Social Comparison and Hierarchy Leadership in Voluntary Public Goods Provision

Transcription

1 Social Comparison and Hierarchy Leadership in Voluntary Public Goods Provision Jipeng Zhang and Huan Xie Abstract This paper investigates the effect of social comparison and hierarchy leadership on prosocial behavior in a setting of sequential charitable giving. The model shows that a follower s (second mover s) response to a leader s (first mover s) giving depends on both the relative social status between the follower and the leader and the giving amount of the leader. We test the theoretical predictions in a field experiment that varies the social distance between the leader and the follower. The experiment consists of three treatments: Leader treatment, in which the followers know the leader s name, affiliation, and giving amount; Peer treatment, in which the followers are only informed of the leader s affiliation and giving amount; and Stranger treatment, in which only the leader s giving amount is revealed to the followers. The empirical findings show that the followers giving critically depends on their perceived social relationship with the leader. The followers giving in the Leader treatment is significantly smaller than those in both the Peer and Stranger treatments when the leader s giving is small ($10 or $20). When the leader s giving is large ($50 or $100), however, the followers giving in the Leader treatment is significantly larger than that in the Stranger treatment and is not significantly different from that in the Peer treatment. The followers giving in the Leader and Peer treatments responds positively to the leader s giving amount with the slope coefficients between 0.15 and 0.34, but the response in the Stranger treatment is not significant. Our structural model estimation shows that all three motives for giving altruism, social comparison, and social image concern are important in explaining the giving decisions of the followers. Key words: charitable giving; hierarchy leadership; social comparison; field experiment Jipeng Zhang: Nanyang Technological University, Singapore, jpzhang@ntu.edu.sg. Huan Xie: Concordia University, CIRANO and CIREQ, Montreal, Canada, huan.xie@concordia.ca. We thank Zhao Chen, Jimmy Chan, Yan Chen, Jaimie Lien, Yohanes Eko Riyanto, Walter Edgar Theseira, Lise Vesterlund, Maoliang Ye, Jie Zheng, and the seminar participants at the ESA Asia-Pacific Meeting, the 2013 ESA World Meeting in Zurich, Fudan University, Shanghai University of Economics and Finance, Tsinghua University, for helpful comments. We thank Jasmine Huang, Tian Hao Neo, Albin Wan, and Jia Xin Xu for their excellent research assistance. We gratefully acknowledge the financial support from Concordia University, CIREQ (Montreal), and Nanyang Technological University. 1

2 1 Introduction Social comparison has been long recognized as a driving force underneath behavioral decisions since we often compare ourselves to others for guidance or to increase self-esteem (Festinger 1954). A large literature is developed to examine the effect of social comparison on economic decision making, as summarized by Chen, Harper, Konstan, and Li (2010). People exhibit the tendency to mimic prevalent behavior or conform to social norms (Asch 1956, Akerlof 1980, Bernheim 1994). When information of others payoffs is available, people show distributional concerns, such as inequality aversion and peer-induced fairness (Fehr and Schmidt 1999, Bolton and Ockenfels 2000, Ho and Su 2009). In the theories on interdependent preferences, utility depends not only on the absolute level of consumption, but also on the average consumption (Pollak 1976), or the rank of consumption (Frank 1985, Robson 1992, Hopkins and Kornienko 2004). Social comparison induced by information provision has a direct effect on charitable giving (Frey and Meier 2004; Shang and Croson 2006). The models on social pressure assume that individuals incur a cost when their contribution is lower than the social standard (Akerlof and Kranton 2000, DellaVigna, List and Malmendier 2012). In this paper, we study how social comparison between donors with different social distances may affect their behavior of charitable giving differently. In particular, we are interested in the case when the donors possess real hierarchy positions and there is a real hierarchy leader as the first mover in the setting of sequential charitable giving. Leadership giving is widely used in fundraising campaigns (Andreoni 2006). However, the effect of hierarchy leadership is not easy to comprehend since it introduces elements that can affect the follower s giving in different directions. On one hand, it may have a negative effect on the follower s giving if the follower is concerned with outperforming their leader. On the other hand, the follower may feel obligated to contribute more if the leader sets a good example. Hence, an empirical investigation on the effect of hierarchy leadership and social comparison on charitable giving can be valuable. Our study can also generate insights on the fundraising strategy of charities that target an organization or a group of people with different hierarchy positions and social status. We develop a model of sequential charitable giving to show the effect of the first mover s (leader s) hierarchy leadership on the second mover s (follower s) prosocial behavior. The follower responds to the leader s giving by conforming to a social norm that depends on both the relative social status between the follower and the leader and the giving amount of the leader, while taking into account the follower s altruistic and social image concern. We show that the follower s contribution is larger when the marginal benefit from the warm-glow effect, the public good provision, and the social approval is larger, and when the contribution is made more public. Importantly, a larger social comparison cost may increase or decrease the follower s giving, depending on whether the follower s contribution is smaller or larger than the social norm. Empirically, it is challenging to identify the effect of the hierarchy leader on the follower s prosocial behavior because of the endogeneity problem associated with the peer effects (Manski 1993; Brock and Durlauf 2000). In this paper, we adopt the approach of randomized field experiments to create exogenous variation in the social distance between the leader and the followers. The hierarchy effect is potentially compounded with two additional effects, namely, the anchor point effect and the peer effect. We define the hierarchy effect as the influence of the leader s hierarchy position or social status on the giving of someone who has a lower rank or status within the same organization. The peer effect is defined as the effect on charitable giving when solicitation induces comparison to those socially near the solicitee or potential donors in the fundraising campaign. The effect of an anchor point is the influence on the follower s contribution with the presence of a suggested contribution level from a previous giver. When the follower is informed of the contribution amount of their hierarchy leader, this information automatically serves as a reference point and a contribution by a peer, in addition to the hierarchy effect of the leader. In order to identify the pure hierarchy effect, that is, 2

3 whether soliciting high-status individuals first and publicizing their giving information can induce low-status people to contribute more or not, we have to separate out the peer effect and reference point effect. For this purpose, we need to design a field experiment that not only varies the distance in the social status between the leader and the follower, but also keeps the same giving amount of the leader (first-mover) across treatments. To create variation in the social distance between the leader and the followers we vary the information disclosed to the followers. In the first treatment, Leader treatment, the leader s name, affiliation (department), and giving amount are all revealed to the followers. In the second treatment, Peer treatment, we only show the affiliation and donation amount of the leader to the followers, without revealing the leader s identity. Hence, no hierarchy effect influences the followers contributions. However, the follower knows that the presented contribution is from one of their colleagues in the same department since the affiliation of the leader is disclosed. In the third treatment, Stranger (Baseline) treatment, only the giving amount of the leader is revealed to the followers as a reference point. The followers do not know the leader s name and department. We conducted the experiment at a large public university in Singapore by soliciting the university faculty members for contributions to a charity that provides financial aid to low-income families supporting a family member with blood cancer. We approached department chairs (leaders) first, following which we approached other professors (the followers) in those departments for contributions to the same charity. The followers within a department are randomly assigned into the three treatments. We report the following findings. First, the followers average giving critically depends on their perceived social relationship with their leaders. The average giving in the Peer treatment is 21 and 12 percent higher than that in the Leader and Stranger treatments respectively. Second, the effect of hierarchy leadership depends not only on the existence of the hierarchy leadership but also on the size of the leader s donation. The followers giving in the Leader treatment is significantly smaller than those in the both Peer and Stranger treatments when the leader s giving is small ($10 1 or $20). However, when the leader s giving is large ($50 or $100), the followers giving in the Leader treatment is significantly larger than that in the Stranger treatment and is not significantly different from that in the Peer treatment. Therefore, the hierarchy leadership discourages the followers giving only when the leader gives a relatively small amount, but it has a positive effect on the followers giving when the leader s giving is large. Third, the followers giving responds positively to the leader s giving amount at significantly higher rates in the Leader and Peer treatments than that in the Stranger treatment. According to the theory, the estimated responsiveness of the followers giving to the leader s giving reflects the specific social norm that the followers make social comparison. Relative to the Stranger treatment, the followers in the Leader and Peer treatment give and dollars more for every one dollar increase in the leader s giving. The estimated responsiveness in the Stranger treatment is not significantly different from zero. These estimates imply that the followers tend to proportionally match the amount given by their leader or their colleagues, but not a stranger. In order to better quantify the relative importance of different motives behind the prosocial behavior of the followers in our environment with hierarchy structure, we provide a structural model estimation. We estimate the model using a classical minimum distance estimator (MDE) from Chamberlain (1984). Based on the structural parameters, we find that altruism, social comparison, and social approval are all important factors beneath the followers donations. The essential trade-off for a follower to give is between the utility from altruism and social image and the disutility from social comparison. However, the components in the model can only explain part of the donors giving decision. Other factors, such as group fixed effects and academic position, also play important roles in the determination of the followers giving. 1 For simplicity, $ in this paper refers to Singapore Dollar (SGD). The exchange rate between Singapore Dollar and US Dollar (USD) in the second half of year 2012 is 100 SGD for 80 USD approximately. 3

4 This paper closely relates to the literature on social comparison and decision making. Chen, Harper, Konstan, and Li (2010) find that effective personalized social information can increase the level of public goods provision in a field experiment on MovieLens. After receiving behavioral information about the median user s total number of movie ratings, users below the median demonstrate a 530 percent increase in the number of monthly movie ratings, while those above the median decrease their ratings by 62 percent. When given outcome information about the average user s net benefit score, above-average users mainly engage in activities that help others. There are also several papers that look at the effect of social information by conducting dictator games in laboratory. Cason and Mui (1998) study how social influence affects subjects choice in a sequential dictator game and find that observing the contribution of one other subject constrains some subjects from moving toward more self-regarding choices. Duffy and Kornienko (2010) show that ranking information has significant effects on giving in sequential dictator games. Bardsley and Sausgruber (2005) find that conformity accounts for around one third of the increase in contributions due to an increase in others contributions. Our study demonstrates that social comparison is critical for us to understand prosocial behavior in real life where people are often part of a hierarchy organization. The leader s behavior can have the opposite effects on the behavior of group members, which is very similar to what Chen et. al. have found, though our experimental setting is completely different. This paper contributes to the literature on the effect of social information on altruism and charitable giving. 2 Frey and Meier (2004) analyze the behavior of students in Zurich who had the opportunity to contribute to two social funds every semester. They find that information about the average contribution in the past has a significant impact on contribution. Shang and Croson (2006) demonstrate a positive social information effect on individual contributions in a public radio fundraising campaign. They find that the most influential level of social information is drawn from the 90th to 95th percentile of previous contributions and the effect is significant for new members but not for renewing members. Alpizar, Carlsson, and Johansson- Stenman (2008) conducted a field experiment by showing the potential donors different levels others typical contribution and find that lower reference point ($2) encourages participation but decreases conditional contribution and higher reference point ($10) increases conditional contribution only. List, Berrens, Bohara, and Kerkvilet (2004) analyzed charitable contributions to a Center for Environmental Policy Analysis at the University of Central Florida. They used three different information treatments: (i) completely anonymous responses, (ii) the experimenter knows the response, and (iii) the whole group knows the response. The largest share of yes responses was obtained when the whole group was informed of the response, followed by the case when only the experimenter knew the response. In all previous studies, there is no real social status or hierarchy leadership and no variation of social status. The main difference between our paper and this literature is that we introduce hierarchy leadership and create variation in social distance between the leader and the followers, in addition to the different level of the leader s giving amount. Another closely related literature on charitable giving studies how status concern and hierarchy structure affect prosocial behavior. The major challenge for an empirical study is to establish a causal relationship between prosocial behavior and social relationship. Carman (2004) uses donations to United Way by the employees of a large national bank and finds the causal effect of social influence on charitable giving through an instrumental variable approach. Another paper by Andreoni and Scholz (1998) find that donors respond positively to an increase in contributions by others in their reference group that is defined in terms of 2 Several papers examine the effect of social information in other types of games. For instance, Duffy and Feltovich (1999) look at how observation of one randomly chosen pair affects the play in repeated ultimatum and best-shot game. Bohnet and Zeckhauser (2004) show that presenting the average offer in ultimatum game activates the social norm of equal split. Ho and Su (2009) find that people exhibit peer-induced fairness concerns when they look to their peers as a reference to evaluate their endowments in ultimatum game. Eckel and Wilson (2006) investigate the observation of the move of another player in coordination game. 4

5 socioeconomic variables. Though these papers did careful analysis of the field data, the endogeneity problem and identification remain a challenge. Experimental design can prevent the endogeneity challenge by creating an exogenous change of social status. Bracha, Heffetz, and Vesterlund (2009), using a laboratory experiment, find that exogenous status significantly increases donations among high performers, but find little evidence on endogenous status-seeking behavior. Ebeling, Feldhaus, Fendrich (2013) find that, in a fundraising field experiment where they systematically varied the status of the first giver, people are also more prone to donate to a homeless individual when the previous donor has a higher social status. In this study, the leader has no hierarchy position, however, our study incorporated real hierarchy leadership and real social status concern among the followers. This paper is also part of the broad literature on sequential charitable giving. Previous literature has shown that the giving behavior of first-mover (leader) has a significant effect on charitable giving, mostly using laboratory experiments. Potters, Sefton, and Vesterlund (2005) find that sequential moves in a public good game results in a larger provision of the good, because the follower mimics the action of the leader. Potters, Sefton, and Vesterlund (2007) find that leading-by-example increases contributions and earnings in an environment where a leader has private information about the returns from contributing. Reinstein and Riener (2012 ) find that the giving of leaders are influential when their identities are revealed along with their donations. In most of these studies, the leader is just the first mover in the laboratory game. In our study, although it is also our interest to investigate the overall effect of the sequential contribution, our focus is on the behavior of the follower conditional on the information of leader s contribution, in a field setting with real hierarchy structure between the leader and the followers. The remainder of the paper is organized as follows. Section 2 presents the model. Section 3 explains the experimental design and procedure. Section 4 reports the empirical results. Section 5 concludes. 2 The Model We model a situation where a leader and a follower make charitable contributions to a cause sequentially. The model focuses on whether the follower is willing to give to a charity, and if so, how much to contribute. We adopt an additive formalization for a potential giver s decision problem, similar to the models in Alpizar, Carlsson, and Johansson-Stenman (2008) and DellaVigna, List and Malmendier (2012). The follower cares about his private income Y F g F, where Y F is his initial income and g F is his contribution to the public good, the total amount of public good G, his self image, I F, and his social approval, S F. The follower s utility function is given by U F = (Y F g F ) + α G G + I F + S F, (1) where we assume that 0 < α G < 1 so that the marginal utility of private income is larger than the marginal utility of money given to the public good. The amount of the public good is the sum of the follower s own contribution, g F, and the contribution by other donors, G F, so that G = g F + G F. We assume that the follower s self image depends on two components: his own contribution (due to for instance the warm-glow effect defined by Andreoni (1988)) and whether his contribution conforms with the social norms. On one hand, when he is given the leader s contribution g L as a reference point, he would like to be as generous as his leader. On the other hand, if within a culture where people have a hierarchy consideration, the follower may choose his contribution corresponding to his social status relative to the leader s. Consequently, the follower faces an ideal level of giving relative to his leader s giving that depends on social culture and norms. The follower incurs a disutility from deviating from these norms. Formally, we 5

6 write the follower s self image as follows: I F = α W g F β(g F γg L ) 2, (2) where α W > 0 implies that the follower s self image increases with his own contribution. β > 0 is the marginal cost of deviating from a particular social norm and measures to what extent the follower cares about the comparison with his leader or conforming to the social norm. The parameter γ > 0 represents different social norms the follower would like to follow in different scenarios. For instance, γ < 1 may reflect the idea that, when the hierarchy consideration is salient, the follower wants to keep his contribution lower than the leader s contribution in order to match their corresponding social statuses. We model social approval in a similar fashion as DellaVigna, List and Malmendier (2012) model social pressure in a door-to-door fundraising campaign. Social approval is written as follows: S F = (1 a)θ(g F g S ), (3) where g S is the cut-off contribution level that induces a positive (negative) utility when the follower s contribution is greater (smaller) than g S. The more one gives below g S, the more disutility one gets; the more above g S, the more utility one derives. a denotes the degree of anonymity of the follower s action. The follower s action is completely anonymous when a = 1 and completely public when a = 0. Finally, θ denotes the marginal benefit (marginal cost) of social approval. Substituting Eqs. (2) and (3) into Eq. (1), we have U F = U 0 F (1 α G )g F + α W g F β(g F γg L ) 2 + (1 a)θ(g F g S ), (4) where U 0 F Y F + α G G F is the follower s utility level without contributing to the public goods. Assuming an interior optimum and given the level of anonymity a, we can solve for the follower s optimal contribution level g F : g F = (α G 1) + α W + (1 a)θ 2β + γg L. (5) 2.1 Comparative Statics From equation (5), gf increases when α G, α W, or θ is larger, that is, when he derives higher marginal benefits from the total public good provision, the warm-glow effect, or the social approval. The follower contributes more when the contribution is less anonymous, based on the following derivative: gf a = θ < 0. (6) 2β When α W + (1 a)θ = 1 α G, i.e., the marginal utility of warm-glow and social approval, α W + (1 a)θ, is equal to the net cost of contribution, 1 α G, the follower s optimal contribution perfectly matches the social norm regardless of β, i.e, g F = γg L. Alternatively, we can rewrite the condition as α W + α G + (1 a)θ = 1. Hence, when the sum of pure and impure altruistic concern and social approval is equal to the marginal cost of giving, in terms of private consumption, the follower always gives according to the social norm, in which case the follower can avoid any punishment from not conforming to the social norm. When α W + (1 a)θ > 1 α G, the follower s contribution g F > γg L, in which case g F is decreasing in 6

7 the marginal cost of social comparison, β. On the contrast, when α W + (1 a)θ < 1 α G, the follower s contribution gf < γg L, and gf is increasing in β. Therefore, the social comparison cost, β, tends to force the follower s contribution gf to move toward the social norm γg L. The follower s contribution responds to the leader s contribution by a rate of γ, according to the first derivative of the follower s optimal contribution: g F g L = γ. (7) Proposition 1 The follower s optimal contribution, gf, has the following properties: i. g F increases in the marginal benefit from the warm-glow effect α W, the public good provision α G, and the social approval θ, and when the contribution is more public (a is smaller); ii. The response of gf to the social comparison cost β can be zero, positive, or negative, depending on whether gf is equal to, smaller than, or larger than the social norm γg L; iii. g F increases with the leader s contribution g L in a constant proportion of γ. 2.2 Zero Contribution and Anonymity Choice In this subsection we consider under what conditions the follower will choose to contribute zero. If the follower chooses to contribute zero before knowing the existence of the leader s contribution, we can take β = 0, that is, there is no social comparison. The follower will choose to contribute zero if α W + α G + (1 a)θ < 1, (8) that is, the marginal utility from the pure and impure altruism and the social approval is smaller than the marginal benefit of private consumption. If we further assume that zero contribution will not be recorded or publicized, then the follower will choose to contribute zero without the presence of social comparison if α W + α G < 1. (9) If the follower chooses to contribute zero after knowing the leader s contribution, then from equation (5) we have α W + α G + 2βγg L + (1 a)θ < 1, (10) that is, the sum of the marginal utility of altruism, conformity to the social norm, and social approval is smaller than the marginal benefit of private consumption. As a summary, the follower will contribute zero if and only if: α W < 1 α G, under anonymity and without social comparison; α W < 1 α G θ, under non-anonymity and without social comparison; α W < 1 α G 2βγg L, under anonymity and with social comparison; α W < 1 α G θ 2βγg L, under non-anonymity and with social comparison. 7

8 Proposition 2 If the follower chooses to contribute a positive amount without social comparison, he must contribute a positive amount with social comparison, given the same anonymity condition. If the follower chooses to contribute a positive amount in an anonymous setting, he must contribute a positive amount in a non-anonymous setting, given the same condition for social comparison. Now consider the case when the follower can choose being anonymous or not endogenously. Given the follower s optimal contribution in equation (5), denote gf,0 = α W (1 α G )+θ 2β + γg L as the follower s optimal contribution under the non-anonymous environment, and gf,1 = α W 1+α G 2β + γg L as the follower s optimal contribution under the anonymous environment. Correspondingly, we can derive the follower s maximized utility UF,0 and U F,1 when g F is equal to gf,0 and g F,1 respectively. Denote g F g F,0 g F,1 = θ 2β. U F U F,0 U F,1 = (α W 1 + α G ) g F β(g F,0 + g F,1 2γg L ) g F + θ(g F,0 g S ) = (α W 1 + α G ) θ 2β β(g F,0 + g F,1 2γg L ) θ 2β + θ(g F,0 g S ) = (α W 1 + α G ) θ 2β + (g F,0 g F,1) θ 2 + θγg L θg S = (α W 1 + α G ) θ 2β + θ2 4β + θγg L θg S The follower chooses to be non-anonymous if U F > 0, that is, θ > 4βg S 2(α W 1 + α G ) 4βγg L. This condition implies that, given a fixed set of altruistic and social comparison parameters, if the marginal benefit from being perceived as generous (θ) is high enough or the cutoff contribution level, beyond which the follower is regarded as generous (g S ), is low enough, the followers will choose to be non-anonymous. A higher altruism concern (α G + α W ) or a larger contribution amount given by the social norm also leads to a higher probability of choosing non-anonymous. Proposition 3 Given the altruistic and social comparison parameters, the followers will choose to be nonanonymous if the marginal benefit from being perceived as generous, θ, is high or the cutoff contribution level for social approval, g S, is low. 2.3 Treatments with Different Relative Social Status According to equation 5, both social comparison parameters, β and γ, play an important role in determining the follower s giving and his response to the leader s contribution. Parameter β captures the extent to which the follower cares about following the social norm (or social comparison) associated with the leader s contribution. Parameter γ represents the social norm, that is, how the follower s giving responds to the leader s giving. Empirically, we aim to study how social comparison affects the follower s giving decision, when the social distance between the leader and the follower varies. We create the variation of social distance between a leader and her followers through the following treatments. In treatment 1, Leader treatment, the follower is asked to donate in an environment where the leader has real hierarchy leadership; in treatment 2, Peer treatment, the follower perceives that the leader has a similar social status as he has; in treatment 3, Stranger (Baseline) treatment, the follower is not aware of any real social connection with the leader. We summarize the predictions of the three treatments as follows, where γ k and β k (k = L, P, S) denote the corresponding parameters in different treatment k. 8

9 Leader treatment. In this case the follower has a lower social status than the leader, so we conjecture that γ L < 1, that is, the follower likes to conform to a smaller amount than that the leaders has contributed in order to avoid being offensive to the leader by giving too much ; however, γ L might not be too small because the leader s giving also serves as an example of generosity so the follower may feel obligated to respond positively to the leader s giving. We expect a larger γ L if the leader s contribution has a stronger effect of leading by example. On the contrary, we predict a smaller γ L when the discouraging effect of hierarchy concern dominates other elements. We expect a high value of β L since the follower should put a large weight on the social comparison with the leader. Peer treatment. Here, the leader has no hierarchy position and is only a peer of the follower. Since the social statuses of the leader and the follower are similar, we conjecture that γ P is higher than γ L in the leader treatment since the follower is more likely to match or respond positively to the giving of his peer without worrying about giving more than the peer. We conjecture a positive and significant β P because of social comparison with peers, but the effect between peers may be less than that in the Leader treatment, so β P < β L. Stranger treatment. This is the scenario where the leader has no social connection with the follower and is simply a first mover. We conjecture that both β S and γ S are small in this treatment. Since the follower does not have any social information about the leader, the social tie between the leader and the follower is weak. The follower may not care about comparing with the leader s contribution, neither does he care about the exact amount the leader has contributed. The follower s contribution in this case may be driven mostly by warm-glow and/or pure altruism. According to the discussion above, we write down the following hypothesis. Hypothesis 1 We conjecture that (i) the hierarchy leadership has the highest marginal cost from social comparison, and the stranger treatment has lowest marginal cost from social comparison (i.e., β S < β P < β L ); (ii) the follower s contribution is most responsive in the peer treatment and least responsive in the stranger treatment (i.e., γ S < γ L < γ P ). 3 The Experiment 3.1 Experiment Design In order to implement the three treatments discussed in the previous section and identify the effect of hierarchy leadership on the prosocial behavior of the follower, we need a group of followers associated with one leader so that we can have enough observations across the three treatments under one leader. Moreover, it is necessary to have a group of leaders in order to make sure that the treatment effects, if any, do not rely on one particular leader and his/her specific characteristics. In addition, the hierarchical structure between the leader and the followers should be comparable across groups. Based on these considerations, a university seems to be a perfect ground to conduct the field experiment since there exist many departments in a university and each department has a more or less similar hierarchy structure. The additional advantage of conducting the fundraising campaign among academic faculties in a university is the seniority of professor ranks which we can use to control for income effects. Finally, we can avoid direct social interactions among the potential donors during the solicitation process given that every professor has his/her own office. 9

10 We create the variation in the social distance between the leader and the follower for one given leader by varying the information disclosed to the followers. In the Stranger treatment, the leader s giving amount was disclosed to the followers. However, we did not disclose any information regarding the leader s identity including his/her name and affiliation. Therefore, the followers were not aware that the amount shown to them was in fact their chair s giving but only knew that it was from a previous donor. This treatment can capture the anchor-point effect, that is, the impact of the giving amount from the leader (first-mover). In the Peer treatment, we disclosed both the leader s giving amount and his/her affiliation (the name of the department). We did not reveal the leader s name, however. Hence, the follower only knew that one of his peers/colleagues had given at a certain level. The followers response to their peer s giving might be different from their response to the giving of a completely anonymous giver, even if the giving amounts in the Peer treatment and the Stranger treatment were exactly the same. The Peer treatment captures both the anchor-point effect and the peer effect. In the Leader treatment, the leader s identity/name, department name, and giving amount were all disclosed to the followers. This treatment captures not only the effect of giving amount from a peer but also the effect of hierarchy leadership. The essential difference between the three treatments is the information disclosed on the pledge card for donation. Examples of the pledge cards for the followers are illustrated in Figure 1 for the Leader, Peer, and Stranger treatment from the left to the right respectively. 3 Figure 1: Examples of Pledge Cards in Leader, Peer, and Stranger Treatments (from Left to Right) The solicitors approached the leader first within a department. When they approached the leaders, they showed them an empty donation card as the one in Figure 2, which was the actual pledge card with the organization information. No information of other donors appeared on the card. After that, they approached the followers. Each follower saw a pledge card which contained only the leader s giving information varying according to the treatment. 3.2 Experiment Procedure In conducting the experiment, we collaborated with the Leukemia & Lymphoma Foundation (LLF) in Singapore. LLF is registered as a nonprofit organization and has charity status. Its primary mission is to fight blood cancers and provide financial assistance to low-income patients, as well as providing other supports to the patients and/or their families. Leukemia and lymphoma are among the most curable cancers, but the treatment is very expensive and needs long-term care. The cost involves chemotherapy, stem cell transplant or special medication, which needs to be undertaken in the long term and is a huge financial burden for patients to cope with. The financial, counselling, and emotional support from LLF can be crucial for the 3 In the university we conducted the experiment, division is the same as department in most universities in North America. Since department is more commonly used in other places, we use department in the presentation of this paper. 10

11 Figure 2: The Inside and Front Pages of the Actual Pledge Card well-being of patients and their families. We conducted the fundraising experiment at a large university in Singapore. We chose the chair of each academic department (division) as the leader and other professors or teaching faculty in the same department as the corresponding followers. The fundraising was mostly conducted between the 2nd of October and the 12th of December of 2012, with a few observations (17 out of the total 184 donors) from the first two weeks of April of Four solicitors went around in fixed mixed-gender pairs. We started the solicitation process targeting the leader (chair) of a department. After securing the leader s contribution, the solicitors approached other professors (the followers) in a department. In order to ensure consistency, they followed a standardized procedure, including the soliciting message, gender and pair combination, and donation pledge cards. When the solicitors approached the leaders, we printed the leaders name and affiliation on the pledge card so 4 The results are essentially the same if we drop the observations for the fundraising campaign in the April of

12 that they know other donors (followers) might know their identity. All leaders, except one, that the solicitors managed to approach gave a positive amount. None of them asked for keeping themselves anonymous. 5 After obtaining the leader s donation, we then prepared the pledge cards for the followers according to the three treatments, where the followers would see different information of the leader s identity, affiliation, and giving amount. The solicitors then went to a department and knocked on the door of a potential donor (follower). If the professor was in the office and opened the door, one solicitor then presented the soliciting message as the following: Hi Prof, can we get 5 minutes of your time? We are a group of volunteers representing the Leukemia & Lymphoma Foundation (LLF). We are basically going around to create awareness for them, as well as to raise funds at the same time. To highlight some points regarding leukemia and lymphoma, they are blood disorders which are curable but very expensive at the same time. Patients have to go through procedures like chemotherapy, stem cell transplant or special medication taken over long term which is a huge financial burden to cope with. If you are interested in donating to these needy patients, you can do so by penning down your donations on this pledge card. Thank you. At the beginning of the solicitation message, the other solicitor provided the potential donor with a Leukemia & Lymphoma Foundation (LLF) brochure and a proof letter from the LLF showing that the solicitors were volunteers of the charitable organization. The brochure clearly identified the charity status of the organization and the proof letter was used to establish the credibility of the solicitors. Towards the end of the soliciting message, the solicitors showed the pledge card to the professor, making sure that he/she could see the information of the leader on the card. The front of the card was shown first, and after that, the solicitors opened the card and provided instructions on how to fill up the relevant information of the donor and the donation amount. At the end of the session, the solicitors informed the donor that the donation was tax deductible and expressed our appreciation that the donor could help out to create more awareness for the Leukemia & Lymphoma Foundation (LLF). There were some cases that the solicitation process had to be ceased before the potential donor saw the pledge card. Usually in those cases, the solicitors were interrupted and asked to leave the room during delivering the solicitation message. Given the fact that the potential donor did not see the different information on the pledge card before deciding not to donate, we cannot allocate these cases in a particular treatment. As a result, the data we present from each treatment are the cases all with a positive donation. Although we realize that it may be a more informative data set if we observe the zero contribution in each treatment, it was much more natural to describe the cause first to the potential donor before showing the pledge card during the solicitation. Notice that according to our model, people who have chosen to contribute zero without seeing the pledge card may choose to contribute a positive amount if they have had the chance to see the pledge card, since in the latter case the additional effect of social comparison may drive the donor s contribution from a corner solution to an interior solution. On the other hand, if the follower has allowed the solicitors to finish up the entire solicitation message, we can interpret it as the follower s willingness to donate without social comparison. 6 During the fundraising, we tried to have a balanced number of participants across all treatments by reversing the solicitation order for the three treatments. That is, in one round of the solicitation, the first donor is put into treatment 1, and the second and the third donor are in treatment 2 and 3 respectively. In the next round, the first, the second, and the third donor are in treatment 3, 2, and 1 respectively. Before the solicitors randomly chose a potential donor and knocked the door of his office, they prepared the solicitation according 5 If the leader requested for anonymity, we cannot implement the Leader treatment and compare to the other two treatments. 6 By the analysis in Section 2.2, we can infer that, after seeing the information on the pledge card, the follower should contribute a positive amount because of social comparison. 12

13 to the protocol for each treatment in an alternating manner. Only if the professor donated a positive amount, is it counted as one observation in that treatment, for the reason we discussed in the previous paragraph. 3.3 Summary of the Sample In total, we approached 308 professors, among which 212 donated a positive amount. The total contribution is $4224, with an average giving of $20. Table 1 shows the sample distribution across schools and treatments. In the sample, we have 21 leaders (department chairs) who gave $25.4 on average; the remaining are 191 followers who gave $19.3 on average. All but one leader are male, and all the leaders are tenured professors. The donors came from 8 different schools in the university. The Stranger, Peer and Leader treatments have 57, 64, and 70 observations/followers, respectively. In some schools, such as the School of Computer Engineering and Business School, there is an unequal number of observations in the three treatments. This is because the solicitors did not strictly follow the design to reverse the order of the solicitation for the three treatments in some departments, so it leads to more observations from the Leader treatment in some departments and hence some schools. In the regression analysis, we control for the impact of the unequal number of observations in different treatments by adding school and department dummies in the regression. Table 1: Sample Distribution Across Schools and Treatments School Number of Number of Leader Peer Stranger leaders followers treatment treatment treatment Civil and Environmental Engineering Computer Engineering Electrical and Electronic Engineering Humanities and Social Sciences Materials Science and Engineering Mechanical and Aerospace Engineering Physical and Mathematical Sciences School of Business Sum In the analysis, we dropped 3 observations from 2 departments where we have no observations in some treatments. We also dropped 4 outlier observations from the followers who gave $100, which is too large relative to a median giving of $10 or $20 in different treatments. These outliers can significantly affect the aggregate statistics such as the mean and the related comparisons. Finally, 184 observations remained with 67, 62, and 55 in leader, peer, and stranger treatment respectively. The observations can be classified into different groups by the treatments or by the giving levels of followers and their leaders. The last row of Table 2 shows the distribution of the followers giving by each classification. Out of the total 184 followers, 110 gave at the lowest category $10 or below; 46 gave at medium level larger than $10 and smaller or equal to $20; 28 gave at the highest level above $20 and smaller or equal to $50. If classified by the leader s giving, 75, 72, and 37 followers are in the group with the leader s giving at $10, $20, $50/$100 respectively. We further calculate the distribution of the followers giving by their positions, gender, anonymity choice, and solicitor group, as shown in the second column of Table 2. The majority of the observations are from associate and assistant professors. Only 20 of the 184 followers are female. More than one-third of the 13

14 followers chose to be anonymous, and around one-third were solicited by solicitor group B. All these factors are potentially important determinants of charitable giving behavior and need to be controlled for in the robustness checks. Table 2: The Distribution of Followers Giving in Different Groups Total By treatment By leaders giving By followers giving Leader Peer Stranger $10 $20 $50/$100 (0, $10] ($10, $20] ($20, $50] Professor Associate professor Assistant professor Other positions Male Female Non-anonymous Anonymous Solicitor group A Solicitor group B Sum Notes: Other positions are mostly lecturers. One leader who gave $8 is in the group with leaders giving $10. 4 Results 4.1 Summary Statistics Table 3 shows the summary statistics of the contributions from both the leaders and the followers. The followers giving amounts are presented for each level of the leader s donation and in each treatment separately. Among the 19 leaders, six of them gave $10 (one of them gave $8 and is counted in this category for simplicity); eight gave $20; four gave $50; one gave $100. As shown in last 3 rows of Table 3, the followers in the Leader treatment give the least and those in the Peer treatment give the most on average. The average giving of the followers in the Stranger, Peer, and Leader treatments is $17.9, $19.4 and $16 respectively, which results in a difference of $2 (13%) between the Stranger and the Leader treatments and a difference of $3.4 (21%) between the Peer and the Leader treatments. Statistically, the difference is insignificant between the Stranger and the Leader treatments (p > 10%) but marginally significant between the Peer and the Leader treatments (p < 10%), by using one-sided t-tests (122 obs. and 129 obs. respectively). When we take a closer look at the followers giving conditional on their associated leader s giving, there appears a more significant and different pattern of the treatment effects. In particular, we find that, in the Leader treatment, the leaders giving at a small amount discourages the followers giving while the leaders giving at a large amount encourages the followers giving. For instance, in Table 3, when the leaders giving is $20, the followers in the Leader treatment on average gave $5.4 and $6.9 less than those in the Peer and Stranger treatments. This pattern changes when the leaders giving amount is high at $50 and $100. For example, when the leaders giving is $50, the followers in the Leader treatment gave on average $10.1 and $3.1 more than those in the Peer and Stranger treatments. The differential response of the followers giving to the different levels of the leaders giving can be better illustrated in Figure 3. In the Stranger treatment, there appears no significant positive correlation between the followers giving and the leaders giving. In the Peer treatment, there is a consistent and stable positive 14

15 Table 3: Summary Statistics of Donors Giving Leader s Number Treatment Number of Mean Median Standard Minmum Maximum giving of leaders group followers deviation 10 6 Leader Peer Stranger Leader Peer Stranger Leader Peer Stranger Leader Peer Stranger Total 19 Leader Peer Stranger relationship between the leaders giving and the followers giving. In the Leader treatment, there is a negative (not significant) relationship between the leaders giving and the followers giving at the lower levels of the leaders giving, but a positive relationship at the higher levels of the leaders giving. Figure 3: Followers Average Giving in Response to Leader s Giving across Treatments In order to further statistically confirm the pattern in Figure 3, we conduct both t-tests and Wilcoxon- Mann-Whitney (WMW) tests on the differences in the followers giving conditional on the leaders giving between different treatments. Specifically, when the leaders giving is low, at $10 or $20, we test the null hypothesis that the followers giving in the Leader treatment is smaller than that in the both Peer and Stranger treatments. When the leaders giving is high, at $50 or $100, we test the null hypothesis that the 15

16 Table 4: p-value of Tests on Followers Giving Conditional on Leaders Giving g L at $10 or $20 g L at $50 or $100 Null hypothesis g L F < gs F g L F < gp F g L F > gs F g P F > gs F p-value (t-test) p-value (Wilcoxon-Mann-Whitney test) Number of Observations followers giving in the Stranger treatment is smaller than that in the both Peer and Leader treatments. Table 4 below provides the p-value of these tests. Indeed, when the leaders contribution is at the low level, $10 or $20, the null hypothesis is supported by both tests. When the leaders contribution is at the high level, $50 or $100, the difference is significant using the WMW test but not the t-test. This may be due to the smaller sample size associated with large contributions of the leaders, which makes the t-test not appropriate. Some of the features of the followers giving conditional on the leaders giving also holds if we examine the summary statistics within each school, as seen in Table 5. The followers average giving in the Leader treatment is smaller than those in the Stranger treatment, when the leaders average giving at the school level is small. This pattern changes when the leaders average giving within a school is large and the followers average giving in the Leader treatment is larger than those in the Stranger treatment. Table 5: Average Giving of Leaders and Followers across Schools and Treatments School Leaders Followers Followers Followers Followers in all in all in Leader in Peer in Stranger treatments treatments treatment treatment treatment School of Business Mechanical and Aerospace Engineering Physical and Mathematical Sciences Electrical and Electronic Engineering Computer Engineering Humanities and Social Sciences Materials Science and Engineering Civil and Environmental Engineering To better understand the significant difference in the followers giving conditional on the leaders giving across treatments, we further examine the distribution of the followers giving under different levels of the leaders giving. Figure 4 presents the conditional distribution of the followers giving when the leaders giving is at $10, $20, and $50/$100, respectively. On the horizontal axis in Figure 4, 1, 2, 3 refer to the Leader, Peer, and Stranger treatment respectively. The conditional distribution of the followers giving has very different patterns at different levels of the leader s giving. When the leaders giving is at $10 or $20, the followers giving concentrates at the lower range around $10; the Leader treatment has the highest percentage of the followers who gave less than or equal to $10 and has the lowest percentage of the followers who gave more than $20. In contrast, when the leaders giving is $50 or $100, some percentage of the followers shifts from giving less than or equal to $10 to giving more than $20. The percentage of the followers who gave less than or equal to $10 is lowest in the Leader treatment and highest in the Stranger treatment. On opposite, 16

17 the percentage of the followers who gave between $10 and $20 or gave more than $20 is highest in the Leader treatment and lowest the Stranger treatment. The conditional distribution of the followers giving is consistent with those from Table 3 and Figure 3. As a summary for this subsection, we find different patterns in the followers responses to the leaders contribution in different treatments. In the Stranger treatment, the followers giving is least responsive to the leaders contribution, resulting in relatively constant contributions no matter whether the leaders contribution is low or high. In the Leader treatment, the followers giving has a nonlinear relationship with the leaders giving. If a leader gives a small amount, disclosing the identity of the leader seems like a bad idea to raise more donations. However, if a leader gives a large amount, disclosing the leader s identity encourages more giving from the followers. From our model, we have derived a constant responsiveness of the followers giving to the leaders giving, γ. However, the empirical findings suggest that the responsiveness γ may not only depend on the treatment but also depend on the level of the leader s contribution. The leader can positively affect the followers giving when his donation serves as a good example; but it can also negatively affect the followers giving as a bad example. This finding may be due to the followers consideration of social comparison upon seeing the leader s contribution when the leader possesses a real hierarchy leadership. This consideration may have driven the followers to contribute less when the leader s contribution is low and pushed the followers to contribute more when the leader s contribution is high, compared to the case when there is no such consideration. In contrast, the leader s giving in the Peer treatment has a consistently positive effect on the followers giving. This might reflect the fact that the social status or social comparison concern is simpler among peers who do not have hierarchy leadership over each other. 4.2 Treatment Effects In order to estimate the treatment effect, we first need to control for the impact of the leaders giving amount, which is an important factor in affecting the followers giving no matter which treatment the followers are in. Moreover, the responsiveness of the followers towards the leaders giving differs in different treatments, as documented in the summary statistics. Hence, it is important to incorporate the interaction items between the treatment dummies and the leaders giving. In addition, we include other control variables such as the followers position, gender, solicitor group, school and division fixed effects, whether or not the followers choose to be anonymous, and the interaction item between the followers position and the leader s giving. A benchmark model is formalized as follows: g F,i = α + 2 t=1 β t treattreat t i + β e g L,i + 2 t=1 β t interacttreat t i g L,i + X i β X + ξ i, (11) where the dependent variable g F,i is follower i s giving. The independent variables include two dummy variables for the leader and peer treatment treat 1 i and treat2 i, the leader s giving faced by follower i, g L,i, and the interaction items between the treatment dummies and the leader s giving, treat t i g L,i. X i are the other control variables mentioned above. The estimation results are presented in Table 6. Since we include the interaction variable treat t i g L,i in the regression, the coefficient estimated for the treatment dummy captures the treatment effect when the leader s contribution is close to zero, compared to the (baseline) Stranger treatment. We see that the estimate for the leader treatment dummy is significantly negative, implying that the followers in the leader treatment gave significantly less than those in the baseline treatment when the leader s contribution is very low. The estimate for the peer treatment is negative but not significant. Hence, the followers contribution in the peer treatment is not significantly different from those in the baseline treatment when the leader s contribution 17

18 Figure 4: Conditional Distribution of Followers Giving across Treatments 18