Cartwright: what worked there may not work here

In 2012, Cartwright and Hardie published a philosophical study questioning how evidence-based policies rely on RCTs to export policies that have worked in a specific place and particular moment to another place in a different time. Cartwright explains that for evidence-based policies, RCTs are the gold standard for establishing what works (Cartwright, Reference Cartwright2012, p. 298). The main issue, she claims, is that the most a good RCT can prove is that it worked somewhere, not that it will work in a different setting. Building upon that criticism, she provides a theory of evidence to evaluate the claims of RCTs regarding their possible applicability to other settings. She argues that evidence that the policy worked somewhere is only a good starting point. One also needs evidence that the policy will work in the target setting. The relevant question is whether the policy will “make a positive difference in the desired outcome” (Cartwright & Hardie, Reference Cartwright and Hardie2012, p. 5).

RCTs are considered by the industry to be the most rigorous tool for predicting the effectiveness of a policy: “A well-designed randomized experiment makes it highly likely that the effect of the treatment be reflected in the data, but does not guarantee that this is going to be the case” (Guala, Reference Guala2012, p. 615). Cartwright posits that, although policymakers would like to have evidence that those policies will work in other settings, RCTs cannot provide this – they can only inform on the efficacy of a treatment: that the policy worked somewhere (Cartwright & Hardie, Reference Cartwright and Hardie2012, p. 9); that is, “that it worked in the studied situation” (Cartwright, Reference Cartwright2012, p. 299). To check for effectiveness, experts need to understand some facts about the causal role that the policy plays and what type of support factors it needs.

RCTs have a clear advantage: by separating control and treatment groups and creating a blind and random process of selection, experimenters can isolate the effects of specific policies on a group of individuals. But the process warrants some “metaphysical premises”:

RCTs, therefore, try to convert probability to causal explanation and to establish that if we observe a positive effect in the treatment group, it had to be caused by the policy. But, according to Cartwright, the outcomes of an RCT can only be applied in settings that share the same acting principles that produced the effect in the study situation (Cartwright, Reference Cartwright2012, p. 313). Since causal principles are local, we should not expect that the policy will work somewhere else (Cartwright, Reference Cartwright2012, p. 310). To have evidence that the policy will work somewhere else, we need to find causal principles climbing up the “ladder of abstraction” (Cartwright, Reference Cartwright2012, p. 311), using more deliberative and thoughtful tools with the goal of finding a causal claim with the capacity to establish that the “treatment reliably promotes the outcome” (Cartwright, Reference Cartwright2011a, p. 1401); that is, that the treatment is able to produce more cases of the outcome in a variety of circumstances.

Accordingly, Cartwright emphasizes the notion of capacity, considering it a “powerful tool,” since that is what may show the effectiveness of the policy and, consequently, that it will work somewhere else: “Effectiveness is what a cause does ‘in the field’” (Cartwright, Reference Cartwright2009, p. 185). Policymakers, she states (Cartwright, Reference Cartwright2009, p. 203), usually have good reasons to believe that their interventions will have an effect in part of the population before they test it with RCTs. They have theories, and although theories sometimes are uncertain, learning how to deal with those uncertainties should be a part of the task of finding the evidence that a policy will work somewhere else:

Ideal RCTs are not the only methodology for arriving at conclusions; other methods that work deductively can clinch them. Econometric methods, Galilean experiments, probabilistic/Granger causality, derivation from established theory and tracing the causal process can also do the work: “[T]here is no a priori reason to favor a method that is rigorous part of the way and very iffy thereafter over one that reverses the order or one that is less rigorous but fairly well reasoned throughout” (Cartwright, Reference Cartwright2013, p. 31). Cartwright also suggests that sometimes RCTs are not needed and that simple observation can do the job without putting the population at risk (Deaton & Cartwright, Reference Cartwright2018, p. 7). As Judea Pearl posits on a commentary to Deaton and Cartwright's work: “In addition, considering the practical difficulties of conducting an ideal RCT, observational studies have a definite advantage: they interrogate populations at their natural habitats, not in artificial environments choreographed by experimental protocols” (Pearl, Reference Pearl2018, p. 60).

Considering the limitations of RCTs, what can we use them for? Cartwright provides some answers to the question: they may yield a type of falsification test to refute a theoretical proposition, or they may confirm the prediction of theory (not the theory itself). They are also able to show that a specific treatment does, in fact, work in a specific setting (Deaton & Cartwright, Reference Cartwright2018, p. 13).

In evidence-based public policy, the effort should focus on the effectiveness of the treatment, not only its efficacy. The question, therefore, is how to warrant that prediction. As Cartwright posits, a “warrant requires a good argument” (Cartwright, Reference Cartwright2012, p. 15), an argument that has to be sound and valid, with premises that can be trusted, and a conclusion that is really implied by those premises. Good arguments lead us to robust conclusions. Accordingly, to check for effectiveness, policymakers need a good argument based on facts about causal principles and not just simply statistical associations. The type of causal principles that experts need to find in social sciences are ceteris paribus: they require that all conditions remain the same. These types of principles can be deterministic or probabilistic.

Finding the causal principles of a policy requires that we look into the support factors that make the policy work. Policies need contributing structures set in place for them to work – they do not work by themselves. They are part of a team of structures that belong to a setting. To predict effectiveness, experts need to understand those support factors. If that were the case, the study would have external validity, since the treatment obtains the same result as it did in the initial study. As María Jiménez-Buedo explained, while internal validity deals with how reliable causal inferences are, external validity tackles the possibility of generalizing them to events beyond the experimental setting (Jiménez-Buedo, Reference Jiménez-Buedo2011, p. 271). Experts in the industry suggest that we can get such external validity when the same treatment is applied to a population that is sufficiently similar to the one from the initial study. Cartwright argues that this perspective is too vague since it does not help with the effectiveness of a policy: applying the same treatment to obtain the same result in a different setting leaves out the actual causal principles, and the requirement for similarity is too demanding and wasteful (Cartwright, Reference Cartwright2012, pp. 46–49). The external/internal validity dichotomy is not exempt from criticism. Jiménez-Buedo suggests that due to certain conceptual problems, this distinction is prone to ambiguous interpretations. She wonders about the notion of validity and what it refers to: Is it the experiment, the type of experiment, the experimental results, the data obtained from experiments, etc. (Jiménez-Buedo, Reference Jiménez-Buedo2011, p. 274)? Cartwright warned us about the danger of substituting deliberate thinking regarding causal factors with just external validity: “Only by thinking in terms of causal roles and support factors can you begin to see what evidence you need if you are going to bet that the policy will work here. You cannot avoid thinking like that. The notions of external validity and similarity are no substitutes” (Cartwright & Hardie, Reference Cartwright and Hardie2012, p. 49). The notions of external and internal validity are not practical for extracting accurate theoretical interpretations from observed events (Jiménez-Buedo, Reference Jiménez-Buedo2011, p. 279).

Accordingly, Cartwright summarizes that the argument to verify the effectiveness of a policy should go as follows: the policy worked in one setting, the policy can play the same causal role in a different place and the support factors that helped the policy to play a positive causal role are shared by at least some individuals in both settings (Cartwright & Hardie, Reference Cartwright and Hardie2012, p. 54). Those premises would yield the prediction that the policy will work on the other settings.

There are several elements to consider in these situations:

• • Causal principles can change once a policy is in place after an RCT has confirmed its efficacy.

• • The deployment of a policy may affect the causal structure of a setting to the point that the policy itself no longer obtains the desired effect.

• • The same policy may carry positive consequences for a group of individuals and negative consequences for another group (Cartwright & Hardie, Reference Cartwright and Hardie2012, p. 45).

• • RCTs are not exempt from producing moral dilemmas: randomization may expose people to unnecessary issues, especially in medicine or public policy matters, which may dramatically affect the lives of individuals (Deaton & Cartwright, Reference Cartwright2018, p. 7).

There are instances where if we apply a higher level of abstraction we can find causal principles that are common in different settings. If a factor is able to play the same causal role in different situations, that factor is acting at a more abstract level. Since support factors and causal principles can verify whether a policy may work in a different setting, and policies work in tandem with other variables (ceteris paribus); if those variables are not part of the setting, their absence will make the policy ineffective. This idea of causation is well explained by E.J. Mullen:

In order to find support factors, we need to do what Cartwright called a “horizontal search” (Cartwright, Reference Cartwright2012, p. 91). If an RCT provides a positive result about a specific policy, its success would be related to a mix of factors that help the treatment to be effective. The other elements needed to verify that a particular policy may work in a different setting are, as mentioned before, causal principles. Those causes can have different levels of abstraction. A vertical search more or less would go up and down through those levels. Vertical searches should be able to determine the appropriate level of abstraction to retrieve common explanatory factors that may intervene in how a policy may yield a positive outcome (Cartwright, Reference Cartwright and Morgan2011b, p. 23).

Experts usually already have an idea of how the intervention may cause the desired outcomes before the RCT. It is not just trial and error, since those experts rely on previous knowledge to bet that the intervention will work, speculating on the causal principles and the support factors needed for it to be successful. Science will provide for you that type of background knowledge. In social sciences, one needs to find and understand what happens and the causal roles of the intervention. That is what vertical and horizontal research will give you (Cartwright & Hardie, Reference Cartwright and Hardie2012, p. 126). Beyond using RCTs, one must think and deliberate: “Deliberation is not second best … To deliberate in order to exercise discretion requires a rich list of intellectual and practical virtues that cannot be reduced to the virtue of conformity. Thus, the orthodoxy not only discourages deliberation as unnecessary, since the rules are superior but selects in favor of operatives who cannot deliberate” (Cartwright & Hardie, Reference Cartwright and Hardie2012, pp. 158–159). Deliberation may increase the external validity of an experiment, opening up the possibility of answering the question: “Can we use experimental knowledge to understand what goes on in the ‘real world’?” (Guala, Reference Guala2012, p. 612). When researchers conduct an RCT seeking external validity, they should immerse themselves in a process of deliberation to provide inferences that will yield that validity. Deliberation should help with what Francesco Guala called the “justification” of the inference process (Guala, Reference Guala2012, p. 612).

If mistakes are made while thinking, the policies will not be effective, even if RCTs are used. Creating rules to avoid thinking eliminates a necessary part of the process of discovering the effectiveness of policies.

Nudge theory

We have examined Cartwright's analysis of evidence-based policies and RCTs. The following pages will be devoted to nudges as evidence-based policy interventions with the objective of analyzing their epistemological status using Cartwright's philosophical approach. These pages will also show how the consideration of expressive rationality can make nudges more robust while attending to ethical considerations related to people's autonomy. Accordingly, this section starts with a short history of nudge theory and the psychological framework behind its use, to help us examine how nudges conceive human rationality and how this conception may be problematic. It will conclude by showing how the consideration of expressive rationality and the application of Cartwright's approach by weighing the causal principles and support factors of nudges within a deliberative process may allow policymakers to obtain more effective results while respecting the autonomy of those who are nudged.

Most nudges rely on an instrumental view of rationality. Holding a narrow view of rationality within a means–end schema, nudges only focus on how treatments can positively affect the outcome of a decision and if behavior satisfies the goals of the individual. Accordingly, the most rational agent will select the appropriate means to maximize their utility function. Nudges focus only on the instrumental part of rationality and consider irrational the behavior that will not yield maximization. A nudge is not only a tool to help the instrumental rationality of individuals to become more effective, but also an intervention to increase social welfare by changing people's behavior (e.g., increase organ donation). Most policymakers that use nudges tend to neglect the reasons why individuals behave in the way they do or why they modify their decisions. To verify whether those nudges will work during a period of time in the same or a different setting, policymakers test them, as mentioned above, mostly using RCTs (Einfeld, Reference Einfeld2019, p. 509). To provide a measure of the scale of the use of RCTs by nudge units, a recent analysis on nudges conducted by Stefano DellaVigna and Elizabeth Linos (Reference DellaVigna and Linos2020) used data from the two biggest nudge units in the USA: it included 126 RCTs and 23 million people. Anneliese Arno and Steve Thomas, for example, conducted a systematic review of current nudge papers to determine whether these strategies “are successful in changing adults’ dietary choices for healthier ones” (Arno & Thomas, Reference Arno and Thomas2016, p. 1). For the review, they used 37 papers, 31 of which (74%) reported RCTs.

Nudges have the possibility to fail or backfire, and some authors argue that they may even pose a limitation to the autonomy of the decision, since they sometimes manipulate the choice environment without the awareness of those they are trying to nudge (McCrudden & King, Reference McCrudden and King2015).

Ever since Herbert Simon coined and developed the notion of bounded rationality (Simon, Reference Simon1957) to describe the processes of how people decide, numerous authors have approached social sciences from that perspective, considering that our rationality is bounded by cognitive limitations and by the complex structure of the environment. While standard economics presumed an economic agent with unlimited cognitive skills, infinite memory and capable of optimizing their decisions, Simon sustained that a robust empirical foundation of economics contradicted that view. When Simon introduced psychology as part of the theoretical framework for understanding specific phenomena in social sciences, providing a more realistic description of economic decisions, a new type of social agent appeared: one with bounded rationality. Simon claimed that people satisfice instead of optimize. The rationality of our decisions, therefore, had to be understood within the boundaries imposed by our rationality and the structure of the environment.

The idea that our rationality was limited was also later developed by other authors. The work of Amos Tversky and Daniel Kahneman in heuristics and biases (Kahneman & Tversky, Reference Kahneman and Tversky1979; Tversky & Kahneman, Reference Tversky and Kahneman1974, Reference Tversky and Kahneman1981) showed that our rationality systematically errs, concluding that people “suffer” from a set of biases that makes them “irrational.” They claimed that the systematicity of these errors allows for their prediction. Following that view, some public policymakers suggest that certain interventions can be proposed in a way that might respond to those failures without limiting the freedom of the agent. That biases are systematic, epistemologically speaking, indicates that we can scientifically operate with them and predict their occurrence. Social scientists believe that RCTs can be used to predict the effectiveness of their interventions, but as mentioned before, no evidence supports the possibility of that prediction. I will return to this issue below.

Richard Thaler and Cass Sunstein, following the work of Tversky and Kahneman (Reference Tversky and Kahneman1974), proposed a solution to people's irrationality without disturbing their freedom to choose (Thaler & Sunstein, Reference Thaler and Sunstein2009): nudges. Tversky and Kahneman are not themselves nudge theorists; rather, nudge theory rests on the research work on biases and heuristics conducted by these authors. They recognized the systematicity and predictability of people's rational errors. Consequently, since people's decisions are inserted in a choice environment and that it is costly and sometimes impossible to educate them, nudge theorists maintain that the choice architect can organize the environment of the decision in a way that would take into consideration the biases of individuals to increase the efficiency (instrumental) of public policy interventions regarding behavioral change.

Considering that nudges are able to promote change while respecting freedom of choice, different nudge units have been created throughout the world by governments and institutions captivated by their success and promises of efficiency: the Behavioural Insights Team (BIT), the Ontario Behavioural Insights Unit (OBIU), the Behavioural Insights Network Netherlands, MineduLab in Peru, the Behavioural Economics Team in Australia, etc. The BIT emphasizes above all the use of RCTs for testing purposes. David Halpern, chief executive at BIT, promotes their use:

A nudge can be a change in the location of a salad bar in a university cafeteria, the placement of a signature on a tax form or a default option to save 6% of your salary for retirement. Nudges can be successful and, if they fail, proponents argue that problems are limited, since they should be easily avoided by individuals. Cass Sunstein indicates that even if nudges are not effective, they should be kept if they improve people's welfare: “A largely ineffective nudge may have positive welfare effects; an effective nudge might turn out to reduce welfare. A strong reason for nudges, as distinguished from more aggressive tools, is that they preserve freedom of choice and thus allow people to go their own way” (Sunstein, Reference Sunstein2017, p. 22).

Expressive rationality and nudge theory

In economics and social sciences, rationality is commonly defined instrumentally (Nozick, Reference Nozick1993, p. 133): it helps individuals to choose the most efficient means to get to their desired goals. Nudges, from a philosophical perspective, look at decisions through that instrumental lens, aiming at altering the outcomes of behaviors by manipulating the choice environment. Other authors (Boudon, Reference Boudon1998; Álvarez, Reference Álvarez2002; Echeverría & Álvarez, Reference Echeverría, Álvarez and Agazzi2008; Bouwmeester, Reference Bouwmeester2017) claim that rationality is not only instrumental, but also expressive: people have reasons to act in a specific way, sometimes regardless of the consequences. Often people want to express their subjective selves with their actions.

In order for nudges to be more efficient, besides considering the outcome of decisions, they should try to understand rationality comprehensively, by examining its bounded (Bouwmeester, Reference Bouwmeester2017, p. 3), social and expressive components (Hortal, Reference Hortal2019). Conceiving rationality only within an instrumental frame would neglect essential elements of rational decisions, impeding a full understanding of the behavior. Accordingly, nudges should include the instrumental and bounded aspects of rational behavior (as they are currently doing), understanding at the same time the expressive and the social rationality of the causal factors of group behavior (Bicchieri, Reference Bicchieri2017, p. 48).

What is important for our instrumental rationality is that we are aware of the causes and effects of our behavior, so we can decide efficiently. According to this frame, rationality's spectrum would have two poles: more rational, if you choose the means that yield the desired outcomes; and less rational, if you choose something less desirable. Our instrumental rationality focuses on how to be efficient in order to arrive at our goals, but it says nothing about the goals themselves. Nudge theory, examining rationality within just an instrumental frame, views people as fundamentally irrational. Nudge theorists complement this view by using the two-system cognitive model proposed by Kahneman (Reference Kahneman2011) to understand decisions, where system 1 is automatic and intuitive and system 2 is deliberative. That distinction is not exempt from criticism. Gerd Gigerenzer, for example, claims that this division is rather problematic due to its vagueness:

When we behave, we are not only trying to achieve a desired goal (e.g., losing weight, saving more money) – we also have reasons for doing what we are doing. Those reasons can be social or expressive. Social reasons are anchored to intersubjective elements, while expressive reasons are subjective. Behavior is complex and is grounded on the three different components of expressive, social and instrumentally rationality. For example, if we vote in a general election, our instrumental rationality may have the illusion that our vote matters and our action will have consequences (Opp, Reference Opp2015, p. 191). We also have good reasons to vote: our social rationality may see it as a norm, and our expressive rationality may enjoy the act of voting: “Expressive rationality refers to what we value subjectively or as a single person. Expressive arguments can, for instance, explain the rationality of what we do during a holiday, how we dress, what music we listen to, etc. Acting on personal impulses, interests, or motives can make our actions rational from a subjective perspective” (Bouwmeester, Reference Bouwmeester2017, p. 8). Accordingly, we can explain actions within different rational frames, but we should assume a comprehensive model of rationality. In the previous example, the comprehensive approach would conclude that voting can be instrumentally irrational while expressively and socially rational.

The reductionist instrumental approach disregards part of the reasons behind human behavior, explaining everything using a vague two-system theory. This view causes the theory of nudges to rest on an incomplete theory of human rationality. The comprehensive theory, which includes expressive and social rationality as part of the theoretical frame, can make nudges more efficient, and since those expressive aspects of rationality act as causal principles of behavior, an understanding of these aspects can provide much-needed evidence for the use of nudges in settings that go beyond the initial one targeted by the RCT. Cristina Bicchieri (Reference Bicchieri2017) and other researchers (Bicchieri & Dimant, Reference Bicchieri and Dimant2019) are already working on the social aspects of nudges in a way that regards our social rationality as part of the reasons behind behavior without stigmatizing these social aspects as irrational. Some authors, for example, are using these social aspects to test interventions related to energy consumption (Brandon et al., Reference Brandon, List, Metcalfe, Price and Rundhammer2019) or vaccination (Korn et al., Reference Korn, Betsch, Böhm and Meier2018).

There is a major issue when including expressive rationality in nudge theory: its subjective aspect. While nudges consider the instrumental and bounded aspects of rationality, they do so by attending to the systematic biases of human rationality. They rest on universal and systematic aspects of human cognition. When social nudges are deployed, they are intervening in a social context where, more or less, experts can obtain information about the social norms of the setting. It is difficult to gather data or information regarding expressive elements that can be used by policymakers when researching possible nudges.

Nudges should consider expressive rationality in two different ways: to respect the subjective autonomy of individuals and to make interventions more efficient. Our expressive rationality is molded culturally, so it is possible that some expressive components are shared by large groups of individuals in the same setting. If that is the case, research can be conducted to include these expressive components in behavioral interventions.

For example, if the purpose of a nudge is to change people's behavior so that they can eat healthier, the most effective nudge would understand:

• • The bounded rationality of individuals in their instrumental actions

• • Their social rationality and how social norms play a role in how we eat

• • Their expressive rationality and how tastes, likes and dislikes affect what we eat and how we eat it

To this end, that intervention would not only change the environment to position a salad in a cafeteria in a way that may increase its consumption – it would also try to make salad consumption a social norm while making it tastier (or by introducing other expressive elements such as soccer teams for salad names, etc.).

The idea, therefore, is to make sure that nudges are aware of the diverse components of our rationality, and that these components are included in the search for the causes of our behavior and for the causes of why policies work in a particular setting. Accordingly, understanding those other types of rationality should be a must for nudge theorists. Our decisions are a complex system in which those different rationalities interact. Our beliefs, thoughts and tastes – that is, what Pierre Bourdieu called habitus (Bourdieu, Reference Bourdieu1985) – are essential causes of behavior. The notion of habitus represents a group of tendencies that lead our behavior and form our identity, and it is the result of social externalities related to class and other circumstances. Nudges should consider this habitus – the social and expressive components of behavior – to understand the differences in the effects of interventions.

Concluding remarks: expressive rationality, Cartwright's causal principles and nudge theory

In the first section of this paper, we reviewed Cartwright's philosophy and examined how she argued that evidence-based policies should try to understand causal principles and support factors besides using RCTs to provide more robust evidence for their effectiveness. To this end, social scientists should search horizontally and vertically to find those principles and factors. Once they are found, they can have a better and more comprehensive understanding of the policies and the mechanisms that make them work.

Nudges, using behavioral insights focusing on cognitive biases, change people's behavior by manipulating the choice environment. As evidence-based policies, they heavily rely on RCTs. As Jiménez-Buedo argues, an RCT does not guarantee the possibility of forming a robust inference regarding a causal relationship (Jiménez-Buedo, Reference Jiménez-Buedo2011, p. 274). Consequently, I claim that nudges can be improved if we apply Cartwright's epistemological recommendations mentioned above. Another improvement can be achieved by examining people's expressive rationality, which acts as a causal principle regarding their behavior. Understanding the reasons as to why people behave will increase the effectiveness of nudges and provide understanding of how they can be deployed while respecting people's subjective autonomy. This type of understanding, using Cartwright's words, will allow experts to “climb up the ladder of abstraction,” increasing trust in the specific nudge intervention. Our expressive rationality acts as a causal principle of our behavior. Understanding the reasons as to why people behave will increase the chances of collecting more evidence regarding the effectiveness of nudges. For example, a study conducted to increase influenza vaccination rates among employees (Milkman et al., Reference Milkman, Beshears, Choi, Laibson and Madrian2011) showed that mailing people reminders listing the time and location of the appointment with a prompt instructing them to write down the date increased the vaccination rate by 1.5% (which was statistically insignificant). If, instead of just the date, the prompt also asked the recipient to write down the time of the appointment, there was an increase in the vaccination rate of 4.2%. Was the change related to just adding the actual time of the appointment? It may be the case, for example, that what really caused the behavioral change was not related to writing down the time of the appointment, but the possibility of writing something about the appointment so that individuals can spend a few more seconds thinking about the appointment. If we understand what caused the increase, we can implement it in other settings.

Text messages, for example, have been used successfully to increase loan repayments in Uganda (Cadena & Schoar, Reference Cadena and Schoar2011), to improve the collection of overdue fine payments (Behavioral Insights Team, 2012) and to increase commitments to savings (Karlan et al., Reference Karlan, McConnell, Mullainathan and Zinman2016). The same intervention (text message) did not work to increase the number of future students completing FAFSA, a form for obtaining financial aid for college (Bird et al., Reference Bird, Castleman, Denning, Goodman, Lamberton and Rosinger2019). Text messages also did not work to improve academic performance (Oreopoulos & Petronijevic, Reference Oreopoulos and Petronijevic2019). Why do text messages work in certain scenarios and not in others? Understanding what causes the above-mentioned behavioral changes on a more abstract level than the medium of the intervention can improve the effectiveness of policies. People's expressive (and social) rationality is involved in their decisions – understanding their different levels of influence in specific decisions will improve the effectiveness of nudges.

Reminders are also a good example of how nudges work. Trying to increase adherence to medications prescribed after heart attacks, Volpp et al. (Reference Volpp, Troxel, Mehta, Norton, Zhu, Lim, Wang, Marcus, Terwiesch, Caldarella, Levin, Relish, Negin, Smith-McLallen, Snyder, Spettell, Drachman, Kolansky and Asch2017) tested the effect of electronic reminders (nudge) under the assumption that they would increase such adherence. Contrary to their initial hypothesis, an RCT conducted with more than 1500 patients showed that those nudges caused no improvement. These authors thought that since similar nudges had worked in other settings, they would work on their own patients: “Adherence to medications prescribed after acute myocardial infarction (AMI) is low. Wireless technology and behavioral economic approaches have shown promise in improving health behaviors” (Volpp et al., Reference Volpp, Troxel, Mehta, Norton, Zhu, Lim, Wang, Marcus, Terwiesch, Caldarella, Levin, Relish, Negin, Smith-McLallen, Snyder, Spettell, Drachman, Kolansky and Asch2017, p. 1093). As they mentioned, similar approaches “have shown promise,” but this promise is not a warrant for effectiveness. If people's adherence to medication does not improve after specific policies are in place, more RCTs with different nudges are not going to fix the situation without a comprehensive understanding of the reasons behind people's behavior. The solution cannot simply consist of trialing new interventions without trying to understand the expressive rationality of individuals. For example, the research mentioned above claimed that behavioral economics could offer “promise in improving motivation for desirable but difficult activities, such as weight loss, exercise, or smoking cessation by harnessing pervasive patterns of irrational behavior” (Volpp et al., Reference Volpp, Troxel, Mehta, Norton, Zhu, Lim, Wang, Marcus, Terwiesch, Caldarella, Levin, Relish, Negin, Smith-McLallen, Snyder, Spettell, Drachman, Kolansky and Asch2017, p. 1094). Treating human rationality as a black box and establishing nudges by testing their outcomes while disregarding expressive or social reasons for behavior, all while defining non-normative behavior as irrational, can make nudges ineffective. Throughout their research article, not a single word is mentioned on the possible reasons for the lack of adherence. However, the authors did observe that due to the negative results of the intervention, different approaches need to be considered (Volpp et al., Reference Volpp, Troxel, Mehta, Norton, Zhu, Lim, Wang, Marcus, Terwiesch, Caldarella, Levin, Relish, Negin, Smith-McLallen, Snyder, Spettell, Drachman, Kolansky and Asch2017, p. 1099).

After facing negative results in the outcomes of policies based on nudges, authors tend to request more research. Using Cartwright's perspective, what nudges need is a better understanding of expressive and social rationality. They both act as causal principles and as determinants in order for policies to work. Nudges should also pay attention to the support factors involved in policies, not simply testing policies using RCTs. They should use vertical searches to determine the expressive and social reasons underlying people's behavior. They should also use horizontal searches to discover the support factors that might help to explain why some policies fail while others succeed. Even if a causal link has been found in an experimental setting, it will be difficult to extrapolate the finding to the real world, since we may not be aware of all of the possible factors contributing to the effect (Jimenez-Buedo & Miller, Reference Jimenez-Buedo and Miller2010). By understanding the expressive rationality of individuals in their research, policymakers can also ensure that the autonomy of the individual in their decision-making process is respected. A nudge that positively alters the choices people make by tackling the reasons behind their actions is a good method for ensuring respect of their autonomy. Behavioral changes that bypass people's awareness can pose serious moral and political problems.

Consequently, in conjunction with RCTs, policymakers should think and deliberate about the causes of behavior, its changes and the causal mechanisms of policies. Cass Sunstein sometimes has argued for a similar idea: “[I]f a nudge is based on a plausible but inaccurate understanding of behavior, and of the kinds of things to which people respond, it might have no impact” (Sunstein, Reference Sunstein2017, p. 20). I regard this task as being rather difficult if our expressive rationality is not considered as one of the causal principles involved in our behavior and, consequently, as a subject of study for policymakers.

In future research, it would be interesting to define how experts can “climb the ladder of abstraction” to find the causes of behavioral change and how they relate to specific nudges considering the different aspects of rationality. In the vertical search for causes, the application of Pierre Bourdieu's sociology, particularly his work on habitus and social fields, may shed some light on how nudges may affect people differently due to inequality and other social elements that act as causal principles. With habitus, Bourdieu refers to tastes, beliefs, interests and our understanding of the world. These elements, as integral parts of the non-instrumental aspects of our rationality, should be included in any research regarding nudges.