Chapter 3 Frameworks

3.3.1 The outsider’s point of view

Figure 3.4: Man undergoing an EEG

From the outside, cognitive science carries the picture of popular science: white lab coats, strange equipment and experiments in a bland, medical setting. Unfortunately, some research and behavioural insights teams play, perhaps unwillingly, on this representation. For example, here is the header of a post from the Whitman Behavioral Lab.

Pause: What kind of representations of human cognition does it carries? In other words, what vision of our work does it imply?

Figure 3.5: Banner of the Whitman Behavioral Lab.

• Firstly, I think we can safely say that the picture is extremely cluttered. There are close to 100 symbols, most of which do not carry a clear, explicit relation with human cognition.
• Some symbols refer to natural sciences: particle physics, chemistry, a wandering infinity symbol.
• Most other symbols refer to engineering, plumbing, sprinkled with electrical wires.
• The only distinctively human symbol is the big, black, disembodied brain outline. This brain has no organs to feel the world with, no body to interact with its environment. It is also alone, missing the social dimension of human cognition.2323 I admit I am often guilty of using the word “brain” as a shortcut for “human cognition”. I do not claim this should be always avoided, but we must be conscious of what it may imply in terms of representations.

Of course, this is a single example. You’ll observe however that a similar set of symbols is often present in logos and illustrations of behavioural insights. It may be useful in projecting the idea that we’re doing rigorous science, but on the other end, it appears far removed from the daily experience of people and civil servants you’re supposed to help. From your point of view, frameworks are a way to bridge that perception gap.

3.3.2 Corporate and civil service framework culture: expectations

For most people in corporate of civil service, a framework is a conceptual organisation of the way to define processes. Let us unpack these elements. Just as Taylorism defined precise, rigid processes for industrial activities, the influence of the New Public Management has led to a replacement of broad rules and civil servants’ discretion by a growing set of rules and processes. Civil servants thus follow an increasing number of (sometimes conflicting) processes. These processes are theoretically informed by frameworks. One example, again coming from industry, is Lean management, articulated around the dual objective to maximise value and minimise waste — hence the “lean” part — in terms of materials, time and skills.

Figure 3.6: Lean management flow

This is thus the kind of things your counterparts will have in mind and expect when asking about a framework: a kind of map of the steps you’ll follow in designing and making your intervention. At the least, they’ll expect it to make your intervention understandable and predictable to them. Often, the requirement of a framework will come will an expectation of scalability. They’ll expect that by following the same framework, they’ll be able to scale up the intervention, transfer it to other contexts or design their own interventions.2424 Managing this kind of expectations in a key skill for a behavioural project manager. You need to be clear on what is achievable in terms of upskilling through a given intervention.

3.5.1 The Cognitive bias codex

The need for some organisation device is evident in the face of the number of documented biases, as represented in for example in the cognitive bias codex. We can assume each of these biases to be relevant, since they cleared inclusion criteria for the English language version of Wikipedia. The sheer number of them is not cognitively manageable, even when excluding duplicates. Hence the need to form higher-level categories linked with more general behaviours (e.g. We tend to find stories and patterns even when looking at sparse data) and general constraints faced by the brain. While the categories in the codex have been partly constructed using an algorithm, building such a classification is a behavioural science exercises in its own right, and entails an (even implicit) theory of human behaviour.

Figure 3.7: An algotritmically -assisted classification of behaviours labelled as biases in Wikipedia, Jm3, CC BY-SA 4.0

Did you notice the disembodied brain?

3.5.2 Theories of human cognition

Theories of human cognition, such a these we saw previously — Prospect theory, System 1 / System 2, Self-determination theories — can be used as frameworks, both to define and to make sense of the collection of behavioural features considered for inclusion in a given intervention. More often then not, however, a given situation will entail features coming from several theories. This has led practitioners to define empirically-driven frameworks of organisation.

Figure 3.8: Graphical summary of the System 1 / System 2 features.

This time, there are two of them.

3.5.3 The MINDSPACE

The MINDSPACE framework, first presented in (⊕Dolan et al. 2010Dolan, Paul, Michael Hallsworth, David Halpern, Dominic King, and Ivo Vlaev. 2010. MINDSPACE. The Behavioural Insights Team. https://www.bi.team/wp-content/uploads/2015/07/MINDSPACE.pdf.) is perhaps the first influential example of this kind, informing most of the Behavioural Insights Team early interventions (2009-2014). Rather than building a higher-level theory of human cognition, it proposes a practical gathering of behaviours into categories ready to be used to analyse the problem (behavioural diagnosis) and to design a solution:2525 As you imagine, the need to have a catchy acronym has probably influenced the number and the labeling of the categories.

Figure 3.9: Graphical illustration of the MINDSPACE framework.

• Messenger: We are heavily influenced by who is communicating information.
• Incentives: Our responses to incentives are shaped by predictable mental shortcuts, such as the strong desire to avoid losses.
• Norms: We are heavily influenced by what others do.
• Defaults: We “go with the flow” of pre-set options.
• Salience: Our attention is drawn to novel things that seem relevant to us.
• Priming: Our actions are often influenced by subconscious cues.
• Affect: Our actions can be powerfully shaped by our emotional associations.
• Commitments: We seek to be consistent with our public promises and to reciprocate actions.
• Ego: We act in ways that make us feel better about ourselves.

And yet another disembodied brain.

Its self-declared aim is

“[to] condense available evidence at the time into a manageable checklist, to ensure policy-makers take into account the most robust effects on our behaviour.”

This quotation underlines the context of this first framework: there were few qualified experts. As a consequence, it was designed to help non-expert civil servants in including behavioural features in the policy-making, and provided simple rules to ensure minimal attention was devoted to these features. In hindsight, I’d argue it had two drawbacks. The first one was that the list was already too long, leading to some confusion on the best levers to use, and, in a checklist approach, the suggestion that you had to use all of them. The second one would be that it fostered the idea you did not really need someone with behavioural training in the room, and just go by checklist. In my practice, I saw several instances of this (dangerous) idea through the requirement that people we briefly train during and intervention should be able to train their colleagues without additional external support.

Figure 3.10: EAST framework logo

3.5.4 Going EAST

The EAST framework represents the second iteration, introduced by (⊕Service et al. 2014Service, Owain, Michael Hallsworth, David Halpern, et al. 2014. EAST: Four Simple Ways to Apply Behavioural Insights. The Behavioural Insights Team. https://www.bi.team/publications/east-four-simple-ways-to-apply-behavioural-insights/.). Where the MINDSPACE framework was effectively a list of observational lenses (look at the behaviour through a shortlist of common heuristics), the EAST framework puts the onus on an operational approach: how should I design my intervention.

The technical terms composing the MINDSPACE acronym are replaced by four top-level categories of common terms, with descriptions in plain language. Technical terms are pushed to the bottom of the page. If you look at the document itself, you’ll see that the Executive summary has also been crafted to be as accessible as possible to non-specialists, providing a simple description of the phenomena, e.g. “Harness the power of defaults. We have a strong tendency to go with the default or pre-set option, since it is easy to do so. Making an option the default makes it more likely to be adopted”.

Figure 3.11: EAST framework table

The components of this framework are:2626 Why this list? Because the table is an image, which is not accessible to screen readers..

• Easy: Allowing people to “go with the flow” by removing or reducing effort, steps, choices to make action simple and effortless. Relies e.g. on Endowment effect, Status quo bias, Cognitive overload avoidance

• Attractive: Presenting benefits in a way that maximizes perceived value. This include increasing the salience of you offer. Relies e.g. Availability bias, Anchoring, Loss aversion, Optimism bias, Scarcity.

• Social Harnessing social/peer “pressure” by showing desired behaviours are supported by others in a social group and encouraging shared commitments. Relies e.g. on Confirmation bias, Herding, Commitment bias, Authority bias.

• Timely Prompting when people are likely to be most receptive, and structuring/phrasing benefits to make them more immediate. Relies e.g. on Present bias, Hyperbolic discounting, Duration neglect, Hot/cold states.

This framework is clearly a success. It is extremely lightweight, and enables people with little training in behavioural sciences to include behavioural elements in their action. As behavioural scientists yourselves, you should also be aware of its shortcoming: through availability, it nudges you towards a limited set of heuristics, at the expense of others which can be more relevant to the situation at hand. It is also oriented towards the design stage of an intervention, with the temptation to jump to building a solution before having carefully diagnosed the problem, including its structural dimensions or the perception of the stakeholders.

3.5.5 Some EAST application examples

Part of the success of the EAST framework does stem from the fact that we evolve in an environment where a superficial analysis of the behavioural obstacles may be enough to spot large margins of improvements.

Choosing your energy provider An example is a program run in 2011 by the UK government. It started by the observation that few people switch energy providers, even if the process itself is simple, and that they could make significant savings.

Figure 3.12: Easy: a QR code to compare energing provider offers.

The quick analysis spotted the fact that assessing these savings was complicated: tiered systems and peak/off-peak prices made it difficult to know how much you would pay. To help people choose, paper energy bills were endowed with QR codes containing a link to detailed consumption information and to a comparison engine which provided offers from providers based on the actual consumption.

Basically, the intervention directly lifted a recommendation from (⊕Thaler and Sunstein 2008Thaler, Richard H., and Cass R. Sunstein. 2008. Nudge: Improving Decisions about Health, Wealth, and Happiness. Yale University Press.): mandate the energy providers to provide a harmonized, machine-readable, version of consumption data and of their prices, and feed that to a comparison algorithm. The key word here is of course “mandated”: making things simpler to consumers often require imposing significant regulatory pressure on firms (or administrations), which are incurring a direct cost (and, for non-competitive providers, loss on revenue).

In terms of structural analysis, you’ll notice that some market structures may be more conductive to this kind of intervention than others. In France, the smart electricity meter, Linky, is owned and operated by ENEDIS, and provides an API open to all energy providers.

The 2022-2023 energy price crisis also showed a limit of this approach: to be as easy as possible, the price comparison only includes consumption from the previous period. As a result of the absence of any forward-looking element, people have switched to energy providers whose prices rose dramatically following the onset of the war in Ukraine and subsequent disruption of the wholesale energy markets. This is a generic issue of make-it easy intervention: by construction, you simplify things by subtracting elements of choice you deem less relevant at a given point of time. Evolution of the context may make very relevant elements you discarded earlier.

Figure 3.13: Photos from the INudgeyou hand sinitizer experiment.

Clean your hands This said, there is little reason to underestimate the success of this approach. A 2016 experiment in Denmark described in (iNudgeyou 2016) illustrates both the high return on investment of simple interventions combining two aspects to the framework (here, Easy and Attractive). Moving the hand sanitizer dispenser in front of the main doors increased visitors’ compliance from 3% to 20%, and adding a big, red sign on top of it, which signaled both availability and norm, had compliance jump to 67% — 22 times the baseline level of compliance.

3.6.1 Is it a behavioural problem?

The first part of BI work is to select the right projects, that is those which feature a behavioural issue – and not try to patch structural problems. As a starting point, I suggest using the flowchart proposed by (⊕Hansen 2019Hansen, Pelle Guldborg. 2019. “The Concepts of Nudge and Nudging in Behavioural Public Policy.” Handbook of Behavioural Change and Public Policy, May 31, 63–77. https://www.elgaronline.com/view/edcoll/9781785367847/9781785367847.00010.xml.):

1. Do people have information available to guide them?
   • No: This is a knowledge problem. Information and education
2. Would rational agents have the capacity and opportunity to enact the behaviour?
   • No: Provide training, change environment
3. Do people have the internal motivation to enact the behaviour?
   • No: Explain why it would be the right thing to do
4. Are external incentives aligned with the desired behaviour?
   • No: Create or align incentives
5. Are regulatory measures in place to support the behaviour?
   • No: Revise existing regulation
6. It is probably a behavioural problem

This flowchart is completely aligned with the epistemological idea that we deal with what differs between a rational agent’s behaviour and observed actual behaviour. As a result is probably a bit strict of what a BI problem is relative to current practices, but a useful guide to explain what you do and what you don’t.

In most actual projects, you won’t have all the necessary information to answer these questions. Let us take an example: why do people exercise too little? The problem is widely under-specified (as most policy problems are). What is the general level of recommended physical activity for the general public? Guidance exists, but the evidence base is surprisingly flimsy. The problem is different if you’re dealing with young people, or with older people with physical or cognitive impairments (and there you may have more reliable guidance), and so on.

As a result, you’ll often have to argue that some preliminary research is necessary to assess whether of not the behavioural dimension of a policy issue is relevant and amenable to action. Having a clear set of rules and information requirements will greatly help you to explain why time and money is needed at that stage. It makes it clear that an early exit (This is not a behavioural problem) is an eventuality.

3.6.2 Explaining your work in a flowchart

In essence, most if not all behavioural intervention frameworks express the same ideas. The differences are a matter of accent.

Figure 3.14: Flowchart of the British Columbian BI Group.

Figure 3.15: French DITP flowchart.

The British Columbia BI Group example takes a very linear view. It is more comfortable for policy makers worrying about long delays and never-ending feedback loops. Here, the EAST framework guides the I part of the project. The design team choose to give data-collection its own top-level box, probably because this is often expensive and should be planned and partly executed before the intervention to provide a reliable baseline. By contrast, the DITP representation insists on the often circular nature of the process. More often than not, your intervention will lead to disappointing or fragile results. You’ll need intermediate scale-ups to confirm or not your effect. Over the course of an intervention, you also learn a lot of things you’ll wish you’d known in the first place. You end up suggesting something different from what you tested. This representation thus conveys the idea that you’ll probably need several rounds of iteration before you get something that can reliably work at scale.

3.8.1 The OECD-OPSI framework

(⊕OECD 2022OECD. 2022. OECD Public Governance Policy Papers. no. 20. https://doi.org/https://doi.org/10.1787/e19a9be9-en.) provides an attempt at designing a comprehensive ethical framework for behavioural interventions. Its main steps are:

1. Scope: Checking the relevance of behavioural science
2. Design: Integrating ethics into intervention design
   • Consider all relevant stakeholders
   • Have a protocol (e.g. review board)
   • Assess the risk that you stigmatize a population / degrade outcomes for your control group
3. Research and Evaluation: Gathering behavioural evidence responsibly
   • Plan for backfire effects
   • Pre-registration
   • Data privacy
   • Data sharing
4. Policy Implementation: Preserving ethics when implementing and scaling
   • Built-in transparency and accountability

3.8.2 An evolving landscape

This is an evolving field and I cannot summarize it for you, only direct you towards relevant resources. The crux of the matter is that to properly conduct an intervention, you need a set of skills.

• As a project manager, you’ll need a comprehensive view of what’s required. usually, these frameworks are enough for you.
• You’ll need general-level expertise, e.g. the CNIL RGPD MOOC in France, BITSS’s MOOC on Open and Transparent Social Science Research, explicit certification, such as the fhi360 Research Ethics Training Curriculum (RETC).
• You’ll need in your team dedicated expertise on such topics as RGPD, pre-registration, etc. And you can choose to specialize in one of these areas.
• Map and keep up-to-date external actors who can provide you with relevant resources, e.g. which academic institution has a review board you can submit to? Is there the equivalent of a review board within the administration you’re working with?