Defining choice architecture interventions

Prior work in this area has not included a clear definition of choice architecture interventions applicable to public health. In ‘Nudge: Improving decisions about health, wealth and happiness’ [3], Thaler and Sunstein define a ‘nudge’ as “any aspect of the choice architecture that alters people’s behaviour in a predictable way without forbidding any options or significantly changing their economic incentives”. In terms of operationalising this definition, they outline a set of mechanisms by which choice architecture can be altered to change behaviour, such as via incentives and defaults, but there are no precise operational definitions of what these terms mean in an applied sense [2]. Such definitions are instead largely implicit in the illustrative examples chosen, and apply to specific behavioural and environmental contexts that may have limited public health relevance. For example, possibilities for altering defaults are presented largely by reference to opt-in/opt-out systems such as for organ donation or pension schemes – examples not obviously applicable to key sets of health behaviours related to diet, physical activity, alcohol and tobacco use. Continued policy interest worldwide has stimulated efforts to clarify the concept, but to date this has been reflected only in non-systematic attempts to group intervention approaches that have common mechanisms and highlight examples of these approaches that may be scalable to population level [2, 4, 7, 17, 18]. A small number of recent studies in public health have examined choice architecture interventions as applied to health behaviours and conceptualised these as changes to environments to alter the ways in which choices are presented, taking into account that people’s choices are not always rational [19, 20]. However, such examples, in this case being respectively a primary research study in a cafeteria environment [19] and a systematic review of interventions in self-service eating settings [20], focus on highly specific intervention contexts and so contain little examination of the broader conceptualisation of choice architecture interventions for changing health behaviour that we suggest is needed. The fact that terminology continues to be used inconsistently [6] emphasises the need for a more systematic approach to conceptual development.

Towards a definition and typology of choice architecture interventions

We consider each of these aims in the Discussion section. We developed, tested and refined our analysis through a systematic, iterative and configurative scoping process, a short summary of which is provided in Additional file 1. A detailed description of the methods used for the review is provided in the full report (Additional file 2) [21] and in a separate open-access methods paper [22].

Defining choice architecture interventions in micro-environments for changing health-related behaviour

We propose the following operational definition:

Interventions that involve altering the properties or placement of objects or stimuli within micro-environments with the intention of changing health-related behaviour. Such interventions are implemented within the same micro-environment as that in which the target behaviour is performed, typically require minimal conscious engagement, can in principle influence the behaviour of many people simultaneously, and are not targeted or tailored to specific individuals.

This reflects our focus on physical and social dimensions of micro-environments. In practice we did not encounter empirical studies of interventions that involve altering social dimensions of micro-environments (such as those centred on changing social norms [23]) that met our definition. In line with prior formulations of choice architecture [3], we excluded interventions that involved the use of economic instruments (e.g. taxes, subsidies), unless these also included other intervention components that met our operational definition. We conducted a concurrent scoping review of the effects of economic instruments on diet and physical activity, reported elsewhere [24]. Use of the term ‘typically require minimal conscious engagement’ reflects our view (consistent with that of others e.g. [20]) that interventions that alter micro-environments often display this characteristic [13, 25], while also recognising the potential for varying degrees of conscious engagement with such interventions.

Provisional typology of choice architecture interventions

We grouped the available evidence that was consistent with our definition into nine types, each comprising a range of interventions that share similar characteristics or proposed mechanisms of action (Figure 1, left side). These nine intervention types can be aggregated into two higher-level classes of intervention: i. those that involve altering the properties of objects or stimuli within a micro-environment, and ii. those that involve altering their placement (with some interventions involving both). These two broad classes of intervention map well onto one of the more detailed models describing how environmental stimuli elicit behavioural responses outside of awareness [26].

Mapping the available empirical evidence

This conceptual groundwork enabled the systematic identification and preliminary mapping of a large body of empirical evidence for the effects of choice architecture interventions on diet, physical activity, alcohol and tobacco use. The right side of Figure 1 shows the numbers of study reports we identified that met our operational definition, disaggregated by intervention type and target behaviour. Our evaluation of the text mining technologies used to expedite study selection (published elsewhere [22]) engenders confidence that the distribution of studies we assembled is likely to be largely representative of the full spectrum of available evidence for target interventions.

This evidence base is dominated by studies of interventions to change diet-related behaviours such as food purchasing and consumption (70.2% of study reports), with many fewer studies relating to physical activity (19.1%), alcohol (7.3%) or tobacco use (3.4%). The two types of intervention most frequently encountered, together accounting for over 40% of study reports, were those involving point-of-choice labelling, and prompting. These interventions generally involve providing information — about the nutritional content of food, for example, or the health benefits of climbing stairs — similar to that communicated in more conventional approaches to health promotion. Other types of intervention closer to the spirit of the formulated definition of choice architecture, i.e. less reliant on the conscious engagement of the individual, are less well represented in the literature. Further description of the nature of the studies identified for each intervention type is provided in Additional files 2 and 3.

Implications and next steps for intervention development and evaluation

We identified a considerable volume and range of evidence for choice architecture interventions and their potential effects on four key health behaviours. We also found some notable gaps in the evidence base, including a lack of high-quality systematic reviews of the effects of interventions contained within the parameters of our typology (with exceptions, such as interventions to prompt stair use [27]). Most of the research evidence we identified focuses on the effects of interventions on behaviours related to diet. This imbalance in the distribution of evidence between the four sets of behaviours considered is perhaps surprising, given that most of the intervention types that comprise our typology could in principle be applied to alcohol and tobacco use, which (like eating) necessarily involve the consumption of products. The relative scarcity of evidence for impacts on physical activity may in part reflect the fact that some intervention types (such as product sizing) are less applicable to physical activity. We did, however, locate studies of interventions to promote physical activity in six of the nine intervention types identified. This highlights the potential to take interventions that have been applied in one behavioural domain and develop and test them in others. For example, interventions to alter the portion or package size of foods have been widely developed, but we found less evidence for similar interventions to alter the use of alcohol or tobacco products. This may be explained in part by the fact that diet-related behaviours provide a more diverse range of opportunities for intervention spanning a much larger range of products and environments, relative to alcohol or tobacco.

Limitations of the evidence base support our previous assertion that evidence to support altering choice architecture as a population health strategy is currently weak [2], but it is premature to conclude whether or not choice architecture and other potentially non-regulatory interventions are likely to be effective [6]. Formal critical appraisal and synthesis to estimate the direction (health-enhancing or not) and magnitude of effects reported in included studies was beyond the scope of this preliminary work. Further rigorous primary research and systematic reviews, conducted within a conceptually coherent framework, will be necessary to produce reliable assessments of the likely direction and magnitude of intervention effects, and of factors likely to moderate those effects.

This work provides a foundation for understanding the effects of a broad array of interventions to change health behaviour by altering micro-environments. The empirically grounded definition and typology of interventions are intended to inform and frame the scope of further investigations, as well as providing a platform for further conceptual development work. The definition and typology are both provisional and the typology is primarily descriptive. Additional development will be necessary before they can be applied to inform specific aspects of the design and evaluation of such interventions. In particular, ongoing work will need to compare and assimilate relevant theoretical and conceptual accounts of behaviour change interventions and processes with our provisional definition and typology. Such development would benefit from consideration of possible mechanisms of action, including the role of non-conscious processes in health-related behaviour [28, 29]. It will also be necessary to monitor the emergence of new interventions proposed to alter choice architecture to change health behaviour. This will enable assessment of the extent to which their characteristics and proposed mechanisms of action are consistent with, or suggest further refinement of, the definition and typology.

Both the definition and typology were generated through systematic analysis of evidence for interventions that have not previously been evaluated within the specific context of this review. As such, they may not map neatly on to prior conceptualisations of choice architecture or nudging that have been proposed in the wider literature, encompassing various other fields of policy and practice. Despite challenges in translating the details of prior formulations of choice architecture to the different focus of this work, our definition is broadly consistent with their key principles. For example, Ploug and colleagues [8] characterise institutional strategies to shape behaviour based on choice architecture principles as: (1) imposition of trivial costs on those departing from welfare-promoting options, (2) the framing of options i.e. their presentation, and (3) setting of institutional default rules. Our definition and typology encompasses instances of (1), such as interventions altering proximity (e.g. placing less healthy foods further away); (2), such as nutritional labelling interventions; and (3), such as altering default portion sizes.

We found few intervention studies that explicitly linked proposed mechanisms of behaviour change with wider notions of ‘choice architecture’ or ‘nudging’. This lack of overlap between terms currently used in policy and research circles may reflect a time lag between these ideas becoming popular and studies of related interventions being completed and published. It may also signal a lack of conceptual clarity or agreed-upon terminology between different communities and research disciplines. The definition and typology we have presented here can therefore contribute to more fruitful translation between research and policy.

To improve population health and reduce health inequalities, we need to know not only the short-term effects of behaviour change interventions, but whether their effects — both singly and in combination— can be sustained and how these effects are distributed within and between social groups. Whilst we did not systematically extract data on this, our impression is that few studies have reported the long-term durability of behavioural effects of interventions. As many of these interventions rely on brief exposures in time and place, one-off or infrequent exposures to these interventions would not necessarily be expected to have enduring effects on behaviour. It is therefore important to investigate the cumulative effects of repeated exposure over time. This will require alteration of aspects of micro-environments within a long-term evaluative framework. As the evidence base develops, it will also be important to explore issues of implementation, such as public acceptability and financial impacts of interventions. Furthermore, whilst governments may be liable to present these interventions as non-regulatory approaches, it does not necessarily follow that regulatory or legislative frameworks will not be needed to ensure or enhance their implementation [2, 30].