Reflecting widespread interest in concepts of ‘nudging’ and ‘choice architecture’, there is increasing research and policy attention on altering aspects of the small-scale physical environment, such as portion sizes or the placement of products, to change health-related behaviour at the population level. There is, however, a lack of clarity in characterizing these interventions and no reliable framework incorporating standardized definitions. This hampers both the synthesis of cumulative evidence about intervention effects, and the identification of intervention opportunities. To address this, a new tool, TIPPME (typology of interventions in proximal physical micro-environments), has been developed and here applied to the selection, purchase and consumption of food, alcohol and tobacco. This provides a framework to reliably classify and describe, and enable more systematic design, reporting and analysis of, an important class of interventions. In doing so, it makes a distinct contribution to collective efforts to build the cumulative evidence base for effective ways of changing behaviour across populations.
Unhealthy patterns of food, alcohol and tobacco consumption are major contributors to the burden of non-communicable diseases—currently accounting for more than two-thirds of deaths worldwide1,2. It is now widely recognized that the physical environments that surround us exert considerable influence on these patterns of consumption, and that changing these environments holds corollary potential as a catalyst for changing consumption. Whilst not new, the idea that behaviour can be changed in predictable ways, by changing the environments in which people make choices—‘choice architecture’3—has gained traction globally among the public, the research community and policymakers4,5. However, despite the recent popularization and intuitive appeal of these approaches, there has been an absence of definitional and conceptual clarity in characterizing such interventions, particularly regarding applications to public health. The absence of a reliable framework that incorporates standardized labels and definitions has hampered the synthesis of cumulative evidence about intervention effects, resulting in an evidence base that remains uneven and uncertain. It has also hindered the identification and discussion of opportunities to intervene to change environments.
In response to these observations, we present and provide guidance for a new tool—TIPPME (typology of interventions in proximal physical micro-environments)—that aims to improve researchers’ and practitioners’ ability to clearly and consistently classify and describe an important class of behaviour change interventions related to concepts of ‘nudging’ and ‘choice architecture’. The focus of the typology is on interventions that involve altering aspects of physical micro-environments to change health-related behaviour, here specifically applied to the selection, purchase and consumption of food, alcohol and tobacco products. A more detailed discussion of definitions and concepts follows below, but in essence, these interventions involve changing characteristics of the products themselves and the environments in which they are available, within places such as shops, restaurants, bars and workplaces. Examples include altering the portion size of food, alcohol and tobacco products, and changing their availability or position in an environment, such as providing additional healthier options to select from or placing less healthy options further away from potential consumers.
Aims of TIPPME
The aim of TIPPME is to provide a framework for reliably classifying and describing ways in which interventions can alter proximal physical micro-environments to change selection, purchase and consumption of food, alcohol and tobacco products, to achieve the following:
Facilitate the synthesis of cumulative evidence about the effects of interventions that can be mapped onto the areas of influence or responsibility of different potential actors (for example, industry, policymakers and the public), including supporting clearer reporting of intervention content in primary and secondary research.
Facilitate identification and discussion of a broader range of opportunities for interventions to be developed, implemented and evaluated. This is potentially useful for researchers and those in positions to directly alter, or advocate for changes to, commercial, public-sector or domestic environments.
Focus of TIPPME
We define the focus of the typology as follows:
Interventions or ways to alter the properties or the placement of objects or stimuli in proximal (sensorily perceptible) physical micro-environments, to elicit particular behaviours among people in those environments. These interventions are implemented in the same environment as that in which the target behaviour is performed, and are not designed to be interactive or tailored to specific individuals.
Our choice of terminology regarding ‘proximal physical micro-environments’ is intended to reflect the spatial focus of this class of interventions relative to the people exposed to them. It draws on a conceptual distinction made in the ANGELO (analysis grid for environments linked to obesity) framework6 between two levels of environment, micro and macro. Micro-environments are settings that people use for specific purposes (for example, shops, restaurants and bars) and where they interact directly with objects and stimuli in those environments. In contrast, macro-environments are the higher-level systems and infrastructure that influence the characteristics of micro-environments and the relationship between them (for example, the availability of micro-environments themselves, such as the geographical distribution of shops, restaurants and bars in a given area).
ANGELO also distinguishes between four types of environment: physical, economic, political and socio-cultural. As we are concerned with the consumption of food, alcohol and tobacco products that are themselves objects within the physical micro-environment, the stated focus of TIPPME is on the physical micro-environment. We have not attempted concurrently to map economic, political and socio-cultural environments, though we acknowledge their importance and the complex interactive relationships between them, and between interventions and outcomes. For example, these other types of environment may be manifest in any changes made to physical environments (for example, political environments may influence physical environments) or changes made to physical environments may impact the other environment types (for example, providing information may influence social norms). As well as directly influencing the nature of physical environments, economic, political and socio-cultural environments also determine the background conditions in which people are exposed to physical environments, such as times when the physical environments can be accessed, and the economic costs that are imposed on them. Applying a sociological lens, social structures constrain and enable the actions of individual human agents and set the limits of behavioural possibilities. Humans operate in an environment that is simultaneously social, biological and physical7. Here our focus is on the physical, while acknowledging this wider set of parameters.
Because physical micro-environments can be very large and encompass a wide range of functions and purposes for the people in them (for example, neighbourhoods or streets), this term is not specific enough to capture the focus of the interventions we aim to characterize. The addition of ‘proximal’ reflects our conceptual focus, as these interventions are typically implemented close (spatially and temporally) to the point of decision or performance of the people exposed to them, in order to influence behaviour enacted in that same physical environment. We have bounded the parameters of the physical environments characterized by our typology to those that are sensorily perceptible (that is, able to be seen, heard, smelt, touched or tasted) by intervention recipients. In combination with the other elements of our definition, this is intended to give an approximate indication of the likely scale of the interventions of interest, given that precisely and accurately quantifying the range of distances is not practicable. Finally, the focus of this typology excludes interventions that are designed to be interactive or tailored, meaning those in which the intervention content is not standardized for all recipients and is intended or enabled to vary dependent on their characteristics or responses. This may result from an interaction with a person or machine, such as a cafeteria worker or a computer-based system providing personalized nutritional guidance on the basis of food purchasing patterns, demographic characteristics or responses to questions. While interventions included in TIPPME are not necessarily non-interactive—in so far as people could in theory interact with them and change their content—they are not designed to elicit such interaction.
Importance of these interventions
Interventions in proximal physical micro-environments have significant potential to change behaviour for the purpose of improving population health8. This is reflected in current policy and research interest. These interventions have key advantages over many other types of behaviour change interventions. First, the nature of altering characteristics of physical environments means that these interventions have the potential to shape the behaviour of all those exposed to that environment without the need for interpersonal interaction. This means that once an intervention has been developed and implemented, there will probably be minimal ongoing resource costs associated with its continued use. Second, because physical environments have the potential to be modified in a consistent and directly measurable way, an intervention can be readily and reliably transferred to other locations, and scaled up in its application to reach larger populations. Third, because these interventions typically involve altering cues located proximally in time and space to the behaviour, their effects are probably less reliant on people purposefully and consciously engaging with the intervention over time9, or on high levels of personal agency10. This means that they may be less affected by differential, often socially patterned, cognitive or motivational resources. They therefore, in theory, have the potential to be effective across the populations to which they are applied, without widening existing health inequalities. Such potential is reflected in evidence suggesting that interventions that alter the environments to which people are exposed may be less likely to widen inequalities than individual-level education and counselling11. Finally, there is emerging evidence that this kind of public health intervention is more acceptable to the public than economic interventions such as taxes on products12,13, public acceptability being a key determinant of whether an intervention is implemented14.
Characterizing the proximal physical micro-environment
In addition to the ANGELO framework6, there have been several complementary research efforts that make reference to small-scale physical environments. The behaviour change technique taxonomy15 aims at comprehensively describing behaviour change techniques, including ‘restructuring the physical environment’, although it does not further classify interventions within this category. The intervention mapping approach16 describes a series of steps for developing interventions, and includes ‘nudging’ as one possible approach to changing determinants underlying behaviour, but does not disassemble this concept in terms of specific intervention content. Other work has focused on classifying characteristics of ‘nudging’ or ‘choice architecture’ interventions (for example, refs 17,18), but these typically concern broad theoretical principles and do not describe ways of changing physical environments in any detail. Previous attempts to map features of the physical environment that cue our behaviour or ways in which it can be changed 19,20,21,22 are unable to address our aims adequately, as they are insufficiently detailed, not systematically developed and assessed, or are not applied to consumptive health-related behaviours.
In an earlier phase of this research, we developed a provisional typology that focused specifically on the ways in which small-scale physical environments have been altered to influence food, alcohol, tobacco and physical-activity behaviours23,24. This was derived from a large-scale systematic scoping review of the research literature on ‘choice architecture’ interventions, intended to map the parameters of previous empirical research and provide a conceptual map of the evidence base, to delineate and characterize more specific intervention types. In this paper, we describe further development of this work, introducing TIPPME, which is intended to improve on and replace the provisional typology. Such development was needed because the nature of the provisional typology was determined by the existing research literature, and was therefore not designed to be applied beyond organizing that specific body of literature. Furthermore, in the early stages of TIPPME’s development (see Methods, Stages 1–3) we identified various conceptual issues concerning the provisional typology that undermined its validity and usefulness. TIPPME therefore represents an attempt to produce a more generalizable typology with a conceptually and theoretically coherent structure that can accommodate both interventions that have been developed and tested, and those that exist only in theory. Using the original scoping review process as a platform, it has been shaped in accordance with the collective understanding of experienced researchers and practitioners, with the aim of producing as complete and parsimonious an account of the phenomena of interest as possible.
Whilst related to these concepts, the focus of TIPPME has been deliberately distanced from the terminology of ‘nudging’ and ‘choice architecture’, this being potentially contentious in terms of how it has been bound to particular political and philosophical positions, and which has been inconsistently interpreted and applied. As Oliver25 highlights, for a nudge to align with the founding principles of libertarian paternalism3, it should fulfil a set of essential criteria (for example, that it is not regulatory, and does not rely on rational reasoning processes). As interventions that are claimed to represent nudges often do not meet these criteria, continued imprecise use of the term has resulted in the concept it denotes being obfuscated, and hence confusion around its meaning and potential policy value26. While interventions within TIPPME may map onto the concept of nudging in some respects, this is not a necessary feature of the typology. It is therefore instead linked to the more general and readily definable concept of the physical environment and the ways in which this can be altered to change behaviour.
This section describes only the final version of TIPPME (Table 1). The seven sequential stages of TIPPME’s development are detailed in the Methods. These stages are outlined in Table 2, Table 3 illustrates how the typology developed, showing an early version at the end of Stage 1. The results of reliability testing exercises are presented in Table 4.
The final, complete version of TIPPME is provided in Supplementary Information. In addition, it is available at http://www.bhru.iph.cam.ac.uk/resources/TIPPME (training materials are also provided here) and at https://dx.doi.org/10.6084/m9.figshare.5053672. This complete version includes full definitions and guidance, a set of instructions for use, and provides examples of interventions within each category in the typology. For illustration only, a simplified version of TIPPME is presented in Table 1. TIPPME includes and encompasses interventions that meet our stated definition of proximal physical micro-environment interventions. In terms of the wording used and examples provided, we have here applied it specifically to the selection, purchase and consumption of food, alcohol and tobacco, acknowledging that it could potentially be applied and adapted to other behaviours (see Discussion).
TIPPME comprises a matrix classification structure defining six intervention types and three different spatial foci. The rows of the typology represent different intervention types, that is, ways in which the proximal physical micro-environment can be altered to elicit changes in behaviour. There are six different intervention types (rows): availability, position, functionality, presentation, size and information. These six intervention types can be aggregated into two higher-order classes of intervention: (1) those that involve altering the placement of objects or stimuli in proximal physical micro-environments, and (2) those that involve altering the properties of objects or stimuli in proximal physical micro-environments, indicated by the column on the left edge of the table. The typology also distinguishes between three intervention foci representing differences across a spectrum in the spatial focus of interventions: product, related objects and wider environment. The combination of rows and columns means there are 18 possible intervention categories that can be applied to describe an intervention.
TIPPME provides a means of reliably classifying and describing an important class of interventions to change health-related behaviour across populations. TIPPME has benefited from an extensive, iterative and explicit development process that included reliability testing using a sample of people involved in researching and implementing interventions, representing those that are ultimately likely to use it. In line with other conceptual frameworks, this framework will, with use, be found to be imperfect, but it represents an agreement that a point of development has been reached where the typology can usefully fulfil its stated aims.
First, it provides a reliable framework for the synthesis of cumulative evidence about the effects of interventions, with the potential for a shared language. Second, it can facilitate systematic thinking about and identification and discussion of a broader range of opportunities for interventions to be developed, implemented and evaluated. In turn, resulting findings can be integrated with a growing cumulative evidence base to facilitate the development of more effective interventions. In more practical terms, this typology can feasibly be used in tasks such as classifying or organizing bodies of literature; identifying, framing and bounding primary research, as well as systematic and conceptual reviews; and providing a way of listing possible intervention strategies. It is potentially useful both for researchers and for framing the actions of those in positions to alter or influence commercial, public sector or domestic environments. This could include public health practitioners and policymakers, as well as those advocating for such changes. Finally, while the predominant focus of TIPPME is on ways of altering environments, it may also be informative in attempts to describe physical features of environments that (as opposed to being implemented as interventions) already exist and may influence behaviour accordingly (or may moderate the effectiveness of interventions that are introduced).
TIPPME also contributes to and complements ongoing efforts by the wider research community to build the foundations of a cumulative evidence base by developing domain ontologies to encode and curate research knowledge about the effects of interventions, and enable its more efficient identification, synthesis and use. Such domain ontologies include representation of the common and distinct features (or attributes) of different types of interventions, and of the proposed ‘active ingredient(s)’ that determine their effectiveness27,28,29. This encompasses representation of the content of interventions15 and of the mode, or form, of their delivery30. TIPPME contributes to these efforts by specifying the common and distinct features of a specific class of interventions. In particular, it delineates these interventions in terms of (1) their content—this being the proposed ‘active ingredient(s)’ that elicit the behavioural response, which in this case concerns the alteration of attributes of objects or stimuli in the proximal physical micro-environment, such as their size or position; and (2) the focus of that content. Ontological relationships both within TIPPME, and between TIPPME and other relevant typologies or taxonomies, are likely to be complex—particularly as there may be variation in the level of explanation or granularity applied in each case. Further development work is therefore needed to clarify and specify the form and structure of these relationships, as well as the ways in which different frameworks may be usefully applied in combination. Relatedly, TIPPME does not at present attempt to delineate the mechanisms of action that underlie each intervention type but, if it fulfils its stated aims, this should facilitate primary and secondary research directed towards furthering understanding of such mechanisms.
Whether TIPPME is viewed as a typology or, with additional development and validation, a more definitive taxonomy, depends on the epistemological position that one adopts. The way that some authors describe ontologies is unequivocally realist, whereby a properly developed ontology will describe the real world as it is, rather than just as it appears to be to the observer31. Contrary to this, phenomenologists such as Schutz32 view ontologies as theories about the nature of being in the world, and typologies and typifications as the means of seeing and interpreting that world; the plastic nature of such conceptual constructs is emphasized. A true ontology would take full account of both realist and phenomenalist perspectives and indeed our approach to developing TIPPME has drawn on both of these traditions. In the initial phase of development, we derived provisional types from empirical studies; while in later phases we have sought to use a mix of realist and other forms of knowledge to refine the typology to align with various priors (including theoretical understandings) about relations between concepts and ideas, either as we imagine the world appears to be, or as we imagine the world should be if it conformed to our prior beliefs.
We judge the level of granularity of TIPPME to be appropriate to fulfil our aims; being relatively simple but enabling discrimination between multiple intervention types. Evidence of how our provisional typology23 has been used to, for example, frame funding calls, inform policy documents and to characterize interventions in systematic reviews (see Methods), suggests that TIPPME has the potential to be similarly useable. However, its granularity could be increased, should important intervention sub-types or additional characteristics be highlighted. This could be through conducting systematic reviews of specific intervention types in which key intervention characteristics are identified (for example, refs 33,34,35), or mapping relationships between TIPPME and other classification systems.
TIPPME and the methods used to generate it have several limitations. Our first reliability testing exercise, while demonstrating that TIPPME can be reliably used by those outside of the research team, used a relatively small sample of experts, predominantly academic researchers. Whilst such a sample is probably broadly representative of some of the most likely users of the typology, there will be other groups that were inadequately represented. Furthermore, the majority of the development process was conducted by the core research team. While a wide range of disciplinary backgrounds was represented, and many of the group had extensive policy and guidance development experience, the team was weighted towards research expertise. A more rigorous and comprehensive series of assessments, probably also integrating responses from a wider cross-section of potential user groups, will be required for greater confidence in TIPPME’s reliability, particularly for use outside of the research community. A further limitation is that TIPPME is currently only applied to three consumption behaviours, which, while highly important—with the metabolic and dietary risk factors linked to food consumption, as well as smoking and alcohol use, all being amongst the most significant risk factors contributing to global disease burden2—do not encompass all of the human behaviours that significantly impact on health. Most notably, although it was included in the provisional typology23, physical activity was not included here. We judged that its inclusion was not practicable due to it being conceptually distinct. This is because, unlike selection, purchase and consumption of food, alcohol and tobacco, physical activity does not necessarily relate to products that are separable from and placed within a given environment (see Stage 3 in the Methods for further details).
While the current behavioural focus of TIPPME limits its generalizability, the typology is intended to be broad in scope so that it could potentially be adapted to apply to other behaviours. When considering the current and potential future scope of TIPPME in terms of the behavioural domains to which it applies, it may be helpful to map its categories to a systematically-developed framework of behaviours, such as Nudelman and Shiloh’s taxonomy of health behaviours36. Within this taxonomy, TIPPME is currently aligned to nutrition and risk avoidance behaviours, these being related to consumption of products that are linked to non-communicable disease. In theory, TIPPME may be applicable to a wide range of other behaviours, including those unrelated to product consumption (for example, physical activity or gambling), those linked to the prevention of communicable disease (for example, hygiene-related behaviours) and pro-environmental behaviours to mitigate climate change (for example, energy use or recycling). Such translation will require specific programmes of development and testing.
In the process of developing TIPPME, some challenging conceptual issues were encountered. Most notably, to ensure TIPPME had the potential to discriminate, it was intended that each intervention type (typology row) would represent a distinct way in which the proximal physical micro-environment can be altered, with any single discrete intervention component being assignable to a single intervention type. However, it was quickly apparent that there was an inevitable degree of conceptual overlap or co-dependence between the different intervention categories. For example, if we consider an intervention in which restaurant patrons are provided with smaller (versus larger) spoons to attempt to reduce dessert consumption, we would expect most users to characterize this as a ‘size’ intervention. However, manipulating the size of a spoon will probably also affect the way the spoon looks and feels, that is, ‘presentation’, and so either or both types could feasibly be applied for all ‘size’ interventions. While this is a fundamental conceptual issue linked to the nature of the sensory, spatial and morphological attributes of objects and stimuli, for the purposes of the typology it can be addressed by distinguishing between the primary target of the intervention and secondary consequences. As explained in the typology and its accompanying instructions (Supplementary Information), assuming a single discrete intervention component or manipulation is present, the user applying TIPPME will aim to identify a primary intervention type that best captures it. Should there be multiple discrete, separable intervention components implemented in the same environment, multiple different intervention types can correspondingly be applied. In practice, reports of interventions will often support judgements of what the primary target of the intervention is via their stated aims and hypotheses, and the way in which they describe intervention content. Importantly, results of the reliability testing exercises suggest that despite these potential challenges, the typology can be consistently applied by users.
Because reliable application of TIPPME, as with any classification system, is dependent on making informed judgements from the information that is provided, the increased attention that is being given to improving the reporting of intervention studies37,38 will be beneficial. In addition, we would hope that TIPPME will enable those reporting on proximal physical micro-environment interventions to specify the primary target of their intervention, thereby reducing future ambiguities.
TIPPME, provides a framework to reliably classify and describe an important class of interventions, and enable more systematic design, reporting and analysis of interventions to change health-related behaviour at the population level. In doing so, we propose that TIPPME makes a distinct contribution to collective efforts to build the cumulative evidence base for effective ways of changing behaviour across populations.
The development of TIPPME is summarized in Table 2 and involved three phases of work: (1) identifying the need for a typology; (2) developing and elaborating on this typology; and (3) reliability testing and finalizing. These were completed in seven main stages. Throughout, the development process was integrated with formal and informal discussions between the core research team as well as wider academic networks. The core research team (the authors) comprised ten members with a range of disciplinary backgrounds across public health, health policy, psychology and behavioural science, sociology, and evidence synthesis, and many of the group had extensive policy and guidance development experience. They represented varied expertise in developing and applying prominent typologies or classification systems relating to behaviour change interventions and theory, developing, implementing and evaluating public health and behaviour change interventions in a range of behavioural and population contexts, and developing practice and research reporting guidelines.
To generate a refined definition and provisional typology of choice architecture interventions in physical micro-environments, and to map the available empirical evidence for the effects of these interventions on diet, physical activity, and alcohol and tobacco use.
We conducted a large-scale systematic scoping review23, published in 2013. In brief, the methods used to develop the provisional typology involved highly sensitive searches of 15 electronic literature databases, combined with parallel snowball searches, retrieving over 800,000 unique title and abstract records. We used text mining methods to prioritize these records for screening39 and manually screened over 54,000 prioritized records to identify 346 eligible full-text articles.
Data extracted from these 346 articles (reporting primary evaluation studies and reviews of such studies) were then used to configure, describe and synthesize the key characteristics of interventions. This was an iterative process, incorporating regular discussion among members of the review team, and resulted in the provisional typology in Table 3, comprising nine types of interventions: ambience, functional design, labelling, presentation, sizing, availability, proximity, priming and prompting.
The research team sought feedback on the typology from a wider group of potential users concerning its usefulness and the scope for further development, including discussion of conceptual issues that had been identified by the research team over the course of conducting the scoping review.
To elicit expert feedback about issues with understanding and using the provisional typology of choice architecture interventions.
Two typology development workshops were conducted, attended by a total of 45 participants working in areas of behaviour change and public health intervention. Participants were predominantly in research roles (80%, with 20% in policy or practitioner roles), with a range of disciplinary backgrounds represented (psychology or behavioural science (38%); public health or medicine (20%); nutrition (7%), policy (11%) sociology (2%), other or missing (20%)). To encourage engagement with the provisional typology and elicit feedback, the workshops involved completing the same brief intervention description coding task (classifying 14 intervention descriptions by reference to the provisional typology) followed by a questionnaire assessing perceived value of the work and its development, concluding with a structured group discussion.
There was strong support expressed for the value of developing the provisional typology from academics and practitioners who attended: 95% (41/43 responses) of participants agreed with the statement ‘developing this typology is valuable and important’. There was recurrent feedback that further development would be necessary for the typology to be more widely applied. Issues highlighted that would need to be considered in future included identifying some inconsistencies in the intervention types concerning whether they related to the content or the mechanism of the intervention (for example, one of the intervention types, ‘priming’, related primarily to a mechanism of effect and was therefore not equivalent to other intervention types); whether the ‘labelling’ and ‘prompting’ intervention types were clearly distinct; and the difficulty of coding physical activity interventions. The intervention description coding task was principally intended to encourage engagement and general feedback within the workshops. However, in line with qualitative feedback received from participants, its results suggested that the typology categories could be applied consistently (Fleiss’ kappa, 0.83), thereby supporting use of the provisional typology’s basic structure and content as a foundation for future development.
In preparation for future development of the typology, the initial research team was extended to include two behaviour change experts with expertise in developing and applying prominent typologies of behaviour change interventions and theory, and a public health expert with expertise in environmental determinants of health and the evaluation of public health interventions. It was agreed by the research team that development of a new typology would be valuable, informed by both the workshops and other external indicators of potential value. For example, the provisional typology had been used to frame two calls from a national research funding body (National Institute for Health Research (United Kingdom)), had directly informed United Kingdom policy documents and guidelines from bodies such as the Department of Health40 and the National Institute for Health and Care Excellence41, and was being used for characterizing and defining interventions in a range of systematic reviews (for example, refs 33,34,35,42,43).
To generate by consensus a preliminary version of TIPPME that would be subject to further development.
A two-day residential meeting of the core research team (the authors) was held. This meeting comprised a series of structured discussions, informed by feedback received from the expert workshops (Stage 2) and from colleagues in wider networks.
A preliminary version of TIPPME was generated. Principal developments agreed at this stage, representing changes to the provisional typology (Table 3), were as follows:
A new matrix classification structure was created to enable representation of both different intervention types (the rows of the typology), as well as differences in the spatial focus of the intervention (the columns of the typology). This structural change stemmed from agreement that a more conceptually coherent position would be to view any given physical micro-environment as a set of objects or stimuli that could all feasibly be manipulated. This would also allow greater flexibility in thinking about the range of possible interventions within this space, whether these are only theoretically possible or are represented in the current body of empirical literature.
The intervention type ‘ambience’ was removed and its place was taken by the ‘presentation’ intervention type applied on the scale of the wider environment. This was because the new typology structure meant that objects and stimuli in the wider environment were considered subject to the same intervention types as the products themselves.
The intervention types ‘labelling’ and ‘prompting’ were subsumed within a generic intervention type pertaining to the communication of explicit textual, numeric or pictorial information. This more inclusive category of information-based interventions—initially named ‘words, numbers and pictures’ and ultimately ‘information’—was considered more coherent, as previous conceptual distinctions between ‘labelling’ and ‘prompting’ interventions were unclear. This still allowed differentiation from other intervention types that focus on the alteration of sensory, spatial and morphological characteristics.
The intervention type ‘priming’ was removed as there was agreement that this represented a specific mechanism rather than an equivalent intervention type.
Notably, physical activity was excluded as a behaviour of interest. It was agreed that it was not practicable to include this in a coherent and concise typology, given that, unlike food, alcohol and tobacco, it does not involve the selection, purchase and consumption of products that are separable from and placed within a given environment. It would be possible to adapt the typology to physical activity, with the equivalent of the target product or object being the physical space in which, or on which, the physical activity is performed. This physical space may be a permanent part of, or the whole of, the proximal physical micro-environment itself. However, adapting the typology to physical activity would be complex and require its own specific explanation and translation.
To use the preliminary typology to identify outstanding conceptual and practical problems with TIPPME.
The research team (n = 8, excluding the first two listed authors who were responsible for producing the exercise materials), completed a task which encouraged engagement with the detail of the preliminary typology and its application. Each participant was given a link to an online Qualtrics task comprising 40 short (<150 words) intervention descriptions. These represented a sample of descriptions of interventions from the 346 papers that were included in the aforementioned scoping review of choice architecture interventions23. We selected intervention descriptions on the basis of a quota that covered a wide range of intervention content, aiming to include at least five examples that could feasibly be mapped to each of the six intervention types in the typology, with a spread across the three intervention foci and across food, alcohol and tobacco. We used the first example meeting our criteria that was encountered via random searching to ensure that the intervention descriptions were varied in nature and broadly representative of the wider empirical literature. The 40 intervention descriptions were presented to each participant in random order. For each intervention example, participants were asked to assess which category in the typology best captured the example, how much overlap existed between the intervention types they considered selecting, and to describe any difficulties they encountered in coding the example and any possible alterations to the typology that would have ameliorated these difficulties. For each intervention example, they were encouraged to provide further qualitative feedback concerning each intervention example, and the overall structure and content of the typology.
Quantitative and qualitative responses from the task were synthesized. While quantitative results suggested that intervention types could be applied consistently (Fleiss’ kappa, 0.69), the task was principally intended to highlight areas in which there were significant levels of disagreement, comment or criticism, to prioritize focused discussion.
A teleconference of the research team was convened, and a structured discussion was conducted. This involved assessing problematic intervention descriptions in a structured format, in order to reach agreement on steps to be taken to improve the conceptual coherence and ease of use of the typology. Intervention examples where ≥50% of responses were discordant were flagged for prioritized discussion. Discussion began in order of the intervention descriptions that were coded least consistently, and terminated after all flagged examples had been discussed. For each flagged example, individuals were encouraged to justify their responses and propose and discuss solutions which could overcome the reasons why individuals coded discrepantly. A vote then took place to assess if participants could converge on the same answer (1) without any further changes to the typology and (2) with specific changes to the typology (if agreed upon). As a result of this process, various changes were made to the wording of definitions, including a clarification of the distinction between the columns in the typology. Additional guidance text was added where it was agreed that there was a greater chance of perceived overlap between intervention types.
Reliability testing exercise 1 aimed to test whether participants likely to use TIPPME, namely external experts involved in researching or implementing interventions to change health-related behaviours, were consistent in identifying its intervention types and foci in short descriptions of interventions.
We recruited external experts with backgrounds in public health and behavioural science as researchers or practitioners, meeting the following criteria (adapted from ref. 15): ‘active in their field and engaged in designing, delivering and/or evaluating interventions to change health-related behaviour that could be delivered at scale to impact on population health’. Recruitment was via email and Twitter enquiries to possible participants in our wider academic networks. A similar exercise to that described in Stage 4 was used, involving coding 40 short intervention descriptions (<150 words) selected on a quota basis to represent a range of intervention content and targeted products. These were presented in a random order, using the question ‘Which intervention category best captures the above description?’. Quantitative reliability statistics were calculated for the pre-specified primary outcome of discrimination of intervention type (the rows of the typology), as well as for intervention focus (the columns of the typology) and a combined total. As some agreement would be achieved by chance alone, two different ‘chance-corrected’ agreement measures were used, Fleiss’ kappa and prevalence and bias adjusted kappa (PABAK)44,45.
Potential participants (n = 52) who had initially expressed an interest in participating were contacted via email with a link to the exercise. All participants (n = 33) who started the exercise completed it. Of these, 58% were female, and most were in research roles (94%, with 6% in policy or practitioner roles), with a range of disciplinary backgrounds represented (psychology or behavioural science, 45%; public health or medicine, 18%; nutrition, 15%; epidemiology, 6%; economics, 6%; other (urban planning, marketing and human factors), 9%). Inter-rater reliability values for the exercise are provided in Table 4. A kappa value of 0.41–0.60 is conventionally considered to represent ‘moderate’ agreement, a value of 0.61–0.80, ‘substantial’ agreement, and a value of >0.80, ‘almost perfect’ agreement46. Fleiss classified a kappa between 0.400–0.075 as ‘intermediate to good’ and >0.75 as ‘excellent’45. Therefore, the observed values indicate that the intervention types in the typology in its current form are strongly distinguishable from one another. Furthermore, category-wise statistics indicate that each of the six intervention types and each of the three intervention foci could be applied reliably.
A two-day residential meeting of the research team was held to further refine the typology. Although the structure of the typology was not altered, each intervention type was given a concise single-word title (for example, ‘words, numbers and pictures’ was changed to ‘information’), and some minor changes to the wording of definitions were made. It was agreed that, because intervention descriptions used in this exercise comprised short passages focused on the intervention characteristics, this was not representative of how these might more typically be encountered in full-text papers, where details may be spread disparately within papers, in potentially complex formats. Furthermore, it was important that participants were able to code the presence of multiple discrete intervention types identified within one paper, where previously they had been asked to identify a single category that best captured an intervention. Therefore, a second reliability testing exercise intended to be less artificial and more generalizable to real-world use was planned.
Reliability testing exercise 2 aimed to assess whether interventions described in full-text papers can be reliably coded to categories in TIPPME.
Following a pilot phase to develop the exercise, members of the research team (n = 4) each coded 24 study reports within full-text articles that were randomly selected (using a random number generator) from those included in the initial scoping review23 but excluding review papers and those not describing any intervention, those concerning physical activity interventions, and those that had been used in previous stages of the typology development process. To ensure that a range of behaviours was covered, randomization was stratified by behaviour so that half of the papers related to food and half to alcohol or tobacco. Where there were multiple eligible separate studies within a single full-text paper, the first was used. The exercise involved each participant coding which (if any) categories in the typology were identified in each study. The number of study reports to be coded in this exercise was derived using the KappaSize R package47, given that, to our knowledge, there are no gold-standard methods to precisely estimate the required sample sizes for determining reliability kappas in cases where there are both multiple coders and a large number of coding categories. We estimated an approximate, conservative sample size on the basis of the following parameters: an alpha value of 0.05; power of 0.80, using 4 coders; an assumption that categories will not be perfectly balanced and instead may be moderately unbalanced; a null hypothesis of a kappa of 0.4 (that is, the lower bound of ‘intermediate to good’ agreement on the Fleiss’ Kappa Benchmark Scale); and, an expected kappa of 0.7 (based on observed kappa values from reliability testing exercise 1). This suggested that at least 22 study reports would be required to test whether the kappa exceeds 0.4.
Inter-rater reliability values are provided in Table 4. The observed values indicate that the intervention types in the typology in its current form are strongly distinguishable from one another when full-text papers are coded. Furthermore, category-wise statistics indicate that underlying the summary statistics, each of the six intervention types and each of the three intervention foci could be applied reliably.
To reach consensus on a final version of TIPPME including terminology, wording and presentation.
Research team members completed a final check of the typology and its wording to ensure it was clear and consistent throughout. A teleconference of the research team was convened to discuss any identified issues.
Further descriptive notes and additional examples were added to the full version of the typology to aid in its use. A final version of TIPPME was agreed on by the research team; described in the Results section.
The data that support the findings of this study are available from the corresponding author upon request.
How to cite this article: Hollands, G. J. et al. The TIPPME intervention typology for changing environments to change behaviour. Nat. Hum. Behav. 1, 0140 (2017).
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The study was funded by the United Kingdom Department of Health Policy Research Programme (Policy Research Unit in Behaviour and Health (PR-UN-0409-10109)). D.O. is supported by the Medical Research Council (unit programme number MC_UU_12015/6). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Supplementary Notes, Supplementary Figure 1.