Heterogeneous Effects of Informational Nudges on Pro-social Behavior

1 Introduction

Over the past decade, informational nudges, a type of information provision with the purpose of altering social behavior, have become increasingly prominent in the literature (Allcott and Mullainathan, 2010; Thaler and Sunstein, 2008; Cialdini et al. 2006). Moreover, real-world implications have been acknowledged in multiple spaces. In the marketplace, research has shown that peer comparisons can induce pro-social behavior in energy consumption (Ayres, Raseman, and Shih 2013). Nudge, a New York Times bestseller written by Cass Sunstein and Richard Thaler (2008), introduced and advocated the concept of “nudges,” a type of government policy that helps people make better choices without limiting their freedom to choose. These policies have seen widespread popularity: the Behavioral Insights Team, launched in the UK in 2010, has successfully nudged tax debtors to pay more of their taxes owed by mailing them letters telling them about the behavior of others^[1] while the White House has kicked off several federal projects in the second half of 2013 to nudge Americans to save more money for retirement, to curb energy uses, and to trim energy costs.^[2]

Despite their increasing popularity, evidence for the success of programs using informational nudges to induce pro-social behavior has been mixed (Schultz et al. 2007). In particular, making the most of these policies requires understanding the heterogeneity in responses across subpopulations and across intervention designs. One literature focuses on transient situational factors that affect the short-term response to a nudge such as the importance of the wording of the message, for example, whether the message is injunctive, “this is what you ought to do,” or descriptive, “this is what others are doing” (Ferraro and Price, 2013; Merritt, Effron, and Monin 2010; Cialdini et al. 2006). Another literature focuses on long-run dispositional difference across different groups like differences in social responsibility, group identity, or prior behavior (Akerlof and Kennedy, 2013; Allcott, 2011; Beshears et al. 2011). The heterogeneity in impact can lead to unintended spillovers. For instance, a nudge toward good behavior in one domain may license individuals to behave badly in a different domain (Ho etal. 2015). Even successful nudges such as the letters to tax debtors sent in the UK are instructive of the limitations of the nudge approach. The UK policy focuses only on a subset of the worse offenders, and was only effective when the message carefully selected the comparison group (Hallsworth et al., 2014).

Therefore, despite the popularity of nudges in popular discourse, important questions remain. How should we account for the heterogeneous impact of nudges across different populations or different interventions both in terms of designing behavioral policies and also in terms of evaluating their welfare impact? More importantly, as the magnitude of change observed in most field studies is small – for example, 1–2% savings in the case of electricity (Allcott, 2011), or less than 1% in the case of water savings (Ferraro and Price, 2013) – do the benefits of the nudge outweigh the feelings of guilt they induce? Also, as nudges by themselves seem to have small effects, would nudge-based policies work better in conjunction with traditional economic incentives like Pigouvian taxation?

In this paper, we organize what we know from the empirical literature into a coherent framework to think about the heterogeneity of nudges. The framework allows us to address questions about how behavioral interventions interact with traditional economic policies and to assess impacts on social welfare. We focus on the interaction between two dimensions of heterogeneity, the effect of situational nudges and individual disposition, developing a model that considers short-term situation and long-run disposition as two separate channels for inducing the warm-glow or cold-prickle associated with impure altruism. While some of the past literature we have noted does acknowledge the heterogeneity, there has been no systematic study of how different kinds of nudges interact with disposition. Existing research focuses on each nudge intervention independently. Also, while the distinction between disposition and situation is a fundamental one in psychology, it is a relatively underutilized conceptual framework within the realm of economics.

We parameterize the situational impure altruism induced by nudges and the dispositional impure altruism inherent to the individual using cross-derivatives, and predict behavior and welfare based on this relationship. By organizing the existing literature in this way, we hope to provide guidance for the empirical research needed to properly assess the welfare of nudge-based policies.

The main policy context we are interested in is informational message campaigns targeted toward reducing consumption of goods that involve negative externalities: e.g., reducing consumption of gasoline in order to mitigate environmental damage. Therefore, the main disposition we will focus on is whether people are more or less inclined to have altruistic preferences toward curtailing their own consumption for the good of the public. Since the natural policy in this context is Pigouvian taxation, we will consider the interactions between our two dimensions of impure altruism and optimal tax policy, although the model applies to any traditional economic incentive scheme designed to change consumption behavior. Even in domains where changes to tax policy are politically infeasible, design of behavioral interventions should depend on the existing tax environment.

The main findings of the paper include the following:

1. The effect of pro-social nudges on welfare can be either positive or negative.

2. The observed heterogeneity across behavioral interventions can be explained by the cross-derivative between situational and dispositional effects.

3. The optimal tax is a function of disposition and situational nudges.

   1. Targeted taxes on certain subpopulations may be welfare enhancing.

   2. Optimal economic policy accounts for interactions with situational nudges.

   3. The inefficiency of a tax policy that does not take into account impure altruism and situational nudges depends on the complementarity between situational altruism and dispositional altruism.

2 Literature Review

The gaining popular use of informational nudges is a result of the emerging behavioral economics and psychological literatures. Bergstrom, Blume, and Varian (1986) introduce the idea of impure altruism and acknowledge its role in driving individual choices. Andreoni (1990, 1995) extends the previous analysis and incorporates impure altruism into a more general model, reasoning that nudges work by inducing the feeling of a cold-prickle when creating negative social externalities. Ayres, Raseman, and Shih (2013) find in two large field experiments partnered with private companies, Positive Energy and oPower, that by providing feedback to customers on home electricity and natural gas usage with a focus on peer comparisons, utilities can reduce energy consumption at a low cost. Berger and Rand (2008) also acknowledge the power of peer information provision in redressing heavy drinking in universities through laboratory and field experiments.

However, informational nudges are not always effective and their effects heavily rely on the type of people targeted and the framing of the message. While some nudges do effectively reduce production of a negative social externality, others have no impact or may even be counterproductive. Cialdini, Kallgren, and Reno (1991) explain the heterogeneous effects by arguing that injunctive nudges, which guide behavior by shifting the perception of how most others would approve or disapprove of a person’s conduct, are more likely to lead to beneficial social conduct than the descriptive types, which guide behavior through the perception of how most others would behave. Cialdini et al. (2006) show how some nudges backfire. In Arizona’s Petrified Forest National Park, while the rate of petrified wood theft decreased when people were given an injunctive message about what behaviors were not considered acceptable, another treatment using a descriptive message noting that others routinely violated the prohibition proved to be counterproductive. Merritt, Effron, and Monin (2010) explain how one act of pro-social behavior can lead to more pro-social behavior when it is seen as a commitment to the cause, but can lead to more antisocial behavior if it makes people feel like they have done “enough.” Ferraro and Price (2013) revisit a field experiment about water consumption and find that messages based on social norms and social comparisons are more effective than instructive descriptions for reducing water usage.

The type of nudge is not the only source of heterogeneity in terms of effectiveness. Different groups of people are often affected differently by the same informational messaging for two general reasons. The first reason has to do with different people beginning with different status quos. Allcott (2011) notices the heterogeneous effects by evaluating oPower’s Home Energy Report letters comparing residential utility customers’ electricity usage to that of their neighbors. Residents using less than the average increased their energy usage, while though those using more decreased usage – mean energy usage was therefore largely unchanged. Akerlof and Kennedy (2013) further point out that campaigns deploying messages describing public levels of drug and alcohol use, recycling, and littering, etc., have often had little success in heightening adoption of pro-social behaviors. They attribute the ineffectiveness of informational nudges to their heterogeneous impact on people whose conducts are above social norms (socially desirable) and people whose conducts are below social norms (socially undesirable). The second source of heterogeneity comes from differences in individual disposition for pro-sociality due to factors such as group identity, perception of moral obligations, and individual altruism. Beshears et al. (2011) find that peer information increased retirement savings of nonunionized recipients but decreased savings of unionized recipients, attributing the difference to differences in norms between the union and nonunion workers. Jordan, Mullen, and Murnighan (2011) find that people who were asked to recall their immoral behavior reported greater participation of moral activities than people who recalled moral behavior. Ho et al. (2015) find in paired lab and field studies of green electricity purchases that informational nudges have a stronger effect on those who were intrinsically inclined to be pro-social.

Our paper considers situation and disposition as two ways to induce the warm-glow or cold-prickle associated with impure altruism and parameterizes their effects into two temporal dimensions. Formally, disposition is a tendency by an individual to act in a specified way and is stable over time. We denote the effect of disposition by θ and call it the parameter for dispositional altruism. A person with high θ is more likely to be pro-social whereas a person with low θ is less likely. A nudge, on the other hand, has a situational effect that changes each period. We denote the effect of nudges by ω and call it situational altruism. In the case of informational nudges, someone who receives a nudge and thereafter perceives his behavior to be more socially desirable than the average (above the norm) will experience low ω. Someone who perceives his personal behavior to be less socially desirable than the average (below the norm) will experience high ω.

Our theory shows how the heterogeneous effect of behavioral interventions depends on a third factor: the interaction between situation and disposition. We examine the individual psychic cost function for the relationship between situation and disposition. Situation and disposition are complements when people that are more altruistic are more receptive to information that confirms their prior inclinations. Situation and disposition are substitutes when a situational nudge causes resentment in those most likely to give.

Our findings are consistent with the self-completion theory and moral regulation model in psychology. According to the self-completion theory of Jordan, Mullen, and Murnighan (2011), “recalling one’s (im)moral behavior will lead to compensatory rather than consistent moral action as a way of completing the moral self.” Sachdeva, Iliev, and Medin (2009) explain the compensatory moral action differently and argue that moral or immoral behavior can result from an internal balancing of moral self-worth and the cost inherent in altruistic behavior. When moral identity is threatened, moral behavior is a means to regain some lost self-worth (moral cleansing); however, affirming a moral identity leads people to feel licensed to act immorally (moral licensing). An experiment conducted by Darlington and Macker (1966) shows that participants who were led to believe that they had harmed another person were more likely to subsequently engage in altruistic behavior such as donating blood to a local hospital, indicating guilt-induced moral cleansing. On the other hand, moral licensing can happen when good deeds of those who are above the norm establish moral credits that can be “withdrawn” to “purchase” the right to do bad deeds with impunity (Merritt, Effron, and Monin 2010). Monin and Miller (2001) notice that licensing operates by providing a temporary boost in self-concept, and an initial altruistic intent that boosts self-concept can liberate people to choose more indulgent option. Khan and Dhar (2006) and Ayal and Gino (2011) also note that people may feel licensed to refrain from good behavior when they have amassed a surplus of moral currency.

Formally, our model shares much with that of Andreoni (1990) as we both aim to extend the analysis of Bergstrom, Blume, and Varian (1986) by decomposing their original idea of impure altruism. Andreoni’s model characterizes people as either pure altruists whose preferences only depend on the total consumption of a public good or as egoists whose preferences only depend on the private consumption of the public good, with impure altruism combining the two. We add to Andreoni’s idea of egoism and impure altruism by focusing on how impure altruism may arise from two distinct sources, i.e., situation and disposition, modeling the effect of situation and disposition through a psychic cost function. Moreover, just as Andreoni (1995) notices the behavioral asymmetry between the warm-glow of doing something good and the cold-prickle of doing something bad we uncover a different type of behavioral asymmetry, one not based on the consequences of the behavior but based on the disposition of people and the interaction between situation and disposition.

We also endogenize our measure of impure altruism by allowing the situational effects of a nudge to vary with the individual’s own choices. A common behavioral policy intervention is to tell individuals how their behavior compares to the behavior of others. However, if implemented on a large scale, the “behavior of others” also depends on the intervention. Rotemberg (1994) examines how altruism may arise endogenously among a small set of strategically related individuals. Bowles (1998) discusses how markets and other economic institutions may give rise to endogenous preferences by changing the exogenous determinants in a cultural equilibrium. Casadesus-Masanell (2004) studies how principals may utilize motivational schemes that rely on social influences such as norms, ethical standards, and altruism to foster intrinsic motivation and trustworthy behavior. Our interest differs from theirs in that we are less interested in how these preferences arise. Instead we are interested in how the government needs to account for these preferences when considering behavioral policy interventions. We formalize such endogeneity by deriving the optimal message that policy makers should send regarding the behavior of others and show that the optimal message should take into account the population’s heterogeneity as well as the complementarity between dispositional and situational altruisms.

We discuss above examples of heterogeneity in response to such messages in trials involving messages about water usage (Ferraro and Price, 2013) or electricity usage (Allcott, 2011) or carbon footprint (Ho et al. 2015). Other studies in the lab and field shed light about the mechanisms for how information about the behavior of others influences behavior. For example, Azmat and Iriberri (2010) examine the effect of a natural experiment where high school students learn their relative rank. They find the strongest effect of such information on the tails of the distribution and argue that such information works because people have competitive preferences that make especially high or low rankings more meaningful.

A different type of heterogeneity has been demonstrated in field data for cashiers (Mas and Moretti, 2009) and envelope stuffers (Falk and Ichino, 2006) who observe the performance of peers. Both find that working next to higher performing peers induces higher performance, but primarily for low-performing workers. Mas and Moretti (2009) consider the alternative mechanisms of social pressure, contagious enthusiasm, and knowledge spillovers, and argue that social pressure to conform to high performance is the driving mechanism.

Masclet et al. (2003) identify a different source of potential heterogeneity that operates through the mechanism of social pressure. They show that nonmonetary punishments, i.e., the verbal expression of disapproval, will induce some people to contribute to a public good. Individuals of different dispositions would likely respond differently to such nonmonetary sanctions. A similar nonmonetary effect can be seen in Ball et al. (2001) and Eckel, Fatas, and Wilson (2010), which show that people respond to arbitrarily assigned status in both market and public goods games. Knowing how you rank in some meaningless task domain affects behavior in other domains.

The main limitation to all of these studies is that even those that consider heterogeneity tend to focus on a single intervention of fixed size. This paper highlights the need to consider how the heterogeneity in the response to a nudge varies as the size of the nudge varies. This paper studies such variation in a general model that accounts both for economic mechanisms such as social disapproval, peer comparisons, and status and also for more psychological mechanisms like moral regulation and guilt.

Our paper further explores the policy implications of behavioral nudges in light of optimal tax treatments. Our tax model builds on Sandmo’s (1975) model for optimal commodity tax, but with a damage that comes from both the externality and a psychic cost. Johansson (1997) and Diamond (2006) also consider warm-glow preferences to optimal tax calculations, but unlike Diamond, we focus on the cold-prickle from the creation of a negative externality rather than warm-glow, and additionally explore the interaction effects between taxes and impure altruism.

3 Main Model and Propositions

4 The Role of Government: A Model of Taxation

As noted in the general model in Section 3, social welfare loss due to choices made by self-centered individuals may serve as the rationale behind governmental intervention. One possible way to alter choice of consumption is through the use of informational nudges, as discussed in the previous sections. Such interventions are less conventional but are favored by many researchers and policy makers over commands, requirements, and prohibitions (Thaler and Sunstein, 2008). Nevertheless, more conventional governmental interventions have been in place for a longer time and are still worth studying. This section will be devoted to the discussion of one specific type of traditional intervention– taxation. This section aims to provide insights into the optimal tax in the presence of dispositional and situational pro-sociality.

We first consider the first best policy, where government chooses both taxes and nudges optimally. However, policy makers often see nudges as a way to influence behavior when taxes are politically infeasible. In such cases, we hope our model serves as a reminder that nudges are not designed in a vacuum, and should be designed to complement existing tax policies.

5 Discussion: Implications for Policy

As mentioned in the introduction, local authorities and central governments in the UK, the USA, and France have been avidly engaged in the use of informational nudges to increase tax pay-offs, to encourage more retirement savings, to improve energy efficiency, just for a start. However, Akerlof and Kennedy (2013) note limited success over the past 15 years of campaigns using messages of social comparison in areas including drug and alcohol use, recycling, and littering. Our model shows that informational nudges may have very limited impact for three reasons. First, a nudge is less effective when it fails to induce high situational altruism. Cialdini et al. (2006) point out that informational nudges, when purely descriptive, may not be effective in mitigating socially disapproved conduct and may even worsen the problem. To address this counterproductive licensing effect, policy makers should pay attention to how social messages are conveyed.

Second, our model shows that nudges may have very limited impact if low dispositional altruism is dominant in the targeted population. Hence, it is crucial for the government to understand the disposition of certain groups of people when designing public programs to induce pro-social behavior through nudges. Moreover, better understanding of people’s disposition will allow the government to target certain subpopulations and use its resources more efficiently.

Third, even when a nudge successfully induces high situational altruism and when people’s disposition can be identified, the effectiveness of nudges still depends on the relationship between nudges and disposition. While existing work on mechanisms behind situational nudges can provide guidance on this relationship, experiments should be reanalyzed and redesigned to test theories about how certain subpopulations’ situation and disposition interact, and how different types of nudges target different subpopulations. If a nudge can be designed so that people view them as a substitute for disposition, then it may be used to target people with a bad disposition. If a nudge can be designed so that people view it as a complement to disposition, then that nudge should be used to target people inclined to be good.

Some types of dispositional impure altruism include conformity, moral obligation, group identity, etc. Some examples of informational nudges that can induce situational altruism include the use of heuristics (anchoring, availability, and representativeness), framing, emotion arousing, feedback, depiction of social norms, and peer comparison (Thaler and Sunstein, 2008). Others are based on social disapproval, status, or social pressure (Kandel and Lazear, 1992). While our model suggests that pro-social behavior following a nudge is affected by the relationship between disposition and situation, we have not addressed how to identify whether disposition and situation are complements or substitutes in a specific real-life case. However, our model provides a foundation for testing hypotheses based on field experiments. For instance, if situation and disposition are complements, then according to our model, a nudge that induces situational guilt will be more effective for those with high disposition. Costa and Kahn (2010) find in a field experiment that environmentalists are more responsive to green nudges than the average person, implying that situation and disposition are very likely complements in this case.

This paper also considers taxes as an alternative or supplemental tool of public intervention in inducing socially optimal behavior and that the optimal taxes depend on the relationship between the tax and impure altruism. Consequently, two important implications arise. First, if we consider the effect of taxes on dispositional altruism, then identifying subpopulations with different distributions of θ and targeting them with different optimal taxes can be eventually welfare enhancing. Alternatively, if we consider the effect of taxes on situational altruism induced by nudges, then the optimal taxes should be responsive to the behavioral policies being implemented. Also, while in principle, it would be possible to implement a tax system that induces truthful revelation of private information about one’s disposition for altruism along the lines of Mirrlees (1971), in practice such a tax policy would likely require large information rents and would be unlikely. Our tax results, therefore, also provide guidance regarding the size of the welfare loss, when tax policy cannot account for the altruism of the population. In our framework, it is also worth noting that given a choice between a tax and a nudge, tax is generally preferable. Assuming the government revenue is efficiently allocated, the tax costlessly reduces the externality, while the nudge imposes additional psychic costs.

6 Conclusions

Ideally, informational nudges can be used as tools to induce pro-social behavior at the individual level and to reduce negative externality at the public level. The ultimate question is whether or when the nudges are effective. When nudges fail to induce high situational altruism, they are less effective or can even have counterproductive moral licensing effect. But even when nudges do induce high situational altruism, the pro-social impact of a nudge can still be foiled if the intervention targets low disposition individuals. Furthermore, the interaction between situational nudges and disposition serves as an additional explanation for the heterogeneous effects. An effective nudge has a larger impact on people inclined to be bad if people view situation and disposition as substitutes, but the nudge has a larger impact on people inclined to be good if people view situation and disposition as complements. Our conclusions can also be extended to environments with multiple goods or multiple domains.

From a government standpoint, taxes serve as the standard tool used to enhance social welfare given externalities. We show that the optimal tax level is dependent on the distribution of preferences for altruism (both dispositional and situational) as well as the interdependence between people’s altruistic preferences. We show that limiting the tax base by targeting taxes on certain subpopulations may be welfare enhancing.

Finally, the relationship between situation and disposition is unobservable and relatively unexplored. So is the relationship between taxes and impure altruism. This paper aims to fill the theoretic gap in the literature for the policy implications of nudges and to provide a framework for future empirical research uncovering these relationships.