FIGURE 5.1 The experimental interface (social information treatment)
VISIBILITY VERSUS SOCIAL INFORMATION
In the summer of 2014, social media were peppered with video clips of people sitting in gardens and courtyards having buckets of iced water poured over their heads. Before they did so, they nominated three other people to undergo the same ordeal or donate an unspecified sum to a charity benefitting amyotrophic lateral sclerosis (ALS), a subtype of motor neurone disease, or do both. This was the Ice Bucket Challenge—celebrities and politicians took part, but so did civilians, in droves. There were 2.4 million tagged videos circulating on Facebook alone in August,1 and more than 14 million videos overall,2 while hits to the English Wikipedia’s article on ALS grew from an average of 163,300 views to 2.89 million in August, with similar increases in other languages. The ALS association received over $100 million in additional donations, while the Motor Neurone Disease Association received an additional £7 million. Although a number of criticisms were made of the Ice Bucket Challenge, with claims of ‘slacktivism’ for those who did not donate and ‘funding cannibalism’ for those who did, in terms of the opportunity costs for other causes, it did represent a mobilization with social good aims, where large numbers of people used social media to raise awareness of and generate funds for the treatment of an important health issue.
The form of social influence at work here was visibility. For the thirty or so seconds of every video, participants were highly visible to their social networks and beyond, undertaking a charitable act. Conversely, the fact that they had not done so after being nominated would be visible at least to their own social networks. This is the strongest form of visibility discussed in Chapter 2, where people’s names and faces are clearly associated with their actions. Social information, the other form of social influence exerted by social media that we identified in Chapter 2 and discussed in the last chapter, played less of a role. When deciding whether to take up the challenge, nominees would not know how many other people had done so, other than their nominator, their friends whose videos they had seen, and some high-profile celebrity Ice Bucket Challenges (perhaps those of George Bush, Bill Gates, and Donatella Versace, flanked by two shirtless assistants). So what sort of social influence works best for encouraging collective action? Having identified and examined social information effects in the last chapter, we now extend our analysis to include visibility, comparing these two forms of social influence on both individual participatory decisions and collective capacity to contribute to public goods. Visibility on social media comes in part from the ease with which individuals can exhibit themselves, share information, or express opinions that have at least a chance of being viewed by large numbers of people, stretching far beyond the individual’s own social networks. Even people who are passive rather than active users of social media are more likely to be visible to others through the ease with which any kind of information about them can be found using search engines both inside social media sites and on the Internet more generally. As discussed in Chapter 2, visibility is likely to be a factor at work when individuals are deciding whether to contribute to collective action in a social media environment.
As for social information, previous research has shown visibility to be an important influence on individuals when they decide whether to participate, eliciting higher collective action responses. For example, Gerber et al. found that telling people that their neighbours will know whether they voted or not (as well as giving them social information about their neighbours’ voting behaviour) increases the effect of a vote mobilization treatment,3 a finding confirmed in other studies.4 Likewise, as we demonstrated in the last chapter, providing potential participants with social information about how many other people are participating has been shown to increase other types of participation, although the extent of social influence depends upon the level of participation reported. Most existing research either treats these two social influences in isolation (that is, looks in depth at either visibility or social information) or follows a research design that makes it difficult to separate out the effects. Gerber et al., for example, gave voters social information at the same time as telling them that their voting behaviour would be visible.5 Using an experimental design that tests for both social influences allows for a comparison of their relative impacts, as we do here.
In this chapter we first review previous work on visibility and collective action and lay out our expectations for the relative effects of social information and visibility. We then report a public goods laboratory experiment in which subjects were invited to contribute to a range of collective goods in randomly ordered rounds. These rounds were subject to three experimental conditions: visibility of individual contributions, social information about contributions within groups, and a control group where visibility was not applied, nor was social information received. By exposing experimental subjects to these conditions, we hope to simulate a social media environment and estimate the absolute and relative effects of both forms of social influence. We report the results and discuss the implications both for our understanding of social media influences on collective action and for the design of platforms geared at encouraging civic engagement and campaigning.
FAME AND SHAME: HOW VISIBILITY WORKS
Why should visibility influence participation? The explanation is based on the concept of social pressure whereby the fact of an action being observed can invoke two distinct mechanisms: shaming and prestige seeking. Indeed, social pressure may be defined as ‘communications that play upon a basic human drive to win praise and avoid chastisement’.6 Research using large-scale field experiments has shown that citizens will be more likely to vote when they are informed that their behaviour is public and that it can or will be monitored or publicized due to a desire to appear compliant to social norms.7 ‘Shame aversion’ for failing to vote also boosts compliance levels significantly.8 Research using the idea of ‘image motivation’ describes how citizens may be motivated by how others perceive their behaviour, seeking their approval. Ariely et al. elaborate two dimensions of image motivation: ‘Actions that imply personal traits such as being pro-social, fair-minded, or caring yield a positive image, while actions that imply personal traits such as being unfair or greedy reduce a positive image or even result in a negative image’.9 Their laboratory and field experiments demonstrated that if participants liked the prestige they got from having their contributions made public, they acted in a more pro-social manner.10 Similarly, when donations are advertised in categories (e.g., gold, silver or bronze donors), people will more often give the minimum amount needed to appear in a higher category.11 A field experiment seeking to test the effect of a pledge to donate books for charity found that listing participants’ efforts increased the level of donations.12
Both harsh shaming tactics and soft applications of visibility seem to have comparable effects in the various experiments that have been carried out, mostly focusing on voter turnout.13 Panagopoulos found a shaming intervention to increase voter turnout to be more effective than simply listing out the names, but also showed that simply thanking voters can improve turnout by implying that someone has observed whether they voted or not.14 Panagopoulos divides such interventions into two categories:15 negative social pressure, which activates shameful feelings,16 and positive social pressure, such as positive feedback, reinforcement, or recognition, where the intention is to engender positive feelings.17 Some of these studies explore the effect of positive social pressure on different demographic groups. The most recent finding appears to show a similar motivational effect on both higher propensity voters and those in groups with lower propensity for voting, such as Hispanics and African Americans.18
Overall these studies reveal that whichever mechanism is enacted, making the pro-social actions visible to others is likely to increase contributions, leading us to predict that in the laboratory experiment we report next, visibility will increase the amount that individuals contribute. The difference made by visibility on propensity to vote in the field experiments discussed above is significant. Gerber et al., who used both negative and positive social pressure, reported an increase in turnout of around 4 percent,19 while Panagopoulos, using positive social pressure, reported increases in turnout of 2 to 3 percentage points on average compared to control conditions.20
In the previous chapter, we identified the effect of varying levels of social information on people’s decisions whether to participate, finding evidence that high levels of social information make an individual more likely to participate, while low and medium levels do not. Likewise, in Chapter 3 we showed how evidence of relative popularity makes people more likely to sign petitions, at the expense of petitions for which there is no such evidence of popularity.
BEING SMART WITH SOCIAL INFLUENCE
What would we expect to find when we compare the effect of social information and visibility? The need to coordinate individual actions to provide collective goods means that the effect of social influence could vary depending on whether the unit of observation is the individual or the collective. At the individual level, the research outlined above would suggest that visibility has a greater effect, given the effect sizes observed in experiments that have explored the impact of visibility, compared to most of those observed for social information in Chapter 4.21 Such an impact comes in part from the fact that we might expect a given level of visibility to have a more uniform effect across different points in a mobilization, while Chapters 3 and 4 suggest that the provision of social information will have different effects at different stages. For example, people may be negatively affected by low levels of participation in the early stages of a campaign to raise funds for a new scanner at their local hospital when it provides no signal of viability and they feel their resources will be wasted if the initiative is not going to get off the ground. Likewise, news that enough contributions have been made for the scanner to be provided could discourage participation, because people could feel their contribution was no longer needed. In contrast, information that the amount of funds raised was just below this point might give participation an extra boost as people feel their contribution will really make a difference.
In short, the provision of social information allows participants to use their resources of time, effort, and money more efficiently. Such causal processes may not be observable when contributory behaviour is viewed at the individual level. But at the collective level (and in the rounds in our experiment), under social information we would expect the threshold for success to be met more efficiently, given the incentives to act strategically, particularly when the provision point is nearly reached, or has already been reached and further contributions will not benefit the collective good. Under visibility, particularly when it acts through some kind of shaming mechanism, the total contribution amount may exceed the threshold or resources may be devoted to non-viable initiatives, leading to inefficiency. We would expect visibility to have a greater positive effect than social information on the amount that individuals contribute, while social information will have the same or greater positive effect as visibility on the likelihood of a round being funded in our experiment or a collective good being provided.
APPLYING SOCIAL PRESSURE IN THE LAB
To test our predictions we use data from a public goods game laboratory experiment undertaken by the authors, which also provides data for the two chapters that follow. Public goods games are a common method of experimental economics to model collective action problems. Subjects are provided with tokens that they contribute to a public pot, which is shared between participants at the end of the game, according to the payoff structure that forms part of the design. Subjects choose whether to cooperate by contributing tokens or whether to free-ride and maximize their own benefits. The information environment in which they do so can be manipulated as part of the experiment, while all other conditions are controlled in the laboratory; so, it is a good way to test the different forms of social influence that we are interested in here. However, laboratory experiments have limited external validity (that is, the difficulty of generalizing from a laboratory simulation to the outside world),22 which we have attempted to mitigate through our research design, as detailed below.
The experiment took the form of a one-shot public goods game in which 186 subjects were invited to attend a series of sessions in the laboratory, with between 14 and 20 in each session. In each session, subjects were seated at a computer screen (see Figure 5.1, which shows the interface), provided with a number of local public goods scenarios (one at a time), and asked to contribute tokens to each good in a series of twenty-eight rounds, which they played anonymously in small groups (7–10). Participants were randomly reallocated into two groups at each round, so that subjects never interacted with exactly the same group and did not know which group they were in, so there was no opportunity for learning the behaviour patterns of other participants.
At each round, subjects were shown one of six different scenarios (Figure 5.1 shows an example), phrased as a request to fund a local initiative, namely a local campaign for bicycle lanes; a local fund-raising initiative to send aid to victims of an earthquake in Haiti; a health campaign to save the local accident and emergency department; a local campaign to provide free public Wi-Fi; a group effort to clear snow for elderly neighbours; a campaign for more street lights to be installed by the council; and a collective effort to organize a street party (full wording of the scenarios is shown in the Appendix). These scenarios were chosen to capture elements of collective action around different types of public goods (donating money to Haiti is a different kind of public good from campaigning for street lights, for example, because participants receive direct benefit from the latter but not the former). By using scenarios (in contrast to most experimental economics research),23 we tried to approximate real-world situations while preserving the incentive structure of a public goods game.
Subjects were endowed with ten tokens at the start of each round and could donate an amount from zero to ten to the collective effort in each round. They were informed that a fixed bonus would be redistributed among all participants in the group if the provision point—corresponding to 60 percent of the maximum amount collectable for that round—was met (the Appendix shows the instructions to participants, including a table of payoffs for each contribution). Subjects could keep any remaining tokens left to them after they had made their contribution to a round. So when the provision point was reached, those who had contributed nothing would receive the maximum payoff of £15.00, while those who had contributed all their ten tokens would receive a payoff of £12.50. If the provision point was not reached, those who had contributed nothing would receive £8.50 and those who contributed the maximum amount would receive £6.00, the lowest possible payoff. Subjects were notified that their payment was made according to the result of one randomly selected round, which is a design that has been shown to impel subjects to treat each round as if it were the only one.24 As the groups formed in each session were different sizes, the provision point was adjusted to meet 60 percent of the maximum amount that could be collected. Each round was planned to last fifty seconds; but to avoid deadline effects, if there was activity during the last five seconds of elapsed time, the round was extended by another five seconds.
Two treatments were implemented in addition to a control condition and all subjects would experience all conditions more than once in the course of their session. In control rounds, the subjects received no social information about the contributions of other subjects, and their own contributions were not visible to other subjects. Under the social information only treatment, subjects were shown real-time information on their computer screens about the total amount raised within their subgroup, and the number of individuals in their group who had already contributed (see Figure 5.1). This information was updated continually in real time throughout each round, as contributions were made; when the round was over, the information was set back to zero. Under the visibility only treatment, subjects were shown real-time information about the amount the individual participants in the room were giving on a screen at the front of the laboratory, showing the desk plan of the room (so contributions were highlighted on the screen as they were made). They were not told, however, which contributions were from participants in their own subgroup, which would have provided them with social information and contaminated our aim to keep the two treatments separate. Under this treatment therefore, all subjects knew that their contribution amounts would be visible to all other subjects as a number against their position in the laboratory and thereby identifiable to themselves. Taken together, these conditions ensured there was no possibility that reputations based on repeated interactions between the same individuals could develop. The combination of these two treatments resulted in the last treatment, a combination of the social information and visibility conditions.
After the experiment, the subjects completed a questionnaire (shown in the Appendix), which included various demographic questions and a short personality survey (to allow us to carry out the analysis presented in Chapter 6). Individual contributions were not returned if they were made in excess of the provision point (i.e., no rebate), nor if the threshold was not reached in the given round (no refund). As a consequence, this public goods game has two sets of Nash equilibria, that is, two states where all participants are making the best decision they can, taking into account the decision of the other players. The first equilibrium is inefficient (in terms of providing the good) and corresponds to the case in which no individual gives tokens and the group bonus is not paid out; the second equilibrium is when the provision point is exactly met and the public good is provided.25 This experimental design allowed for both strategic and non-strategic behaviour, according to treatment. That is, under the control condition, we would not expect strategic behaviour as participants receive no information as to how likely the provision point is to be reached. Likewise, under the visibility treatment, although participants saw the donation amounts of other participants in real time, they had no way of knowing which subjects were in their particular subgroup for that round, so could not use this information to play strategically, although it may have had some conditional cooperation effects, in that they felt like contributing if they saw that other subjects were doing so.
Under the social information condition, in the early stages of the round, when no one or few had contributed within a group, there was no information available to any participants to make a strategic decision. In these contexts, an individual contribution provides information about that individual’s personality and general propensity to give (discussed in Chapter 6). Under the social information condition in the later stages of the round, there was much more opportunity for strategic play, for example contributing tokens to push the total amount of contributions over the provision point or withholding contributions because the provision point had been reached. So later contributions are more likely to reflect strategic play, and we would expect that under the social information condition there would be more chance of reaching the efficient equilibrium.
VISIBILITY GETS PEOPLE GIVING
The overall effect of the two main treatments is compared in Figure 5.2, which shows the frequency distributions of individual contributions broken down by treatment conditions (across all scenarios). The y-axis displays the frequency of each possible contribution amount that subjects made across rounds (note that the lines are constructed to join up points indicating whole contributions; it was not possible to give a part contribution). The black line representing the control treatment has a tri-modal shape. A large proportion of subjects contribute either zero or ten tokens (although these are not necessarily their consistent strategy across rounds), and others provide what could be termed the ‘fair’ share (60 percent of their endowment), which corresponds to the amount each should contribute for the provision point to be reached at equal cost to all participants. Frequency of contributions under the social information condition is represented by the dark gray line, showing that a larger proportion of participants are inclined to free ride on the contributions of others, compared to the control treatment. When contributions become individually visible (light gray line), the largest proportion of individuals give the ‘fair’ amount of six tokens. This is almost twice as many as under the social information treatment. Likewise, far fewer individuals under this treatment were likely to free ride by contributing zero tokens.
FIGURE 5.2 Frequency distribution of the normalized amount contributed by individuals across treatments
Overall, these results show that visibility will increase the amount that individuals contribute. We confirmed this result by carrying out a left- and right-censored Tobit regression on a collapsed form of the experimental data (with standard errors clustered around individual subjects), in which each observation refers to one person-round. Preliminary analysis using the information from the questionnaire showed that the contribution amount increased with the importance ascribed by subjects to the particular issue, and decreased with the size of the group in which they participated, as we would expect,26 and we used these as control variables in subsequent analyses. We excluded all rounds run under the social information treatment, which means that those held under both treatments are also excluded. The results showed clearly that visibility predicts individual contributions, with a clear significant effect (p < .01). Including variables for the scenarios increased the fit of models for the regressions (increasing the pseudo-R2 by .005 for social information and .013 for visibility), but did not change the results at all, with the coefficients changing by less than 0.1 only and never gaining or losing significance.
In contrast, when we carried out the same regression analysis for social information, we found no overall effect of social information on individual contributions. In fact, under the social information treatment participants were more likely to free ride by contributing no tokens at all, and people were less likely to contribute six tokens than the control group (Figure 5.2). So visibility seems to have a greater effect on individuals’ willingness to cooperate and contribute to the collective good than either social information or the control.
SOCIAL INFORMATION TRUMPS VISIBILITY IN THE ROUND
We then analysed the data at the round level, using ‘funded’, a binary variable that indicates whether the provision point was collectively met. Figure 5.3 shows the effects of treatment conditions on the likelihood of a round being funded. Looking at the aggregate effect for all scenarios at the right of the graph, we can see that both treatments have a very similar impact on the likelihood of a round’s being funded. Logit regressions with ‘funded’ as the dependent variable (shown in Figure 5.4) confirm this finding: both visibility and social information are significant, with similar coefficients and levels of significance. In the regression analysis, we controlled for the differences in importance ascribed by subjects to the issues involved, by using subjects’ responses to the postexperiment questionnaire, where they were asked to rate the scenarios for importance.
FIGURE 5.3 Effect of social information and visibility treatments on the likelihood of rounds being funded, shown across scenarios and for all rounds
FIGURE 5.4 Round level analysis (logits)
Note: For the top model, we included all control rounds (154) and all rounds run under social information only (266). For the bottom model, we included all control rounds (154) and all rounds run under visibility only (42). We excluded all rounds run under both social information and visibility treatments (42).
Figure 5.3 also shows variation in the likelihood of rounds being funded and treatment effects across the six scenarios used in the study. It shows that the scenarios relating to saving an accident and emergency department (under both treatments) and providing aid to earthquake victims in Haiti (social information treatment only) were considerably more likely than the other scenarios to be funded, although social information also had a strong effect for the scenario relating to street lights, while visibility had a strong effect on the street party scenario. These results show that the scenarios used were affecting subjects’ behaviour. However, the scenario effects (and the variations in the control condition shown in the graph) are to a large extent explained by the different levels of importance that subjects ascribed to the scenarios, for which we controlled in all regression analyses.
THE EFFICIENCY OF SOCIAL INFORMATION
As discussed in Chapter 2, social media exert social influence on individuals deciding whether to participate in collective action, notably social information and visibility. In this chapter, we have aimed to simulate a social media environment with a laboratory experiment to investigate the relative impact of these two forms of social influence.
We confirm the importance of visibility of people’s actions as a powerful determinant of their propensity to participate in collective action, something that has been demonstrated in offline contexts in previous research.27 We also show that the impact of social information about the actions of others is not revealed at the individual level; social information can have both positive and negative effects at different stages in a mobilization, which can cancel each other out. Our subjects appear to have behaved strategically by using social information to participate if their contribution would make a difference to the good being provided, and not to participate if the contribution to collective action would be wasted, either because the good was clearly not going to be provided, or because the good was going to be provided regardless of their contribution.
In contrast, we find that both social information and visibility have similar impacts at the aggregate or round level, which can correspond to the community or collective. The provision of social information can help communities provide collective goods just as much as making potential contributors visible, in spite of the greater positive effect of visibility at the individual level. Such a finding shows social information to be the more efficient form of social influence in terms of the collective, encouraging individuals to target their contributions where they feel they will make the most difference. Basically, although the proportions of rounds funded under both treatments are approximately the same, those run under the social information treatment were funded with fewer overall contributions than those run under visibility. Assuming that any contribution to a cause has some kind of opportunity cost, in terms of time, money, or effort, that could have been devoted elsewhere (as suggested by the findings of the natural experiment in Chapter 3), then social information emerges as the optimal form of social influence for maximising the chances of providing the public good.
Our findings are to some extent born out by the Ice Bucket Challenge, which used visibility to good effect in shaming people to partake. Visibility pushed people to opt to undertake an unpleasant and even painful act, operating through the twin influences of shame and desire for prestige, or at least attention. But visibility was also conspicuous in its absence as soon as the video finished, melting away as people contemplated donating away from the gaze of their social networks; in the United Kingdom, while one in six people took part in the challenge, only 10 percent of those went on to donate, and although this figure was higher in the United States, the majority there also did not donate. In the aggregate, as in our experiment, visibility exhibited inefficiency, with allegations that the ALS association spent most of the donations on advertising, promotion, and exorbitant salaries for top officials, while the rest went to ‘the pockets of Big Pharma’ already investing heavily in ALS research, with huge profits in sight.28
These findings could have relevance for the design of platforms geared at civic engagement, or for decisions over which social media platforms to use for which endeavour. Making participants visible as they participate (as is the case on some charitable donation sites and petition platforms) may be useful for increasing participation. But it may not be the most efficient way to do so in some contexts, such as when sufficient levels of contributions have been collected already or mobilization has already shown itself to be non-viable. Here social information can be a more valuable tool to encourage participants to behave strategically and direct their resources where they will have the utmost effect.
Likewise, the effect of visibility will be very context-dependent and needs to be adjusted accordingly. In the context we simulated in this experiment, participants (or at least most of them) clearly wanted to be perceived as having contributed. Other work has shown that the visibility of contributors’ names makes people more likely to make charitable donations, as discussed above. The Ice Bucket Challenge relied on the strongest form of visibility discussed in Chapter 2, that is where participants’ actions are visible but also the participants themselves are identified in person and linked to the action. In other contexts, as in the early stages of the Egyptian and Tunisian revolutions of 2011, for example, it might be crucially important for participants to remain anonymous while their actions remain visible. In this case, their actions constitute an anonymous form of social information that, as it scales up, may exert influence on other people considering whether to participate.
Up until now in the empirical part of this book, we have assumed that all individuals will react similarly to these different forms of social influence. However, with both social information and visibility, we might also expect various forms to exert different influences on different types of people, even within the same context. For example, while some individuals may appreciate the visibility enabled by social media, proudly displaying their ‘I voted today’ button on their Facebook page, for example, others may shy away from having their contributions made public (these people might have donated to the ALS association, but would not have participated in the Ice Bucket Challenge). We might also expect some people to be particularly susceptible to high (or low) levels of social information, with others more willing to act unilaterally, irrespective of what other people are doing. We investigate these possible sources of individual-level difference in reactions to social influence in the next chapter.
