6.2 Social comparison	0.90	1.2 Problem-solving	0.95
	6.3 Information about others’ approval	0.89
	16.3 Vicarious consequences	0.80
Subjective norms			2.7 Feedback on outcomes of behaviour	0.80
Attitude towards the behaviour	13.2 Framing/reframing	0.81	4.2 Information about antecedents	0.81
	9.2 Pros and cons	0.81	7.1 Prompts/cues	0.86
			12.2 Restructuring the social environment	0.80
Motivation	1.1 Goal-setting (behaviour)	0.80
	1.9 Commitment	0.84
	5.3 Information about social and environmental consequences	0.86
	9.2 Pros and cons	0.86
	9.3 Comparative imagining of future outcomes	0.86
	10.1 Material incentive (behaviour)	0.84
	10.3 Non-specific reward	0.84
	10.4 Social reward	0.81
	10.10 Reward (outcome)	0.90
Self-image	13.1 Identification of self as role model	0.90
Needs			1.1 Goal-setting (behaviour)	0.80
			1.2 Problem-solving	0.90
			2.1 Monitoring of behaviour by others without feedback	0.89
			12.3 Avoidance/reducing exposure to cues for the behaviour	0.89
Values			1.2 Problem-solving	0.90
Feedback processes	1.5 Review behaviour goal(s)	0.95
	1.6 Discrepancy between current behaviour and goal	1
	2.2 Feedback on behaviour	0.95
	2.3 Self-monitoring of behaviour	0.80
	2.4 Self-monitoring of outcome(s) of behaviour	0.81
	2.6 Biofeedback	0.89
	2.7 Feedback on outcome(s) of behaviour	0.80
	6.2 Social comparison	0.80
Social learning/imitation	6.1 Demonstration of behaviour	0.84
	16.3 Vicarious consequences	0.85
Behavioural cueing	4.2 Information about antecedents	0.81
	7.1 Prompts/cues	1
	8.3 Habit formation	0.85
	12.1 Restructuring the physical environment	0.85
	12.1 Restructuring the social environment	0.80
	12.3 Avoidance/reducing exposure to cues for the behaviour	0.95
	12.5 Adding objects to the environment	0.95
General attitude and beliefs			1.2 Problem-solving	0.81
Perceived susceptibility/vulnerability	5.1 Information about health consequences	0.84	1.1 Goal-setting (behaviour)	0.90