After following the procedures above, our sample consisted of 89 published and unpublished articles that contained a total of 92 studies reporting the results of behavioral change interventions based on social norms in field settings. As already discussed, the studies are journal articles or unpublished manuscripts that explicitly use a social norm approach and that measure some type of behavioral outcome.
Regarding data collection methods, the vast majority of studies used surveys and questionnaires. Only four studies used qualitative methods, while the rest used quantitative. Regarding outcome measures related to behavior, over two-thirds of the studies (70%) used self-report measurements only, while less than one-third (30%) used some kind of “objective” recording of behavior (either traces or direct). Additionally, around half of the studies used a wide variety of non-behavioral outcome measures, the most common of which are normative perceptions, risks and consequences of behavior, motivation, intentions or self-efficacy, and intervention exposure or relevance.
3.1. Situated and Remote Interventions
A first fundamental distinction in the way that researchers apply behavioral change interventions based on social norms is the one between those interventions that are situated, applied at the point of delivery of behavior (in situ), and those that are applied remotely, in a different context from the one in which the target behavior happens. As Suchman [82
] noted, action is situated, and so is cognition [83
], therefore it is likely that interventions situated where the action take place might have a different impact from remote ones. To use a typical example in our sample, it’s not the same to distribute fliers with normative information in the context and moment in which the target behavior is happening (for example through coasters and table mats in bars to tackle drinking behaviors—see Moore et al. [84
]), compared to distributing the same information remotely, one week before in an unrelated context (for example through an email or a postcard mailed at home—see Schultz et al. [15
Social norm interventions that are applied by researchers on the context where the target behavior happens (situated) rely on the idea that immediate contexts have some influence over behavior. Consequently, modifying in some way the context (which includes the physical environment but also the information that is available to people and the interactions that happen on it) should lead to behavioral changes. As already mentioned, researchers in psychology have long explored how the effect of social norms on behavior is not uniform over different contexts or times. In their famous “Focus Theory of Normative Conduct”, Cialdini, Reno and Kallgren [42
] argue that “norms should motivate behavior primarily when they are activated (i.e. made salient or otherwise focused on)” [42
]. Similarly, Aarts and Dijksterhuis [44
] explore the concept of situational norms, showing that certain environments can automatically activate normative behavior, especially when (i) goals to visit the environment are active and (ii) strong associations between environment and normative behavior are established. Interventions that change local contexts have been shown to have important effects on transforming behavior. Among many examples, Lahlou and colleagues [85
] showed how in-situ interventions to increase the water intake of children at home can achieve increases of more than 700% in the target behavior by leveraging the social (via a social support forum), physical (by providing water bottles) and psychological (providing training) determinants of behavior.
In our sample, less than half of the studies (39–42%) applied interventions on the context where the target behavior happens. Of these, around half (21–54%) were applied only in situ, while the other half (18–46%) were combined with remote interventions (away from the context were the behavior happens). Most interventions in this group were conducted in the U.S. (24–61%), targeted alcohol-related (10–26%), pro-environmental (11–28%), or health-related behaviors (8–20%) behaviors, and were applied on students (15–38%) and residents (10–26%). Most interventions that were applied in-situ did so by introducing specific bits of normative information messages in marketing materials such as fliers, cards, letters, posters and signs. Some interventions also used a wider range of materials, including t-shirts, water bottles, coasters, and stickers, and others also included media and digital platforms such as TV, radio, magazine, and theatre ads, as well as email messages, websites, and even interactive software.
These interventions explicitly or implicitly followed the strategies of the Social Norms Marketing —SNM approach [86
], which is “a means of correcting norm misperceptions that involves publicizing (marketing) the actual rate of the misperceived behavior via the media, posters, emails, etc.” [23
] (p. 341). For example, Schultz and colleagues [15
] used door hangers and letters delivered at home to try to increase recycling and reduce water consumption in households, Lapinski and colleagues [24
] used posters in toilets to increase handwashing behaviors, and Payne and colleagues [22
] used signs in shopping carts to try to increase vegetable and fruit expenditure. In our sample, this information was often complemented with non-normative pieces of relevant information, such as the risks or benefits of target behaviors, its related rules, regulations, and existing initiatives, and tips and goals to direct action.
Apart from providing information about actual rates of behavior (and the opinions around it), interventions have also attempted to communicate that a behavior is or isn’t typical or desirable using marketing materials that are allusive to social vigilance or to broader campaigns. For example, Bateson and colleagues [33
] and Brudermann and colleagues [80
] used posters with images of eyes to influence littering and payment behaviors, and Paluck and Shepherd [35
] used wristbands to express support for a broader social norms campaign against harassment.
Rather than relying only in these normative effects, some researchers also complement normative information with other mechanisms that support behaviors and interactions in the context where behaviors happen. In our sample, one way they do this is by supplying objects that support a target behavior (if the goal is to increase its occurrence). For example, Flüchter and colleagues [17
] provided bikes to try to increase cycling to work, and Lahlou and colleagues [85
] provided water bottles to increase water intake by children. In other cases, researchers also created interaction and social control processes with their interventions, including digital forums to discuss normative information and provide mutual support [85
], mime-artists that made fun of traffic offenders and praised norm-compliers [6
], or thumbs-up and thumbs-down cards for people to express approval or rejection of observed behaviors on the street [6
Of course, some of these mechanisms to support behaviors, interactions, or control aren’t linked to social regulation processes per se. Nevertheless, as happens in the studies of our sample, they can transmit powerful normative messages as well, or make the norm salient, and as such they can have powerful effects on social norm perceptions and more broadly in social regulation processes. Specifically, they actively support social control, they make visible the preferences and opinions of group members, and they communicate that someone is willing and able to act in favor or against a change initiative.
Other researchers that attempt to leverage social norms to change behavior in the real world do not apply their interventions in the context where the behavior happens, but rather in different and often un-related contexts. These interventions are based on the idea that, if people’s behavior is influenced by their “subjective perceptions of the norm” ([3
], p. 181) which “become a reality and a guide for their own behavior, even when the perceptions are inaccurate” ([3
], p. 183), then changing those perceptions could change behavior in an enduring manner. In practice, this means that providing people with certain pieces of relevant normative information can ultimately persuade them of acting differently in the future (whenever they encounter the context where that norm and that behavior is relevant). Researchers have extensively used successful remote interventions to change behaviors, for example by providing normative information via a web-based survey that is related to decreases in overall alcohol intake [37
]. While the connection to relevant behaviors and situations might sometimes be more difficult to achieve through these interventions, their advantage is that participants are usually much more focused on the intervention mechanisms than when performing the target behavior at the same time. Another advantage of this approach is practical: it is easier to give a message to subjects where one can easily gather them, than to install the message in all the sites where the behavior may occur. That is especially the case if, for example, those making the intervention are academics and their target is students.
As a matter of fact, most interventions in our sample (66–72%) applied remote interventions, with the majority of them (53–80%) not including any direct action on the target context. This means that while around half of situated interventions included some remote action as well, only a fifth of remote interventions include some situated initiative. As for their basic characteristics, most remote interventions were also conducted in the U.S. (48–73%), targeted alcohol-related behaviors (48–73%) and were applied on students (52–79%). Although they show similar characteristics to situated interventions in this respect, most remote studies were concentrated in these three categories (compared to situated interventions, which were distributed more evenly).
Here also, the most common way in which remote interventions were applied is by providing participants with normative information messages. Along with widely used fliers, letters, ads, and other marketing materials including very similar messages, a large number of studies used email communications and web-based surveys to apply a Personalized Normative Feedback—PNF [40
] approach. PNF is a popular intervention method that attempts to correct “norm misperceptions” by “collecting participants’ self-reported incidence of some behavior and their perception of the incidence of this behavior among their peers and then providing them with the actual incidence of the behavior” [23
]. In our sample, it was mostly used in interventions focused on alcohol consumption among students, where participants received an email, completed a survey, and then received feedback about their norm perceptions and about their consumption levels (which were then compared to the ones of a reference group).
And just as with situated interventions, remote initiatives attempted to communicate normative information in ways that are different from the usual group’s rates of behavior and opinion. For example, some authors used face-to-face interaction of experimenters with participants to discuss normative information [19
], while others used videos with real-life stories [92
]. Also, in a similar way to situated interventions, studies very often complemented this with the same type of non-normative relevant information (e.g. why this behavior is “better” than the alternatives).
In addition to remote interventions transmitting information to participants, some studies also generated discussions between participants about normative information they had previously provided. This was also complemented with discussions about other relevant information and personal experiences. For example, discussion topics led to participants’ “methods to keep themselves safe in party environments” ([93
], p. 6), “accurate information about the prevention of HIV transmission” ([94
], p. 443), or “explanations for men’s perceptions of false accusations of assault” ([36
], p. 724).
And finally, some remote interventions in our sample were also complemented with mechanisms designed to support or facilitate target behaviors through the embodied competences [27
] and motivations of participants. In this respect, the most popular one included training and skills building sessions that were relevant to the target behavior. This was sometimes done with participants, but also with key actors or referents that might influence the rest of the group. For example, Mogro-Wilson et al. [92
] taught drug refusal skills to participants, Lahlou et al. [85
] conducted online coaching sessions on water intake benefits, and Paluck and Shepherd [35
] trained social referents (defined as “highly connected and chronically salient individuals in a community” [35
], p. 899) to write and perform drama skits about common types of harassment in a school assembly.
There are also examples of interventions that generated commitments to action in participants or imposed relevant financial incentives and penalties. For example, this included students signing a contract with strategies to reduce smoking [95
], monthly letters to hospital employees giving small incentives for buying healthy food at the cafeteria [21
] and imposing fines to households that consumed significantly more water than average relative to their occupants [6
]. Finally, one intervention aimed at reducing alcohol and drug consumption among adolescents also embedded normative information in psychological therapy sessions for participants [96
In summary, the first dimension opposes what seems to be a majority approach of attempting to transform internal perceptions away of the contexts where the target behavior happens (remote), to a less popular approach of modifying the environments in which those behaviors actually happen (situated). Despite their differences, the vast majority of both types of interventions attempted to transform behavior by either distributing pieces of normative information messages, or by creating interaction processes between participants. We now turn to exploring this second relevant dimension in the way in which interventions are applied.
3.2. Interventions Based on Group Summary Information and on Exposure to Behaviors and Opinions
The second fundamental distinction we found in the way that researchers apply these interventions has to do with the way in which they attempt to communicate normative information (the typical or desirable norms of behavior that are meant to produce the change). On the one hand, most interventions chose to give participants a persuasive message about the behavior or the opinions of the group (group summary information) [3
], in some statistical format such as the “Four out of five college students wash their hands EVERY time they use the bathroom” in Lapinski, Maloney, Braz, and Shulman ([24
], p. 27). On the other, many interventions also attempt to transmit this same information by creating the conditions for participants to be exposed to other people’s behaviors or expressed opinions. This includes, for example, generating collective discussions between participants in the spirit of Lewin’s [97
] seminal work [36
], representing drama skits with personal experiences or simulated situations [35
], having social referents or well-known figures publicly endorsing or rejecting target behaviors [35
], and even creating public demonstrations [6
]. In a way, this dimension is about “cold” (anonymous information only) vs “hot” (involving exposure to behaviors and opinions) communication of the norm.
A similar distinction is defined by Tankard and Paluck [3
] as the difference between normative information that is communicated through “summary information about a group” [3
], and that which is acquired by perceiving group member’s “behavior or expressed opinions” ([3
], p. 185). Researchers have explored extensively how receiving summary information about the behaviors and opinions of a group can shape the behavior of individuals. Reading simple messages about “how many people”, “how often” and “how positively my group feels” [3
] (p. 189) about specific behaviors has been shown to have important effects on individual normative perceptions and behaviors, as many landmark social norm experiments have shown [18
]. These “argument-based messages” ([99
], p. 113) are cognitively-oriented modes of influence [99
] close to paradigmatic modes of thinking [101
Although they are popular in the general psychological literature about persuasion, argument-based messages are also “at odds with lived experience” ([99
], p. 113). In the case of social influence and normative processes, most of the information that shapes our normative perceptions in everyday life comes from our interactions with others and from paying attention to their behaviors and opinions. Researchers in this respect have focused on how there are certain individuals that are especially influential over our perceptions of norms [35
]. These social referents, as they call them, are “highly connected and chronically salient actors in a group” ([35
], p. 899) that are weighted more heavily in their behavior and opinions when people “form their impressions of the norms of their reference group” ([3
], p. 185). These referents can be real group members, but they can also be imaginary or role-models as well. Researchers have applied effective interventions based on these ideas, some even successfully changing the behaviors of a whole group by targeting social referents only [35
]. This is supported by Kareeev’s [102
] research on how limitations of working memory mean that people usually rely on small samples of 7 ± 2 elements to make inferences about whole populations. In our case, for example, this means that when a person sees a group member around her recycling or expressing favorable opinions about it, she will infer that “recycling is typical and desirable for her group” ([3
], p. 184).
As happens with remote interventions, group summary information interventions also seem to have a practical advantage: it is easier to design interventions that deliver information only, rather than setting up interaction contexts where the subjects are directly exposed to influence or persuasion by other group members. But while many studies assume that the two types of interventions are equivalent, they are based on different conceptions of social influence and persuasion processes. Moreover, they can often convey very different normative information, as statistics that describe average or majority trends in a population can differ greatly from the effects of confronting actual individuals that will likely hold different positions, as Michaeli and Spiro´s [103
] research points out (As Reviewer 2 rightly argued, for example, "if half the population ranks the acceptability of abortions on a 1 to 5 scale as ’1’ while the other half ranks it as ’5’, the group-summary statistic that reports an average acceptability of 3 could have a totally different effect on behavior compared to exposure to individuals from both sides of the debate, which could lead, in some cases, to the emergence of biased norms”).
Interventions that are applied using group summary information rely on the idea that people’s perceptions of the norm can be transformed by providing them with information related to the rates in which group members engage, approve or disapprove a specific behavior. Maybe because of how simple it can be to produce and transmit these messages, and because it is “in some ways the most straightforward manipulation of a perceived norm” ([3
], p. 189), this type of approach is by far the most widely used in our sample when applying social-norm interventions (82 studies—89%).
As the studies in our sample show, interventions based on group summary information tend to be somewhat effective in transforming a very wide range of behavior in real-world contexts. Nevertheless, around one-fifth (17–19%) of the studies in our sample (most of which used group summary information) did not find effects on behavior linked to social norms. In the broader psychological literature, there is evidence supporting the idea that interventions based only on giving normative information can be ineffective to change behavior [104
], and there is even evidence that message content or quantity of information can be unrelated to the effectiveness of social norm interventions [24
]. On the other hand, combining normative information with other change tools (such as awareness of the problem, tools for action and prevention and a written promise- see Elias-Lambert and Black [108
]), as well as making people conscious about the frequency in which they engage in a behavior and providing feedback about it [109
] have both proven to be effective. Indeed, the vast majority of interventions in our sample mixed normative information with other mechanisms such as discussions, factual or context information, or calls for action and proposed goals.
In our sample, most studies (82–89%) relied in one way or another on delivering “argument-based” [99
] messages with group summary information. This amounts to persuasion (using the logical, argumentative route) rather than influence. Most of those interventions (61–66%) used only this type of information without deliberately exposing participants to the behaviors or opinions of others. Because of this distribution, the profiles of the interventions that use summary information are very similar to those of our whole sample.
The two most common ways in which these interventions were applied in our sample was either by disseminating group summary information to the target participants as a whole (a strategy close to the Social Norms Marketing approach) or by adding to this information some feedback about how the participant´s individual behavior or perceptions compare to them (a strategy close to Personalized Normative Feedback). As was reviewed above, this included both situated and remote interventions. For example, studies that use the former strategy typically use marketing materials (ads, fliers, posters…) to distribute messages like “Most Northwestern Montana’s Young Adults (88%) Don’t Drink and Drive” ([111
], p. 869). On the other hand, studies in the later group use more personalized communication channels (especially emails linked to web-based surveys) to provide graphs and messages like “ the number of occasions you drank was 4 times a week… You told us that you believed that the average student drank five times a week … The actual drinking norm for students at the University of Washington is 1.5 times a week … you drink more than 91% of other college students” ([89
], p. 436).
Researchers seemed to choose opportunistically the materials that worked better for their specific context and hypothesis: for example, signs were placed in shopping carts to increase produce demand [22
], or coasters and glasses were marked with campaign information to reduce drinking [84
]. Also, while some researchers relied on a single message and material, for example by only displaying normative feedback after completing a web-based survey through PNF [109
], others combined several promotional materials that went from fliers and posters, to coasters, stickers, glasses, and meal planners [84
Also, when applying these interventions, researchers often manipulate the way in which these messages are presented in order to test different effects on participants’ perceptions. For example, some researchers have explored the differences between using descriptive or injunctive norms [98
], or how private or visible the target behavior is [24
]. And just as with situated and remote interventions, researchers combined these normative messages with other actions that support or enable the target behaviors, such as providing non-normative information, tips for action, or different objects, for example.
In addition to providing generic summary information, some interventions include different resources to allow participants to collect information about behaviors and opinions of others so they can experience social norms through personal interactions [112
]. This is the first source of normative information described in Tankard and Paluck’s [3
] research, and is based on the idea that “individuals’ subjective perceptions of norms are not derived directly from a comprehensive survey or a census” [3
], but rather from “their unique and local experience” [3
]. Instead of just providing a message with rates of behaviors or opinions, these interventions are based on creating interactions between participants which make visible the way group members are acting, their opinions on a topic, or their willingness and efforts to change (which arguably is how most normative information is transmitted and enforced in everyday life—see Paluck [112
] and also the “vigilante effect” in Lahlou [27
] p. 126). Interventions based on making visible certain group behaviors and opinions have obtained important results in city-wide scales, for example managing in a 7-million-people city to reduce indicators like homicides by 70% [113
], deaths in traffic accidents by 65% [114
], and per capita water consumption by 46% [115
] in less than a decade (some of them included under Mockus [6
Around one-third (29–31%) of the interventions in our sample were based on direct exposure to the behaviors and opinions of others. Most of those (21–72%) did this in addition to generic summary information, were conducted in the U.S. (20–69%), targeted alcohol-related behaviors (14–48%), and were applied on students (17–59%).
In our sample, the type of actions used by interventions in this group included collective discussions between participants, face-to-face interactions and workshops between researchers and participants, and online support forums, but also theatre skits, videos, law enforcement, selling of bracelets by social referents, cartoons, and even public demonstrations. For example, Balvig and Holmberg [95
] used discussions between pupils of a school about their own normative “misperceptions regarding cigarette smoking among their peers” [95
], McCoy and colleagues [39
] used a publicly displayed poster in a clinic for patients to paste a sticker when they attended three consecutive antiretroviral therapy visits, and Agha and Van Rossem [94
], Mogro-Wilson et al. [92
], and Paluck and Shepherd [35
] used videos or theatre skits portraying relevant personal experiences and real-world situations.
These varied actions made visible three main types of information: group member’s opinions, group member’s behavior, and group member’s willingness and efforts to achieve a certain change. For example, interventions based on collective discussions and forums between participants make visible their approval or disapproval of certain behaviors (desirable behavior), and in some cases also their reported behaviors on the matter (typical behavior). In a similar manner, theatre skits, videos, and other materials that make visible participants’ behavior, their personal experiences, or simulated stories (such as cartoons–see Lapinski [24
]), all exemplify behavior. Finally, actions such as law enforcement, social referents endorsement of the change interventions (through using and selling campaign bracelets, for example—see Paluck and Shepherd [35
]), and public demonstrations, all show that people in the group are willing and actively trying to change the target behavior.
In summary, this second dimension describes two different ways in which normative information, which describes the typical and desirable behaviors in a group, is communicated to achieve behavioral change. On the one hand, the most popular approach is based on providing summary messages through a wide variety of marketing materials (group summary information), while, on the other, a less popular approach relies on creating direct interactions between participants that make visible their behaviors or opinions (exposure to behaviors and opinions).
3.3. How are These Dimensions Combined in Interventions?
Of course, the two dimensions described above are not applied independently, but are rather combined in very different ways in order to achieve their desired effects. Some studies used both remote and situated interventions, and others used both group summary information and exposure to behaviors and opinions. By representing these dimensions in a simple Cartesian system presented in Figure 3
, we propose an analytic framework that can help understand and characterize how normative information is transmitted in behavioral change interventions based on social norms.
As displayed in the diagram, where the squares are proportional to the number of sampled interventions in that quadrant, interventions that use group summary information are much more popular in the literature, especially those that do it away from the context in which the target behaviors are performed. This distribution is not surprising, considering the lower difficulty of creating normative messages compared to creating more complicated situations, and of broadcasting it where participants already are or can easily be assembled instead of the contexts where the behavior happens. This first result suggests that there is a practical bias in intervention design: easier interventions are more popular.
Exposure to behaviors and opinions, which arguably is how we collect most normative information in our everyday social interactions [3
], has been much less explored in academic research. More than half of studies (54–57%) used remote interventions only, more than two-thirds (67–71%) used only group summary information. Very few interventions (11–12%) were applied in-situ and using exposure to behaviors and opinions, while even less (5–5%) opted to test all four quadrants of our framework.
In addition to that, the studies in these four quadrants used a wide range of intervention mechanisms that leveraged social, psychological and physical determinants of behavior to achieve their ends. In our sample, we found 16 mechanisms that studies used to achieve the desired effects, and that complement the ones described by authors like Miller and Prentice [23
], Paluck [34
] and Bicchieri [2
]. These are presented in Table 2
according to which layer of installation theory they target.
In order to explore further how these interventions were applied and their characteristics, we conducted a Multiple Correspondence Analysis—MCA to explore what experimental characteristics tend to be combined in our sample. MCA is a technique to analyze the pattern or relationships between different categorical variables which produces the principal components that account for the variance in the data [77
]. The first two or three components often account for most of the variation in the data (or inertia), and this produces a map in which proximity denotes that certain modalities of variables tend to occur together.
To conduct our MCA, we included the two dimensions described above (situated/remote/Group Summary Information/Exposure to Behaviors and Opinions), together with the most popular (specifically, those with 10 or more occurrences in the sample) intervention mechanisms in Table 2
and the information on whether the study registered significant effects on behavior linked to social norms. We also included information on the target behaviors, types of participants and year of the study as supplementary (passive) variables. The results show two main components that account for 61.5% and 12.5% of the variance respectively, as the simplified diagram in Figure 4
Because of simplicity and space considerations, only the general associations that were drawn from the analyses will be reported here at the risk of losing some of their details and nuances. The complete MCA output and coordinate plot can be found in the Supplementary Materials
(Figures S1 and S2
respectively). Two general conclusions emerged from this analysis:
The main axis, which accounts for most of the variance in the sample, opposes “lighter” interventions applied remotely, using GSI and digital platforms mainly on students (which as we have seen constitutes the bulk of our sample and is associated with quadrant III), with more complex interventions done in-situ, using exposure to behaviors and opinions, and relying on a wider variety of interventions mechanisms (associated with quadrant I). Characteristics associated with each of these groups are presented in Table 3
When including only studies that recorded behavior (excluding self-reported measures) finding relevant effects on behavior appears more frequently among interventions in the group 2 (quadrant I). On the other hand, not finding effects is more frequent among interventions in the group 1 (quadrant III). Nevertheless, these results should be taken with caution, as they are based on a limited subsample of 28 studies (in which 25 found effects and three didn’t), and there is a great diversity of experimental contexts, targets and treatments in them.
Due to the low number of studies in this subsample that are linked to quadrants I and III, and because of the configuration of the main axis in the MCA, we hypothesize that these associations might be linked to the opposition between remote and situated interventions, as displayed in Table 4
. This of course must be systematically tested based both on previous literature and new experimental data in order to assess its plausibility.