THE QUARTERLY JOURNAL OF ECONOMICS Vol. 127 February 2012 Issue 1 TESTING FOR ALTRUISM AND SOCIAL PRESSURE IN CHARITABLE GIVING* Stefano DellaVigna John A. List Ulrike Malmendier Every year, 90% of Americans give money to charities. Is such generosity necessarily welfare enhancing for the giver? We present a theoretical framework that distinguishes two types of motivation: individuals like to give, for example, due to altruism or warm glow, and individuals would rather not give but dislike saying no, for example, due to social pressure. We design a door-to-door fund-raiser in which some households are informed about the exact time of solicitation with a flyer on their doorknobs. Thus, they can seek or avoid the fund-raiser. We find that the flyer reduces the share of households opening the door by 9% to 25% and, if the flyer allows checking a Do Not Disturb box, reduces giving by 28% to 42%. The latter decrease is concentrated among donations smaller than $10. These findings suggest that social pressure is an important determinant of door-to-door giving. Combining data from this and a complementary field experiment, we structurally estimate the model. The estimated social pressure cost of saying no to a solicitor is $3.80 for an in-state charity and $1.40 for an out-of-state charity. Our welfare calculations suggest that our door-to-door fund-raising campaigns on average lower the utility of the potential donors. JEL Codes: C93, D03, H41. *We thank the editor, four referees, Saurabh Bhargava, David Card, Gary Charness, Constanca Esteves-Sorenson, Bryan Graham, Zachary Grossman, Lawrence Katz, Patrick Kline, Stephan Meier, Klaus Schmidt, Bruce Shearer, Joel Sobel, Daniel Sturm and the audiences at the Chicago Booth School of Business, Columbia University, Harvard University, HBS, University of Amsterdam, University of Arizona, Tucson, UC Berkeley, USC (Marshall School), UT Dallas, University of Rotterdam, University of Zurich, Washington University (St. Louis), Yale University, the NBER Summer Institute (Labor Studies), the San Francisco 2009 Applied Micro Conference, the 2009 Berkeley Conference in Behavioral Economics, the 2010 UCSB Conference on Experimental Economics, the Munich Workshop on Natural Experiments and Controlled Field Studies, and the ASSA 2009 Meetings for helpful comments. We also thank Daniel Acland, Diane Alexander, Jim Cai, Matthew Levy, Vera te Velde, Xiaoyu Xia, and especially Gautam Rao for excellent research assistance. © The Author(s) 2012. Published by Oxford University Press, on behalf of President and Fellows of Harvard College. All rights reserved. For Permissions, please email: journals. permissions@oup.com. The Quarterly Journal of Economics (2012) 127, 1-56. doi:10.1093/qje/qjr050. Advance Access publication on January 18, 2012. 1 2 QUARTERLY JOURNAL OF ECONOMICS I. Introduction In the United States, approximately 90% ofindividuals donate money each year. There is at least one capital campaign to raise $25 million or more under way in virtually every major population center in North America. Smaller capital campaigns are even more numerous, with phoneathons, door-to-door drives, and mail solicitations increasing in popularity. Despite the ubiquity of fund-raising, we still have an imperfect understanding of the motivations for giving and the welfare implications for the giver (see, e.g., Andreoni 2006). In this article, we consider two broad classes of motivations. First, individuals may enjoy giving. For example, they care about a specific worthy cause or like the warm glow of giving. Second, individuals may give, despite not liking to give to the charity, because the solicitor effectively placed them under social pressure to give. Such givers would rather avoid the personal interaction with the solicitor. The two motivations have very different welfare implications. The altruism (or warm glow) model (Becker 1974; Andreoni 1989, 1990) posits that giving is mostly supply-driven, and that it is utility-maximizing for the giver to give. Under this model, donations unambiguously enhance the giver's utility as well as societal welfare. The social pressure model (Akerlof and Kranton 2000) posits that giving is mostly demand-driven, and that giving may be utility-reducing for the giver. We test for these two types of motivations in the context of in-person, unsolicited donation requests. Building on a theoretical model, we design a field experiment that allows us to test whether giving is welfare-enhancing or welfare-reducing for the giver. We complement the reduced-form experimental evidence with structural estimates of the model parameters. The structural estimation allows us to decompose the share of giving that is due to altruism versus social pressure and to quantitatively evaluate the welfare effects for the giver. In this way, the empirics and theory are intertwined in a manner that is rare in this literature. To our knowledge, this article is the first in the behavioral literature to provide structural estimates of welfare implications of a field experiment. Moreover, while the fund-raising set-up is specific, it showcases a general methodology and provides a first step toward better understanding the underpinnings for giving more generally. TESTING FOR ALTRUISM AND SOCIAL PRESSURE 3 Our field experiment revolves around a door-to-door fund-raising drive for two charities, a local children's hospital, which has a reputation as a premier hospital for children, and an out-of-state charity, unfamiliar to most solicitees. Between April and October 2008, we approached 7,668 households in the towns surrounding Chicago. The crucial aspect of the experimental design is to allow individuals to sort, that is, to either seek or avoid the solicitor. In our first treatment, a flyer on the doorknob notifies households one day in advance about the one-hour time interval in which a solicitor will arrive at their homes the next day. In the second treatment, opt-out, the flyer also includes a box to be checked if the household does not want to be disturbed. We compare these two conditions to a baseline treatment, wherein solicitors approach households in the usual manner without a flyer. We estimate the treatment effects on both the share of households that open the door and the share that give. This design allows for a simple test of (pure or impure) altruism versus social pressure in door-to-door giving. If altruism is the main driver of giving, the flyer should increase both the presence at home and giving. Because giving is utility-enhancing, givers should sort into staying at home, provided that alternative ways of donating to these charities require more effort. In addition, givers who would like to give in response to the flyer but who find it too costly to be at home should give to the charity via other means, such as mailing a check. Conversely, if social pressure is the main driver of giving, the flyer should lower both the frequency of opening the door and the frequency of giving. Since being asked to give is welfare-diminishing, potential givers should sort out of opening the door and should not give via Internet or check because these forms of donation are not subject to social pressure. We report four main results, which are similar across the two charities. First, the flyer lowers the frequency of opening the door. Relative to a baseline rate of 41 percentage points, the share of households opening the door is 9% lower after receiving the flyer and 23% lower after receiving the flyer with an opt-out box (including the households that check the opt-out box). Second, the simple flyer does not reduce giving. However, the flyer with an opt-out checkbox decreases giving significantly, by 28% relative to a baseline of 7% for the local charity and 40% relative to a 5% baseline giving for the out-of-state charity. Third, the decrease in giving in the opt-out treatment is driven by small donations up to $10; donations above $10, instead, increase slightly (not 4 QUARTERLY JOURNAL OF ECONOMICS significantly). Fourth, there is no effect on donations via mail or Internet. In contrast to the substantial donation rates in person, only one household out of 7,668 gave through these other means. Overall, the reduced-form estimates indicate that both altruism and social pressure are important determinants of giving in this setting, with stronger evidence for the role of social pressure. The lower frequency of households opening the door after receiving a flyer indicates that households are, on average, trying to avoid solicitors, consistent with social pressure. The lack of an effect of the baseline flyer on giving is consistent with opposing effects of altruism and social pressure approximately canceling each other out. The decrease in giving after a flyer with opt-out box supports the role of social pressure. When the cost of avoiding the solicitor is lowered (a simple check on a box suffices), giving due to social pressure decreases. This interpretation is consistent with the reduction occurring almost exclusively among small donations, which are more likely due to social pressure than large donations. The social pressure interpretation is also consistent with the lack of donations via mail or Internet. We consider several alternative interpretations. First, flyers could be taken as a signal of lower quality of the charity. This interpretation can explain the reduction in answering the door and in giving with a flyer, but does not immediately explain why only small donations decrease, not large donations. Second, individuals might donate to send (costly) signals to themselves or to others that they are generous (Bodner and Prelec 2003; Ben-abou and Tirole 2006; Grossman 2010). This interpretation can explain our findings to the extent that avoiding a solicitor does not send the same negative signal as saying no to the solicitor. This explanation, however, is less consistent with the fact that 12% of households in the opt-out treatment check the opt-out box, which is a signal to the solicitor, as well as possibly to the neighbors. Third, people may dislike spending time with the solicitor, for example, because of the time involved, despite wanting to donate in principle. These potential donors, however, when alerted of a campaign by a flyer, should seek alternative ways to give that do not involve personal interaction; instead, we observe no such substitution. To assess the welfare effects of giving, we structurally estimate the model parameters. We combine data from the treatments with complementary field experiments on the value of time, run in the same geographical areas in 2008 and 2009. These TESTING FOR ALTRUISM AND SOCIAL PRESSURE 5 experiments are designed to estimate a key parameter in the fund-raising treatments, the sensitivity of home presence to incentives, which is otherwise identified only parametrically. We ask 11,900 households to complete a survey and vary the payment ($0, $5, or $10), the duration (5 or 10 minutes), and whether the surveys are announced (with a flyer, with or without opt-out option). Higher payments and shorter duration increase the presence at home up to 16% and increase survey completion by 17-82%. We use a minimum-distance estimator on the combined data from the charity and the survey experiments. The estimator minimizes the distance between the moments predicted by the model and the observed moments. The moments are the probabilities of opening the door, giving different amounts, completing a survey, and opting out. Key parameters are the share of altruists, the curvature of the altruism function, and the social pressure cost of saying no in person to a solicitor. We estimate that 75% of solicitees have no altruism toward the charities, but there is substantial heterogeneity. Among the altruists, the utility from giving due to altruism is steeply concave in the amount given, with almost no predicted donation above $50, consistent with warm-glow rather than pure altruism. The estimated social pressure cost of giving zero is $3.75 (and significantly different from $0) for the in-state charity and $1.44 (marginally significant) for the out-of-state charity. As a result of social pressure, a majority of donors give more than they would have liked to. Half of donors derive negative utility from the fund-raising interaction and would have preferred to sort out. Given the large social pressure costs, our door-to-door campaigns lower the utility of the solicited households on average. In the benchmark specification, a visit is estimated to lower welfare by $1.10 per household contacted for the in-state charity and by $0.44 for the out-of-state charity. The more negative welfare impact for the in-state charity is counterintuitive because more people are willing to donate to this better-liked charity. At the same time, however, the social pressure cost of saying no is also significantly higher for the local charity, and the second force dominates. If we take our fund-raising campaigns to be representative of door-to-door solicitations, unsolicited campaigns lead to utility losses for the givers in the order of hundreds of millions of dollars. The campaigns may still increase overall welfare, though only if the charities spend the money very effectively; in our campaign, 6 QUARTERLY JOURNAL OF ECONOMICS we raised net donations of only $0.24 per household contacted for the in-state charity, and no net donation for the out-of-state charity. An important qualification is that our design identifies reasons for marginal, as opposed to infra-marginal, giving. Households that do not give to our fund-raiser, or give only due to social pressure, likely contributed to other charities. The motives for giving identified in this article may not generalize to infra-marginal giving, which is more likely motivated by altruism and desire for status. By the same token, however, it would be a mistake to ignore the high-pressure giving requests studied herein, or to assume that the motives for infra-marginal giving studied in the literature apply. Small capital campaigns, like the one studied in this article, are common and reveal a different facet of the motivations for giving. Our findings can be used as an argument to introduce a do-not-solicit or do-not-call list for charities. However, they also suggest an alternative: providing households with the opportunity to sort or, even better, opt out. Introducing sorting opportunities in fund-raising limits or eliminates altogether, the welfare losses for the solicitees. Interestingly, introducing sorting can also increase charitable fund-raising, and be a win-win solution: even a limited amount of sorting in of altruistic givers, who give larger amounts, is likely to counterbalance the sorting out of givers motivated by social pressure, who give smaller amounts. A methodological contribution of this article is the close tie between the behavioral model and the field experiment, allowing for structural estimation of the underlying parameters, which is surprisingly rare. Of all field experiments published in top five journals from 1985 to 2010, only two have this feature (Card, DellaVigna, and Malmendier 2011). A small literature in structural behavioral economics estimates behavioral models on observational data, including Laibson, Repetto, and Tobacman (2007) and Conlin, O'Donoghue, and Vogelsang (2007). Our article adds to several other strands of literature. It provides field evidence about social preferences to complement the laboratory evidence (Fehr and Gachter 2000; Charness and Rabin 2002; and especially Dana, Cain, and Dawes 2006; Lazear, Malmendier, and Weber forthcoming). The study also relates to the empirical and theoretical literature on optimal fund-raising (e.g., List and Lucking-Reiley 2002; Andreoni 2006; Landry et al. 2006; Ariely, Bracha, and Meier 2009; Croson and Shang 2009; TESTING FOR ALTRUISM AND SOCIAL PRESSURE 7 Fong and Luttmer 2009). Finally, it adds to the literature on social pressure (Asch 1951; Milgram 1963; Garicano, Palacios-Huerta, and Prendergast 2005; Gerber, Green, and Larimer 2008; Mas and Moretti 2009). The rest of the article proceeds as follows. In Section II we present a simple model of giving with altruism and social pressure. We introduce the experimental design in Section III and discuss the reduced-form results in Section IV. In Section V, we structurally estimate the parameters. In Section VI, we discuss alternative interpretations. Section VII concludes. II. Model We model the behavior of an individual whose home is visited by a fund-raiser. We distinguish between the standard case of an unanticipated visit and that of an anticipated visit. In the latter case, a flyer announces the visit and the individual can alter the probability of being at home and opening the door. We discuss here the setting and predictions. The technical details, including Lemmas are in the Appendix, and the proofs are in the Online Appendix. II.A. Setup We consider a two-stage game between a potential giver and a solicitor. For convenience, we denote the potential giver, or solicitee, simply as giver. In the first stage, the giver may receive a flyer of the upcoming visit and, if so, notices the flyer with probability r e (0,1]. In the second stage, the solicitor visits the home. The giver opens the door with probability h. If she did not notice the flyer (or did not receive one), h is equal to a baseline probability ho e (0,1). If she noticed the flyer, she can adjust the probability to h e [0,1] at a cost c (h), with cQiq) = 0, c'(ho) = 0, and c"{ ■) > 0. That is, the marginal cost of small adjustments is small, but larger adjustments have an increasingly large cost. We do not require symmetry around ho and we allow for corner solutions at h = 0 or h = 1. If the giver is present, she donates an amount g > 0. If she is absent, there is no in-person donation (g = 0). The giver can donate through other channels, such as via mail or online, after learning about the charity from the solicitor or the flyer. The giver has utility (1) U(g,gm) = u (W -g -gm) + av (g + BgmiG-i) -s(g). 8 QUARTERLY JOURNAL OF ECONOMICS The utility of private consumption, u, is derived from the pregiving wealth W minus the donations given to the solicitor (g) and through other channels such as mail igm). Giving through other channels gm involves additional costs, such as finding an envelope and stamp, equal to (1—9)gm, with 0 < 9 < 1. The charity therefore receives 9gm.1 The private utility satisfies standard properties: u'{ ■ ) > 0 and u"( ■ ) < 0. Notice that the utility of private consumption can include the utility from infra-marginal giving to other charities.2 The utility of giving to the charity, v, allows for pure and impure altruism (warm glow), or prestige (Harbaugh 1998). Since the experiment is not designed to separate pure altruism, impure altruism, or prestige but altruism from social pressure, we use a specification that is general enough to encompass both. We also allow for negative social preferences, or spite (Levine 1998), towards the charity. In the case of pure altruism, the agent cares about the total contributions to the charity, G_; + g + 9gm, where G_; is the giving of others. In this case, we can think of v (G_j + g + 9gm) as the production function of the charity, which is increasing in the donation g but has decreasing returns: v'g(-, ■) > 0, v'gg{-, ■) < 0, and lim^oo v' (g, ■) = 0. The parameter a > 0 denotes the level of altruism,3 and the overall utility from giving is av (G-i +g + 9gm). In the case of impure altruism, the agent cares about the warm glow from givingg. Hence, v (■) does not necessarily depend on the giving of others, G_;, and a > 0 captures the intensity of the warm glow. We make the same assumptions v'g > 0, v'g£ < 0, and lim^oo v' (g, ■) = 0.4 Finally, in the case of spite towards the charity, the agent dislikes giving to the charity. The utility is av (G_j + g + 9gm), with a < 0 capturing the intensity of spite. It is natural to assume that the disutility of giving increases with the donation in a convex 1. The key results generalize to a fixed cost of giving by mail, but the algebra is more complex. 2. We allow for giving to exceed current wealth, that is, the case g + gm > W. In practice, this case is unlikely to matter. 3. The parameter a can also capture the belief of the donor about the quality of the charity. 4. Under the warm-glow model, an alternative interpretation of 6 is that the charity receives the full amount gm (i.e., there are no costs of giving via mail), but the impersonal mean lowers warm glow by a factor 6. TESTING FOR ALTRUISM AND SOCIAL PRESSURE 9 manner: v'g > 0 and v'gS > 0. Here, we are abusing notation since the function v differs for a > 0 (altruism) and a < 0 (spite); it is concave in the first case and convex in the second. When the distinction is important, we use v+ to denote the function for a > 0, v~ to denote the function for a < 0, and v to denote the function that equals v+ for a > 0 and v~ for a < 0. Notice that it is important to consider the case of spite because, unlike in a standard model of giving, even spiteful individuals may give to the charity if social pressure is high enough. The third element in the utility function is social pressure. The giver pays a utility cost s(g) = S ■ (gs — g) ■ lg 0 for not giving or only a giving small amount g while the solicitor is present. The cost is highest for the case of no donation (s(0) = Sgs), then decreases linearly in g, and is 0 for donations of gs or higher. The giver does not incur a social pressure cost if she is away from home during the visit. The special case of S = 0 (no social pressure) and a = 0 (no altruism or warm glow) represents the standard model. We further assume that the giver is aware of her own preferences and rationally anticipates her response to social pressure. Giving in Person. We solve the model working backward. In the second stage, conditional on being at home and answering the door, the giver chooses g to maximize (1). Notice that conditional on answering the door, the giver always prefers to donate in person because mail donations involve an additional cost (1 — 0) and do not eliminate the social pressure cost. We characterize optimal giving g* as a function of the parameters a and S in Lemma 1A in the Appendix. (The thresholds a(S), a(S), and a are also defined in the Appendix.) Figure I illustrates the case of linear private utility u and v+' (0) = v~' (0), which are the assumptions used for the structural estimation. Giving g* increases in altruism. When altruism is sufficiently low (a < a(S)), the individual does not give. For a higher level of altruism (a(S) < a < a(S)), the individual gives a positive amount, but less thangs. For even higher altruism (a (S) a), the donor gives more than gs. Any giving above gs is due to altruism (hence the threshold a > 0 does not depend on the social pressure cost S), while donations smaller than gs may be due to altruism or social pressure. Giving can occur also with Figure I Regions of Giving g and Probability of Home Presence h Figure indicates the different regions for giving, no giving [g = 0], small giving [0 < g < g8], giving equal to g8, and large giving [g > g8], and the different regions for the probability of being at home, avoiding the solicitor [h < ho], and seeking the solicitor [h > ho]. The regions are a function of the altruism parameter a and of the social pressure parameter S. spiteful agents (a < 0) if the social pressure cost S is large enough (S > u' (W) and hence a < 0). Giving via Mail. Conditional on not being at home, a giver who was informed about the fund-raising via a flyer decides whether to give via mail gm. Note that the only reason to give via mail is altruism. Giving via mail is increasing in altruism, provided 9 > 0. For given altruism a, the level of giving via mail received by the charity (Og^ (a)) is always smaller than the level of giving in person conditional on being at home (g* (a,S)) (Lemma IB). Presence at Home. In the first stage, the giver opens the door with probability ho if the visit is unanticipated (no flyer or, with probability 1 — r, despite a flyer). If the visit is anticipated (flyer), she optimizes h given her utility from being at home, u (W —g*) + av (g*,G_i) — s (g*), and her utility from not being at home, u(W -g*m) +av (9g*m,G-i): TESTING FOR ALTRUISM AND SOCIAL PRESSURE 11 max h [u (W - g*) + av (g*, G_,) - s (g*)] + (1 - h) [u (W - g*m) /iG[0,l] +av (0g*m,G-i)]-c(h). We characterize the optimal probability of being at home, h*(a, S), in Lemma 2 (see Appendix). It is (weakly) increasing in altruism: the more the giver cares about the charity (or the warm glow), the more likely she is to be at home. The exact pattern depends on the degree of social pressure (Figure I). In the case of no social pressure (S = 0), sufficiently altruistic agents, a > a(0), give if at home and actively seek to be at home (h* > ho). The probability of being at home is increasing in the altruism up to the corner solution h = 1. Less altruistic agents, a < a(0), instead, do not plan to give. They are indifferent as to being at home or not, and hence do not alter the baseline probability h0. In the case of social pressure (S > 0), agents with low altruism a < a (S) do not give and avoid the fund-raiser in order not to pay the social pressure cost. More altruistic agents with a (S) < a < ao (S) give a small amount but prefer to avoid the fund-raiser. Their giving is either entirely or partly due to social pressure. Agents with sufficiently high altruism, a > a0 (S), care enough about the charity that they seek the interaction with the fund-raiser, despite the fact that social pressure may distort their giving upward. Opt-out. So far we have assumed that it is costly to reduce the probability of being at home. We now allow agents to costlessly reduce the probability of being at home to 0, for example, via a Do-Not-Disturb check box on the flyer. Formally, c (0) =0 and c (h) as above for h > 0.5 Opting out does not affect giving g*{a) (conditional on being at home) or g*m (a) (conditional on not being at home) but only the probability of being at home h* (a). As characterized in Lemma 3, h* (a) remains the same as without the opt-out option if there is no social pressure and, hence, no reason to opt out. In the presence of social pressure, however, the agent opts out for low altruism, a < ao (S), since the interaction with the fund-raiser lowers utility. For higher altruism levels, instead, the agent derives positive 5. This formalization allows a costless reduction of h to 0 but not to other levels. This is not a restriction because agents who prefer to lower h below ho (at a positive cost) will strictly prefer to lower h to 0 at no cost. 12 QUARTERLY JOURNAL OF ECONOMICS utility from giving. Hence, she does not opt out and the solution is the same as in Lemma 2. Opting out also allows us to distinguish social pressure from self- or other-signaling. In our model, checking the opt-out box has no cost to the agent. Under self- and other-signalling, instead, opting out is costly since it signals avoidance of giving. If the cost is high, the agent will never opt out, and the opt-out treatment reduces to the simple flyer treatment. Testable Predictions. To complete the model, we assume that the population of agents is heterogeneous in a with c.d.f. F. We emphasize two special cases: (i) Altruism and No Social Pressure, that is, the standard assumption S = 0, but a positive probability of altruistic individuals with a > a(0); (ii) Social Pressure and Limited Altruism, that is, allowing for social pressure S > 0, but requiring 0 probability of altruistic individuals with a > a0 (S).e The following propositions outline testable predictions regarding the key outcomes, home presence and giving. Our first prediction compares the probabilities of being at home in the treatments without flyer, P(H)NF, with flyer, P(H)F, and with opt-out flyer P(H)0o- PROPOSITION 1. With Altruism and No Social Pressure, the probability P (H) is higher with flyer than without: P (H)F = P(H)qo > P(H)NF. With Social Pressure and Limited Altruism, P(H) is lower with flyer and lowest with opt-out: P(H)NF>P(H)F>P(H)00. In the case of Altruism and No Social Pressure, the flyer increases home presence relative to the control group since some agents seek to meet the solicitor. The opt-out option has no differential effect since no one avoids the solicitor. Under Social Pressure and Limited Altruism, the opposite is true: the flyer lowers home presence as agents seek to avoid the solicitor. In this case, the costless opt-out possibility lowers the presence at home further.7 In general, the probability of being at home is higher for the flyer group if the altruism force dominates the social pressure force, but the opt-out option always weakly lowers the presence at home. 6. In case (ii), we also require F(ao (S)) — F(a (S)) > 0 to eliminate a trivial case. 7. A sufficient (not necessary) condition for the inequality P (H)F > P (H)00 to be strict is a positive mass of households with a in the left neighbourhood of ao. TESTING FOR ALTRUISM AND SOCIAL PRESSURE 13 The next proposition illustrates the impact of the different treatments on the unconditional probability of in-person giving, P(G). PROPOSITION 2. With Altruism and No Social Pressure, the probability P (G) is higher with flyer than without: P (G)F = P (G)oo > P (G)nf • With Social Pressure and Limited Altruism, P(G) is lower with flyer and lowest with opt-out: P(G)NF>P(G)F>P(G)00. Under Altruism and No Social Pressure, the flyer and opt-out treatments lead to the same probability of giving, since there is no reason to opt out in the absence of social pressure. The probability of giving in these two flyer treatments is higher than without flyer because some agents seek to stay at home. Under Social Pressure and Limited Altruism, instead, the probability of giving is lower with flyer and lowest with an opt-out flyer. In general, the net effect of a flyer depends on whether the giving is more due to real altruism (which works to increase giving) or to social pressure (which has the opposite effect). The third prediction regards the probability of giving conditional on being at home, P (G\H). PROPOSITION 3. The probability of giving conditional on being at home is higher with flyer than without: mm(P(G\H)F, P(G\H)00)>P(G\H)NF. Altruism and social pressure both lead to increases in the conditional giving probability with flyer: altruistic people are more likely to be at home, and nongivers sort away from home. Hence, conditionally on reaching an agent at home, giving is higher with than without flyer. The next proposition focuses on gift size. We distinguish between large donations, defined as g > gs, and small donations, 8 P(GHI)NF (with strict inequality if F(a) < 1 ). (ii) The unconditional probability of a small donation, P(GL0), is identical under the simple flyer treatment and the flyer with opt out (P(GL0)F = P(GL0)00) if S = 0 , but higher under the simple flyer (P(GL0)F > P(GL0)00) if S > 0 (and ^(a0 (S)) -Fig (S)) > 0). 14 QUARTERLY JOURNAL OF ECONOMICS A flyer (with or without opt-out option) increases large donations given that altruistic donors increase their probability of being at home. The impact of a flyer on small donations is less obvious since small donations can reflect moderate altruism or social pressure. A flyer with opt-out unambiguously lowers the probability of small donations relative to a simple flyer, given that it simplifies the sorting out of donors motivated by social pressure. The last proposition characterizes the probability of giving via mail. PROPOSITION 5. The unconditional probability of a donation while not at home P(Gm) satisfies 0 = P(Gm)jvF < P(Gm)p < P(Gm)oo- Without a flyer, giving via mail is zero because the giver is only informed about the fund-raiser if she is at home. A flyer informs the giver about the fund-raiser and, hence, she may give even if not at home, so long as she is sufficiently altruistic. Giving via mail is at least as high if the flyer offers opting out as with the simple flyer because some of the individuals who opt out because they would have given more than they wish in person give a smaller amount via mail. III. Experimental Design Charities. The two charities in the fund-raising treatments are La Rabida Children's Hospital and the East Carolina Hazard Center (ECU). Although both charities are well-respected regional charities, we chose them so that most households in our sample would prefer one (La Rabida) to the other (ECU). To document these preferences, in our 2008 survey treatments we asked respondents to rank five charities from 1 (least liked) to 5 (most liked). The charity with the highest average rank is the La Rabida Children's Hospital (average rank 3.95) followed by Donate Life (rank 3.79), and the Seattle Children's Hospital (rank 3.47). At the bottom of the rank, below the Chicago Historical Society (rank 2.96), is the East Carolina Hazard Center (rank 2.54).8 La Rabida appears to be highly liked both because it is an in-state charity well known to residents in the area around Chicago, and also 8. We obtain similar results when we ask the respondents to allocate $1 that an anonymous sponsor has pledged to give to one of the five charities. (We followed up and delivered the donations.) Out of 255 respondents, 147 pledge the donation to the La Rabida charity, and only 7 choose the ecu charity. TESTING FOR ALTRUISM AND SOCIAL PRESSURE 15 because it provides health benefits to children; neither condition applies to ECU. Door-to-Door Fund-Raising. Our experiment uses a door-to-door campaign because it offers the easiest implementation of the design. While door-to-door campaigns are common and previously studied in economics (Landry et al. 2006), it is hard to quantify how much money they raise. To provide some evidence, our survey asked respondents to recall how many times in the past 12 months people have come to your door to raise money for a charity. We asked similarly phrased questions about giving via phone, via mail, and through other channels, such as employer or friends. Of 144 respondents who answered all of these questions, 76% stated that they had received at least one such visit, and 48% of respondents reported at least three such visits. This frequency is smaller than but comparable in magnitude to other solicitation forms: phone (86% received at least one call), mail (95% with at least one mailing), and other forms (83% with at least one contact). We also asked how much the respondents gave to these solicitors in total over the last 12 months. Of the respondents, 40% reported donating to a door-to-door campaign, compared to 28% in response to phone solicitations, 53% in response to mail solicitations, and 76% in response to other solicitations. The average reported total door-to-door donation in the past 12 months (including nondonors) is $26, compared to $101 by phone, $1,012 by mail, and $2,063 by other means. However, this estimate is very sensitive to a small number of individuals reporting large sums given (in two cases $50,000 and $60,000) which could be due to measurement error or self-aggrandizing claims. If we cap the donations at $1,000, the numbers are $26 by door-to-door, $66 by phone, $115 by mail, and $295 by other means. Hence, door-to-door solicitations are quite common, at least in the area where the survey took place, and they raise a smaller, but not negligible, amounts. Logistics. We employed 92 solicitors and surveyors, mostly undergraduate students at the University of Chicago, who were paid $9.50 per hour. All solicitors elicited contributions within at least two treatments, and most over multiple weekends, and similarly for surveyors. Each solicitor and surveyor's participation in the study typically followed four steps: (1) an invitation to work as a paid volunteer for the research center, (2) 16 QUARTERLY JOURNAL OF ECONOMICS an in-person interview, (3) a training session, and (4) participation as a solicitor and/or surveyor in the door-to-door campaign. Details about the recruitment process are in Online Appendix B. The field experiment took place on Saturdays and Sundays between April 2008 and October 2008 (both charity and survey treatments) and then again between April 2009 and November 2009 (survey treatments only). The locations are wealthy towns around Chicago.9 Each solicitor is assigned a list of typically 25 households per hour on a street, for a daily workload of either 4 hours (10 am-noon and 1-3 pm) or 6 hours (10 am-noon and 1-5 pm). Every hour, the solicitor moves to a different street in the neighborhood and typically enters a different treatment. Solicitors do not know whether a treatment involves a flyer, though they can presumably learn that information from observing flyers on doors. Solicitors are trained to either do charity treatments or survey treatments (with a small number trained to do both). A solicitor assigned to La Rabida on a given day will only do treatments for La Rabida, and similarly for ECU or survey. Treatments. In the treatments without flyer, solicitors knock on the door or ring the bell and, if they reach a person, proceed through the script (see Online Appendix C). They inform the household about the charity (La Rabida or ECU), ask if they are willing to make a donation, and if they receive a gift leave a receipt. In the survey treatment, the solicitor inquires whether the household member is willing to respond to survey questions about charitable giving. The solicitor informs the household member about the duration of the survey (5 or 10 minutes) and about the payment for completing the survey, if any ($10, $5, or none). In the flyer treatments, the solicitor's script is identical, but in addition a different solicitor leaves a flyer on the doorknob on the day before the solicitation. The professionally prepared flyer indicates the time of the upcoming fund-raising (or survey) visit within a one-hour time interval. Figure II provides examples of two flyers used for the fund-raising treatment and two flyers used for the 2008 survey treatments.10 In the opt-out treatments, the 9. Burr Ridge, Countryside, Flossmoor, Kenilworth, Lemont, Libertyville, Oak Brook, Oak Forest, Oak Park, Orland Park, Park Ridge, Rolling Meadows, Roselle, Schaumburg, Skoke-Evanston, and Willowbrook. 10. For a small number of observations, the flyer does not indicate the exact time of the visit, only that there will be a visit in the next two weeks. Results for this subgroup are qualitatively similar to the results for the flyer with the one-hour TESTING FOR ALTRUISM AND SOCIAL PRESSURE 17 CHEllLlri't HAIIITIL Fund raising Campaign for La Rabida Children's Hospital Fundraisers will visit this address tomorrow ( / ) between and to raise funds for La Rabida Children's Hospital. □Check riiji bos if you Da nnt want tE> he disturbed. LARABIOA Fundraising Campaign for La Rabid a Children's Hospital Fundraisers will visit this address tomorrow ( / ) between and to raise funds for La Rabid a Children's Hospital. TKF. TrN[VT.JlSTTY OF CHICAGO JfJlll UNLYEHSITY OF CHICAGO University of Chicago Study Researchers will will visit this address tomorrow { / ) between and to conduct a 15 minute survey. University of Chicago Study Researchers will visit this address tomorrow ( / ) between and to conduct a 10 minute survey. You will be paid $10 for your participation. Figure II Flyer Samples Two examples of flyers for the 2008 fund-raising treatments (top row) and flyers for the 2008 survey treatments (bottom row). The top-left flyer is for the opt-out treatment, while the top-right flyer is for a flyer treatment. The bottom-row flyers are both for a 10-minute survey with flyer, the left one without payment, the right one for a $10 payment. 18 QUARTERLY JOURNAL OF ECONOMICS a Fundraising Treatments Fundraising Baseline La Rabida Fund raising Flyer La Rabida Fundraising Flyer with Opt-Out La Rabida Fundraising Baseline ECU Fundraising Flyer ECU Fundraising Flyer with Opt-Out ECU Survey Treatments Baseline 10-Minute Survey SO Flyer 5-Minute 10-Minute SO SO $10 b Baseline 5-Minute Survey $0 S5 Figure III Experimental Treatments (Top) 2008, (Bottom) 2009 Summary of the treatments run in the door-to-door field experiments in 2008 (charity and survey) and run in 2009. La Rabida and ECU are the names of the two charities for which the funds were raised. I Survey Treatments 5-Minute Survey 10-Minute Survey Opt-out 5-Minute Survey SO so so flyer has a box "Check this box if you do not want to be disturbed." If the solicitors find the box checked, they do not knock on the door. The charity treatments are summarized in the top part of Figure III. The survey treatments are aimed at estimating the elasticity of the presence at home and of the response rate to incentives. In Section V, we use these elasticities to estimate the social pressure interval of visit. We therefore present the results combining these treatments. Excluding the observations with the two-week window does not change any of the results. TESTING FOR ALTRUISM AND SOCIAL PRESSURE 19 and altruism parameters. The survey questions are mostly about patterns of charitable giving in 2008 and about voter participation in 2009.11 Figure III summarizes the survey treatments run in 2008 (top) and 2009 (bottom). Sample and Randomization. We reached a total of 8,906 households in the charity treatments, 2,018 households in the 2008 survey treatments, and 10,594 households in the 2009 survey treatments. From this initial sample, we exclude 1,391 observations in which the households displayed a no-solicitor sign (in which case the solicitor did not contact the household) or the solicitor was not able to contact the household for other reasons (including, for example, a lack of access to the front door or a dog blocking the entrance). We also exclude 559 solicitor-day observations for 5 solicitors with substantial inconsistencies in the recorded data.12 The final sample includes 7,668 households in the charity treatments, 1,865 households in the 2008 survey treatments, and 10,035 households in the 2009 survey treatments. The charity field experiment took place in 2008 in three waves. In the first wave (April 27 to June 1), we solicited for both charities (with equal weights) and all three treatments (approximate weights of 40% for no-flyer, 35% flyer, 25% opt-out). In the second wave (July 13 to August 23), we solicited only for La Rabida and ran only no-flyer and flyer treatments (with equal weights). In the third wave (September 6 to October 18) we solicited for both charities (the ECU charity is over-sampled with weight 75%) and all treatments (approximate weights of 25% for no-flyer, 50% flyer, 25% opt-out). The script of the randomization and the flyer design is the same throughout. Within each of these waves, the randomization of the treatments takes place for each solicitor-hour and is at the street level within a town. IV. Reduced-Form Estimates We report the differences across the treatments in the share of households answering the door, the empirical counterpart of 11. In the 2009 survey treatments, about half of the flyers specify that the survey will be about voter participation. The results are similar for the two types of flyers and hence we pool them in the analysis. 12. These five solicitors indicate the presence of flyers on the door or on the floor also for households in the no-flyer treatment. 20 QUARTERLY JOURNAL OF ECONOMICS P(H), and the share of households giving to the charity in person, corresponding to P(G). We also present results on giving conditional on being at home, corresponding to P(G\H), on the frequency of small and large donations, P(GL0) and P(GHI), and on giving via mail and Internet, P (Gm). Table I presents the summary statistics on the key treatment outcomes. The rate at which the respondents open the door varies between 41% and 42% in the baseline treatments for La Rabida, ECU, and in the 2008 survey treatment. Since households did not know the task at hand, these averages ought, indeed, to be close. The share answering the door is smaller for the flyer treatment and smaller yet for the opt-out treatment. The share of givers is substantially smaller for the ECU charity than for the La Rabida charity, consistent with our survey evidence that the La Rabida charity is more liked than the ECU charity. For ECU, the share of givers is substantially lower in the opt-out treatment than in the other treatments. For La Rabida, instead, giving is somewhat higher in the opt-out treatment. In the survey treatments, the share opening the door and the share completing the survey are generally larger for the treatments with higher pay and shorter duration both in 2008 and 2009. While the summary statistics provide suggestive evidence on the impact of the treatments, the raw statistics are potentially confounded with randomization fixed effects. As discussed in Section III, treatments were randomized within a date-solicitor time block, but not all treatments were run in all time periods. Hence, estimates that do not control for the randomization fixed effects may be confounded, for example, by time effects—we ran more ECU treatments later in the sample when donation rates also happened to be lower. It turns out that all directional effects indicated in the summary statistics, except for the higher giving to La Rabida under opt-out, are confirmed once we add the randomization fixed effects. The benchmark empirical specifications (Table II) control for solicitor i and day-town t fixed effects.13 The identification comes from within-solicitor, within-day variation in treatment. The vector of control variables Xijt^ includes dummies for the hourly time blocks h (10 am, 11 am, 1 pm, 2 pm, 3 pm, and 4 pm), and dummies for a subjective rating by the solicitor of the quality 13. On almost all days, we visited only one town, so that the day-town fixed effects are essentially equivalent to day fixed effects. TABLE I Summary Statistics Panel A: Fund-raising treatments to I C! S i b to o 8 ťa ťa Variable: Share of households answering the door Share of households giving In person Number of households giving via mail or internet Sample: Pooled ECU La Rabida Pooled ECU La Rabida ECU La Rabida (1) (2) (3) (4) (5) (6) (7) (8) Baseline (no-flyer) treatment 0.4090 0.4228 0.4032 0.0629 0.0507 0.0680 Zero donations One ($25) (AT = 3166) (N = 946) (N = 2220) across all donation across Flyer treatment 0.3753 0.3993 0.3628 0.0585 0.0460 0.0650 treatments all treatments (N = 3432) (AT = 1172) (N = 2260) Flyer with opt-out treatment 0.3355 0.3503 0.3174 0.0514 0.0289 0.0788 (N = 1070) (N = 588) (N = 482) N AT = 7668 AT = 2706 AT = 4962 AT = 7668 AT = 2706 AT = 4962 AT = 2706 AT = 4962 to TABLE I (continued) Panel B: Survey treatments Share of Share of households households Share of hhs. Share of hhs. answering the completing the answering the completing the Variable: door survey door survey 2009 survey 2008 survey treatments (1) (2) treatments (3) (4) Baseline ($0/10min) treatment 0.4135 (N = 607) 0.0972 Baseline ($0/5min) 0.3915 (N = 1420) 0.1472 Flyer ($0/10min) treatment 0.3681 (N = 489) 0.1186 Baseline ($5/5min) 0.3956 (N = 910) 0.1231 Flyer ($0/5min) treatment 0.392 (N = 449) 0.1714 Flyer ($0/10min) 0.3771 (N = 769) 0.1430 Flyer ($10/10min) treatment 0.4156 (N = 320) 0.1719 Flyer ($0/5min) 0.3852 (N = 1703) 0.1679 N N = 1865 N = 1865 Flyer ($5/5min) 0.4057 (N = 1856) 0.2085 Flyer ($10/5min) 0.4428 (N = 673) 0.2600 Opt-out ($0/5min) 0.3420 (N = 1374) 0.1434 Opt-out ($5/5min) 0.3639 (N = 1330) 0.1797 N = 10035 Af= 10035 Notes. Summary statistics for the variables of the experiment. "ECU" and "La Rabida" indicate the two charities in the experiment, "ECU" is an out-of-state research center on hurricanes, "La Rabida" is an in-state children's hospital. TABLE II Results for Fund-Raising Treatments Specification: OLS regressions Indicator for Indicator for Indicator for giving Amount given Dep. var.: answering the door giving Small amount (< $10) Large amount (> $10) (including $0) (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Flyer treatment -0.0387 -0.0011 -0.0033 0.0022 -0.1459 (0.0137)*** (0.0062) (0.0052) (0.0035) (0.1357) Flyer with opt-out -0.0967 -0.0195 -0.0193 -0.0002 -0.3041 treatment (0.0194)*** (0.0084)** (0.0081)** (0.0051) (0.1653)* Indicator ecu 0.01 0.0041 -0.0249 -0.0263 -0.0127 -0.0107 -0.0123 -0.0155 -0.7611 -0.9767 charity (0.0143) (0.0234) (0.0049)*** (0.0085)*** (0.0053)** (0.0085) (0.0032)*** (0.0052)*** (0.1368)*** (0.2014)**' Flyer treatment -0.0365 0.0006 -0.0045 0.0051 0.1154 * ecu charity (0.0313) (0.0094) (0.0076) (0.0045) (0.1240) Flyer with opt-out -0.089 -0.0183 -0.0222 0.0039 -0.0907 * ecu charity (0.0271)*** (0.0100)* (0.0098)** (0.0058) (0.1268) Flyer treatment -0.0396 -0.0019 -0.0028 0.0009 -0.2545 * La Rabida charity (0.0144)*** (0.0078) (0.0066) (0.0046) (0.1841) Flyer with opt-out -0.106 -0.0202 -0.0161 -0.0042 -0.4573 * La Rabida charity (0.0319)*** (0.0132) (0.0128) (0.0087) (0.2885) Omitted treatment No-flyer, La Rabida No-flyer, La Rabida No-flyer, La Rabida No-flyer, La Rabida Mean of dep. var. for omitted treatment 0.413 0.0717 0.0414 0.0414 0.0215 0.0215 1.161 1.161 Fixed effects for X X X X X X X X X X solicitor, date- location, hour, and area rating n n = 7668 n = 7668 n = 7668 n = 7668 n = 7668 n = 7668 n = 7668 n = 7668 n = 7668 n = 7668 Notes. Estimates for a linear probability model, with standard errors clustered by solicitor-date, in parentheses. The omitted treatment is the baseline no-flyer fund-raising treatment for the La Rabida charity. The regressions include fixed effects for the solicitor, for the date-town combination, for the hour of day, and for a subjective rating of home values in the block. * significant at 10%; ** significant at 5%; *** significant at 1%. 24 QUARTERLY JOURNAL OF ECONOMICS of the houses visited in that hour block on a 0-10 scale. The latter controls provide a rough measure of the wealth level of a street within a town. We run the OLS regression (2) yij,t,h = d + rTiAh + f3ECudEcu + Vi + *t + BXi>t>h + where the dependent variable yijjt,h is? alternatively, an indicator for whether individual j opened the door (yH), gave a positive amount to the charity (yG), gave a small amount iyGh0), or gave a large amount iyGm). The vector Tijt^ contains indicators for the various fund-raising treatments, with the baseline no-fiyer treatment for La Rabida as the omitted group. As such, the point estimates for r are to be interpreted as the effect of a treatment compared to the baseline.14 We cluster the standard errors at the solicitor x date level. We also estimate the impact of the fund-raising treatment separately for the two types of charities (ECU and La Rabida), using the following OLS regression model: Jij,t,h = a + ^LaBTi,t,hdhaR + PecuoIecu + ^ECvT"i,t,hdECV (3) +r)i + Xt+ BXith + £iJjtjh, where dch is an indicator variable for charity ch e {LaR, ECU}. The omitted treatment is the no-fiyer treatment for La Rabida. In Figures IVA and IVB, we plot the estimated coefficients from this specification. The estimated impact for the baseline no-flyer treatment for La Rabida is a, from specification (3) with no fixed effects and controls. The estimated impact for the other treatments k are a + iLaR for La Rabida and a + /3 + Jecu for ECU. Answering the Door. For both charities, a flyer announcing the visit reduces the share of households opening the door by about 4 percentage points relative to the baseline treatment with no flyer (Figure IVA). As Table II shows, the difference is statistically significant at conventional levels. The share of households opening the door is further lowered, by an additional 5 to 6 14. The specification assumes that the impact of the fixed effects on the relevant outcomes is additive. We obtain essentially identical results using solicitor-time-date fixed effects, reported in the Online Appendix. These fixed effects, however, do not allow us to identify the difference in outcomes between La Rabida and ECU, because on any given date each solicitor raised money for only one charity. TESTING FOR ALTRUISM AND SOCIAL PRESSURE 25 a 45% 40% 35% 30% 25% 20% 15% 19% 5% 0% Baseline (N=946/2,220) Flyer (N=1,173/2,370) nCen:er ■"or Natural Hazards Mitigation Research at East Carolina Univfirsity ■ La Rabida Children's Hospital Opting out Flyer with Opt Out (N=588/482) Baseline (N =946/2,2 20) Flyer Flyer with Opt Out (N=1r172/2,260) (N=58b7482) — Csi ■ I.L ■ for Natural Hazards Mitigation Research a1 t: rJ A I C rJ ■ C If Tl University ■ La Rabida Children's Baseline (N=946/2220) Flyer Flyer with Opt Out (N=1>172/2,260) (N=588/482) a Canter Far Natural Hazards MM safari Figure IV Frequency of (A) Answering the Door, (B) (Unconditional) Giving, and (C) Giving Conditional on Answering the Door Panel A presents the percent of households that answer the door under different treatment. The third set of bars (opt-out treatment) also shows the percent opting out (shaded colors on top). Panel B displays the percent that give to the charity out of all the households in the treatment group (including those not answering the door). Panel C shows giving conditioned on answering the door, which equals the ratio of the estimated shares of unconditional giving (Figure IVB) and of households answering the door (Figure IVA). All estimates are obtained from regressions that control for randomization fixed effects. 26 QUARTERLY JOURNAL OF ECONOMICS percentage points, by the presence of an opt-out condition (Check this box if you do not want to be disturbed) on the flyer. Hence, the flyer and the opt-out conditions lower the probability of opening the door by, respectively, 10% and 25%, an economically large effect that is similar for both charities. We interpret this evidence as suggestive of social pressure: when informed of a visit by a solicitor, households attempt to avoid the interaction, especially when doing so has little cost, as in the opt-out treatment. Notice that the reduction in the probability of opening the door in the presence of a flyer can be due to two factors: a lower probability of being at home, or a lower probability of opening the door conditional on being at home. The variable we measure captures the sum of these two effects. Opting Out. Figure IVA also presents evidence on the share of subjects in the opt-out treatment that check the opt-out box: 12% of all households for the La Rabida charity and 9.9% for the ECU charity. Households that, in the flyer treatment, are explicitly avoiding the solicitor by not answering the door, instead use the opt-out option when available. This result is consistent with the assumption that checking the opt-out box is a cheaper way to avoid a solicitor. Unconditional Giving. Figure IVB presents the results on the unconditional giving probability, including households that do not answer the door. Not surprisingly, giving is higher for the preferred charity, La Rabida, than for ECU in each treatment. The pattern of effects across treatments is similar for the two charities. The baseline and the flyer treatments have essentially the same share of giving. The lack of a difference between these treatments is estimated quite precisely because we overweighted the baseline and flyer treatments. The opt-out treatment, instead, lowers giving by 2 percentage points for both charities. This difference is statistically and economically significant (see Table II): the effect amounts to a reduction in giving of about a third relative to the other treatments. The first result—that the flyer per se does not affect giving— is consistent with both social pressure and altruism affecting charitable giving. The advance notice increases the presence at home of the altruistic givers and lowers the presence of those who give due to social pressure. To the extent that these two forces are of about the same size, we expect no overall impact. Note that TESTING FOR ALTRUISM AND SOCIAL PRESSURE 27 this result does not contradict our previous finding that the flyer significantly reduces the share of households opening the door. In the presence of social pressure costs, nongivers also avoid being at home when notified with a flyer. This avoidance does not impact the probability of giving, but it lowers home presence. The second result—that the opt-out option significantly lowers giving—points to the importance of social pressure: in the opt-out treatment the cost of avoiding the fund-raiser is substantially lowered, and giving decreases proportionally. If giving was primarily due to altruism, the opt-out option should not affect giving rates or levels. Conditional Giving. For both charities, the share of households that give among those who answer the door15 is higher in the treatments with flyer than in the baseline treatment (Figure IVC). This increase is consistent with Proposition 3 since the flyer allows sorting in by donors who want to give and sorting out by those who do not want to give. In the opt-out treatment, however, conditional giving is lower. This effect is inconsistent with Proposition 3, though not statistically significant at conventional significance levels. Amount of Giving. In our model, individuals who give due to social pressure give at mostgs, whereas individuals who give due to altruism may contribute higher amounts. Hence, relative to the control treatment, the flyer treatment may both decrease smaller donations (sorting out of social-pressure givers) and increase larger donations (sorting in of altruists). The opt-out treatment, which further facilitates sorting out but not sorting in, should lower the share of small donations but not the share of larger donations (Proposition 4). To test these predictions, we split donations based on the median amount given, $10, and label donations smaller than (or equal to) $10 as small and donations larger than $10 as large. Figure V presents the results. In the baseline treatment, 4% of households give small donations, and 2% give large donations. The percentage giving a small donation decreases slightly in the flyer treatment and decreases by 2.1 percentage points in the opt-out treatment. Hence, the opt-out option more than halves the 15. The conditional giving for each treatment is the estimated unconditional giving, Figure IVB, divided by the estimated share of households answering the door, Figure IVA. 28 QUARTERLY JOURNAL OF ECONOMICS 4.5% 4.0% 3.5% 3.0% 2.5% 2.0% 1.5% 1.0% 0.5% 0.0% Baseline Flyer Flyer with Opt Out (N=3,166) (N=3,432) (N=1,070) Figure V Frequency of Giving: Small versus Large (Pooled) Figure presents the results on (unconditional) giving of small (<$10) and large (>$10) donations across treatment. The estimates are obtained from regressions that control for randomization fixed effects. likelihood of a small donation, a significant difference as shown in Table II. The pattern is very different for large donations. The flyer somewhat increases the incidence of larger donations, though not significantly, and the opt-out option has no effect. This pattern is consistent with Proposition 4. Figure VI presents additional information on the distribution of the amount given across treatments. The opt-out treatment, compared to the baseline treatment, induces a decrease in the donations up to $10, but no change in larger donations. The histogram also provides evidence of bunching at $5 and $10. In the structural model, we use this information and consider $10 as the amount that eliminates all social pressure from not giving, gs. Finally, in columns (9) and (10) of Table II we consider the effect of the different treatments on the amount given. There is no significant effect of the flyer treatment, and a marginally significant reduction in giving with the opt-out treatment. Robustness. We examine the robustness of specification (2) in Table II to: (i) omitting the dummy for the ECU charity; and (ii) introducing fixed effects for solicitor-day-location fixed effects, TESTING FOR ALTRUISM AND SOCIAL PRESSURE 29 w o rz CD =1 O 1.6% 1.2% 0.8% 03 =J O CD g> 0.4% 4—» c u CL 0.0% □ Baseline (6.26% positive donation) ■ Flysr (6.14% positive donation) □ Opt-Out (4.28% positive donation) rJk 20-50 50 n 50-i 5-10 10 10-20 20 Amount Donated ($) Figure VI Distribution of Amounts Donated Figure displays the distribution of donation amounts across different treatment groups. Each bar indicates the percentage giving the specified amount out of all households in the treatment. The figure does not display the share of households donating $0. The estimates are obtained from regressions that control for randomization fixed effects. which allows for different solicitor fixed effects on different days or towns. Online Appendix Table 1 shows that the estimates are essentially identical. In Online Appendix Table II, we estimate separately the results for the three experimental waves in 2008: April to June (wave 1), July and August (wave 2), September and October (wave 3). The results in the first and second waves for the flyer treatment are consistent (in wave 2 we did not run an opt-out treatment): in both waves the share answering the door significantly decreases in response to a flyer, and there is no effect on the share giving. In the third wave, we obtain similar effects as in the previous waves for the share answering the door, but somewhat different effects on giving. In this time period, which coincides with the peak of the financial crisis, the level of giving is substantially lower. Given this lower level, a flyer (marginally) increases giving, mostly in the form of larger donations; a flyer with opt-out option still lowers giving, though less so than in previous waves. These results are consistent with the crisis reducing giving due to lower social pressure cost of turning away a solicitor ("sorry, times are tough"), but the sample size in this period is too small to draw firm conclusions. 30 QUARTERLY JOURNAL OF ECONOMICS Giving via Mail or Internet. We also obtained data on mail and Internet donations from the households in our sample over the time period of the fund-raising campaign. The results are reported in columns (7) and (8) of Table I. There was not a single donation to ECU, and only one donation to La Rabida—a striking difference compared to the 3-7% of households donating in person for the same charities. The near absence of donations provides further evidence on the motivations of giving. If giving was due to pure altruism, individuals who see the flyer but cannot be at home during the fund-raiser would donate via mail or Internet. The cost of this form of giving (captured by 9 in our model) attenuates giving, but not likely to be zero. A model of warm glow can better fit the data if we assume that the warm glow arises only from in-person donations (i.e., 9 is close to 0). The lack of mail or Internet donations is also consistent with social pressure: giving arises only under high social pressure. Survey. For the survey treatments, we estimate a specification parallel to equation (2), separately for the 2008 and the 2009 field experiments. In the 2008 experiments (Figure VII (top) and column (1) of Table III), a flyer announcing a $0 10-minute survey reduces the share opening the door by 15% (though not significantly), compared to the same survey without flyer. In addition, flyers for more attractive surveys with either shorter duration (5 minutes) or higher payment ($10) lead to a 10-15% increase in the share of households opening the door, indicating that households sort into shorter and better-paid surveys, though the difference is again not significant. The share completing the survey is comparable (about 10%) for the $0 10-minute conditions with and without flyer (Figure VII (top) and column (2) of Table III). Interestingly, the willingness to complete an unpaid 10-minute survey is higher than the willingness to give money even to an in-state charity. Also, compared to the $010-minute survey with flyer, surveys with shorter duration or payment have a higher completion rate of 17-18%, a 70-80% increase. The increase is very similar for the two groups, indicating a high value of time for survey completion, consonant with the sample population characteristics. Figure VII (bottom) and columns (3) and (4) of Table III report the results for the 2009 survey. Within the treatments with flyer, the share answering the door is increasing in the amount paid (from $0 to $5 to $10). In addition, the share answering is TESTING FOR ALTRUISM AND SOCIAL PRESSURE 31 a the the iring ttie leting ttie out No Flyer No Flyer Flyer Flyer Flyer Flyer Opt-out Opt-out ($0/Emin) ($5/5min) (SO/lOmin) ($0/Emin) ($E/Gmin) ($10/5 min) ($075min) ($5/5min) (N=1420) (N=910) (N=769) (N = 1703) (N = 1856) (N=673) (N = 1374) (N = 1330) Figure VII Surveys: (Top) 2008 Experiment, (Bottom) 2009 Experiment Figures present the percent of households answering the door and the percent completing the survey (out of all households) in the survey treatment run in 2008 (top) and 2009 (bottom). The estimates are obtained from regressions that control for randomization fixed effects. significantly lower for the treatments with opt-out, especially the treatment with no payment. These findings are consistent with the findings for 2008 and confirm a sizable responsiveness of the presence at home to the attractiveness of the task. CO to TABLE III Results for Survey Treatments Specification: OLS: regressions Sample: 2008 Survey 2009 Survey Indicator for Indicator for Indicator for Indicator for Dependent variable: answering the door completing survey answering the door completing survey (1) (2) (3) (4) Flyer ($0/10min) treatment -0.0563 0.0000 (0.0377) (0.0260) Flyer ($0/5min) treatment -0.0021 0.0794 (0.0334) (0.0233)*** Flyer ($10/10min) treatment -0.0105 0.0695 (0.0448) (0.0265)** Baseline ($5/5min) treatment -0.0096 -0.0240 (0.0217) (0.0176) Flyer ($0/10min) treatment -0.0198 -0.0008 (0.0247) (0.0180) Flyer ($0/5min) treatment -0.0193 0.0243 (0.0189) (0.0132)* Flyer ($5/5min) treatment 0.0028 0.0549 (0.0205) (0.0164)*** Flyer ($10/5min) treatment 0.0368 0.1072 (0.0243) (0.0187)*** I 0 s 1 TABLE III (CONTINUED) ZO I C! S i I b o 8 ťa ťa Specification: OLS regressions Sample: Dependent variable: 2008 Survey 2009 Survey Indicator for answering the door (1) Indicator for completing survey (2) Indicator for answering the door (3) Indicator for completing survey (4) Opt-out ($0/5min) treatment Opt-out ($5/5min) treatment Omitted treatment Mean of dep. var. for omitted treatment Fixed effects for solicitor, date-location, hour, and area rating N -0.0583 (0.0227)* -0.0373 (0.0192)* -0.0026 (0.0160) 0.0259 (0.0162) Baseline ($0/10min) 0.4138 0.1025 Baseline ($0/5min) 0.3915 0.1498 X N = 1865 X N = 1865 X N = 10035 X N = 10035 Notes. Estimates for a linear probability model with standard errors, clustered by solicitor-date, in parentheses. The omitted treatment is the baseline no-flyer $0-10 minutes survey for columns (1) and (2) and the baseline no-flyer $0-5 minutes survey for columns (3) and (4). The regressions include fixed effects for the solicitor, for the date-town combination, for the hour of day, and for a subjective rating of home values in the block. * significant at 10%; * * significant at 5%; * * * significant at 1% CO CO 34 QUARTERLY JOURNAL OF ECONOMICS The 2009 treatments also indicate a strong response of survey completion with respect to duration and payment. The survey completion rate in treatments with flyer increases monotonically from 14.8% for a $0 10-minute survey to 25.5% for a $10 5-minute survey. The latter completion rate is remarkably high: over 50% of the people opening the door took the survey. There is a similar increase in survey completion with respect to payment in the opt-out treatments, though not in the treatments without flyer. V. Structural Estimates The reduced-form estimates provide qualitative evidence on the importance of both altruism and social pressure, but they do not allow for a quantitative estimate of the underlying social preferences. We now estimate the model parameters structurally, combining the results of the fund-raising and the survey experiments. Set-up. We estimate the model of Section II, imposing seven additional assumptions, several of which are relaxed below. First, the private utility of consumption is linear, u(W —g) = W—g. This assumption is not unduly restrictive since utility should be locally linear in a standard expected utility framework. Second, the altruism parameter a comes from a normal distribution with mean fi and variance a2 censored below at 0, with the remaining probability mass assigned to a = 0. Third, the altruism function is vigi) = log (r +gi); the parameter r governs the concavity of the altruism function for a > 0: a large r implies that the marginal utility of giving, given by (r"g), declines only slowly in the individual giving gi, consistent with pure altruism—the individual cares about the overall donation and her individual giving is only a small part. A small r instead indicates that the marginal utility diminishes steeply with the individual giving, more consistent with warm glow. Fourth, the social pressure cost S is homogeneous. Fifth, the level of giving gs from which on there is no social pressure cost is $10, the median donation. Sixth, the cost of leaving home c(h) is symmetric around ho and quadratic: c(h) = (fe~^o) . Seventh, to capture the lack of giving by mail, we set 6 = 0. To model behavior in the survey treatments, we assume a baseline utility s of completing a 10-minute survey for no monetary payment, with s ~ Fs, a normal distribution with parameters jjf and as. Hence, we allow s to be negative for households TESTING FOR ALTRUISM AND SOCIAL PRESSURE 35 that dislike doing surveys without compensation. In addition, individuals receive utility from a payment m for completing the survey, and receive disutility from the time cost t of the survey, both of which are deterministic. Given the assumption above of (locally) linear utility, we add these terms to obtain the overall utility from completing a survey: s + m — t. The time cost t equals rvs, where r is the duration of the survey in fraction of hours, and vs is the value of one hour of time (see below). We denote by Ss the social pressure cost of saying no to a survey request. The agent undertakes the survey if s + m — t is larger than —Ss. The threshold sm,t = —Ss — (m — t) is the lowest level of s such that individuals will agree to complete the survey if asked. An increase in the social pressure Ss or in the pay m, or a decrease in the cost of time t will lower the threshold and hence increase the probability of survey completion. The decision problem of staying at home conditional on receiving a notice is Taking into account corner solutions for h*, this leads to a solution for the probability of being at home: h* = max[min[/io + ?]max(s + m — t, —Ss), 1],0]. The vector of parameters £ that we estimate are: (i) Tig008 and Tig009—the probabilities of opening the door in the 2008 and 2009 no-fiyer treatments; (ii) r—the probability of observing (and remembering) the flyer; (iii) r\—the responsiveness of the probability of opening the door to the desirability of being at home; (iv) \xs and as—the mean and standard deviation of the distribution Fs of the utility of completing a 10-minute survey; (v) vs—the value of one hour of time spent completing a survey; (vi) Ss—the social pressure associated with saying no to the survey request; (vii) ffj1 and aah (where ch = LaR,Ecu)—the mean and standard deviation of the censored normal distribution F from which the altruism parameter a is drawn; (viii) r—the curvature of the altruism function, which is assumed to be the same for the two charities; (ix) Sch (ch = LaR,Ecu)—the social pressure cost associated with a donation request; the tables display the social pressure cost associated with giving zero, Sgs = 10S. To estimate the model, we use a minimum-distance estimator. Denote by m (£) the vector of moments predicted by the theory as a function of the parameters £, and by m the vector of observed max h max (s + m — t, —Ss ÄG[0,1] (h - h0)2 2r) 36 QUARTERLY JOURNAL OF ECONOMICS moments. The minimum-distance estimator chooses the parameters £ that minimize the distance (m (£) — m)' W (m (£) — m), where W is a weighting matrix. As a weighting matrix, we use the diagonal of the inverse of the variance-covariance matrix. Hence, the estimator minimizes the sum of squared distances, weighted by the inverse variance of each moment.16 As a robustness check, we also use the identity matrix as weight. To calculate the theoretical moments, we use a numerical integration algorithm based on adaptive Simpson quadrature, implemented in Matlab as the quad routine. As moments m (£) we use the following probabilities (where j = F,NF,00 and ch = LaR,Ecu): (i) the probability of opening the door in the various charity treatments (P(H)jh); (ii) the probability of checking the opt-out box in the opt-out treatment (P(00)Cq0); (iii) the unconditional probability of giving in the various charity treatments (P (G)jh); (iv) the probability of giving an amount of money in different ranges CP (0 < G < 10)f, P(G=gs = 10)f, P (10 < G < 20)j\ P (20 < G < 50)j\ and P (G > 50)f); (v) the probability of opening the door in the various survey treatments k (with varying dollar amounts, minutes, and flyer conditional), P(H)sk, run in 2008 and in 2009; (vi) the unconditional probability of completing the survey in the various survey treatments, P (SV)sk, run in 2008 and in 2009; and (vii) the probability of checking the opt-out box in the survey opt-out treatments (P (00)sk). The corresponding empirical moments m are estimated in a first-stage model using the same controls as in the main regressions, and are listed in the Appendix. To calculate the method of simulated moments estimate, we employ a common sequential quadratic programming algorithm (Powell 1983) implemented in Matlab as the fmincon routine. We impose the following constraints: Sch,Ss > 0 (social pressure non-negative), (jch,(js > 0 (positive standard deviation of altruism), /io008,/io°09>r e [0>1] (probabilities between 0 and 1), and rj e [0,9999] (finite elasticity of home presence). We begin each run of the optimization routine by randomly choosing a starting point, drawn from a uniform distribution over the permitted parameter space. The algorithm determines successive search directions by solving a quadratic programming subproblem based on an 16. Given the large number of moments, weighting the estimates by the inverse of the full variance-covariance matrix is problematic computationally. TESTING FOR ALTRUISM AND SOCIAL PRESSURE 37 approximation of the Lagrangian of the optimization problem. To avoid selecting local minima, we choose the run with the minimum squared distance of 500 runs.17 Under standard conditions, the minimum-distance estimator using weighting matrix W achieves asymptotic normality, with estimated variance (G WG)—(G W^WG^G WG'>—? where G = N'1 Y4L1 ^emi(0 and A = Var[m(£)] (Wooldridge 2002). We calculate V^m(^) numerically in Matlab using an adaptive finite difference algorithm. Identification. Though the parameters are estimated jointly, it is possible to address the main sources of identification of individual parameters. The baseline probabilities of answering the door, Tig008 and Tig009, are identified by the observed probabilities of opening the door in treatments without flyer. The probability of observing and remembering the flyer, r, is identified by two moments in the opt-out treatment: the fraction of households checking the opt-out box (10—12%), which equals rhoF(ao), and the fraction opening the door. The elasticity of opening the door r\ with respect to incentives is identified by the fraction opening the door in the survey treatments for different payments and survey durations. In addition, rj is identified by the amounts given in the different charity treatments. The survey parameters are identified using the survey moments. The survey completion rates for varying amounts of compensation identify the heterogeneity in the willingness to complete the survey, and hence gs, where social pressure plays no role. Without social pressure, an individual with altruism a will give exactly g dollars if the marginal utility of giving, av' (g) = jj^, equals the private marginal utility of consumption, 1, and hence a = r + g. Thus, in this example without social pressure, the mass of households with altruism higher than r + g, that is, 1 — Fig + r), has to equal the observed share of households that give at least g. This pins down the empirical distribution of a for a given r. Figure VIIIA illustrates the identification mechanism for the estimated value of P = 10.61 and giving levels g of $0 and of $10. The identification of r depends on two sets of moments: the sorting in of givers of larger amounts, and the giving of smaller amount. The more concave the altruism function is (that is, smaller D, the more altruistic individuals sort in because of higher infra-marginal utility of giving, and the more frequent are small donations. Finally, the social pressure Sch is identified from two main sources of variation: home presence in the flyer treatment (which, to a first approximation, equals ho — rjrS) and the distribution of small giving (the higher the social pressure, the more likely is small giving and in particular bunching at gs). Estimates. Column (1) in Table IV reports the benchmark estimates of the parameters along with standard errors. The probability of being at home ho is precisely estimated to be 41.4% in 2008 and in 2009. The share r of households that have read (and remember) the flyer is precisely estimated at 34.1%. Although this estimate may appear low, many households may have just disregarded the flyer, or a household member may have seen it, but not informed the person opening the door. The elasticity of home presence rj is estimated to be 0.040 (s.e. 0.011), implying that the cost of increasing the probability of being at home and answering the door by 10 percentage points is = $0.13. The average utility for survey completion is estimated to equal —$26.89, implying that, on average, households dislike completing 10-minute surveys for no pay. There is, however, significant heterogeneity ( Ö CO o > 4—» TD M CC 20 30 Altruism (a) 50 Figure VIII (A) Distribution of Altruism and Cut-offs for Giving; (B) Overall Utility of Fund-Raiser as Function of Altrusim Figure VIIIA plots the estimated distribution of the altruism parameter a > 0. The figure displays the threshold for giving $0 and for giving $10 in the absence of social pressure, given by a v' (T + g.) -1 y- 0 or a y- r + g. Figure VIIIB plots the implied utility in equilibrium of a standard door-to-door fund-raiser, as a function of the altruism parameter a. The parameter values are from the benchmark minimum distance estimates (Table IV). 40 QUARTERLY JOURNAL OF ECONOMICS TABLE IV Minimum-Distance Estimates: Benchmark Results Common parameters Benchmark estimates (1) No social pressure (2) Prob. of home presence 0.414 0.383 (h) year 2008 (0.004) (0.003) Prob. of home presence 0.414 0.392 (h) year 2009 (0.007) (0.008) Prob. of observing flyer (r) 0.341 0.426 (0.012) (0.017) Elasticity of home 0.040 0.008 presence (eta) (0.011) (0.003) Implied cost of altering 0.126 0.656 prob. home by 10 pp. Survey parameters Mean utility (in $) of doing -26.863 -17.203 10-minute survey (4.204) (3.466) Std. dev. of utility of 29.591 28.347 doing survey (5.129) (5.374) Value of time of 80.656 83.039 one-hour survey (22.762) (24.898) Social pressure cost if saying 6.197 0.000 no to survey (1.492) (-) Charity parameters La Rabida ECU La Rabida ECU Share with zero altruism a 0.753 0.763 0.723 0.747 (0.048) (0.071) (0.01) (0.024) Mean altruism a, 12.786 9.659 14.167 10.272 conditional ono>0 (1.444) (1.485) (0.452) (0.876) Std. dev. of altruism a, 10.545 7.994 11.569 8.455 conditional ono>0 (1.038) (1.103) (0.389) (0.773) Curvature of altruism 10.606 10.606 function (4.466) (-) Social pressure cost of 3.751 1.438 0 0 giving 0 in person (0.581) (0.784) (-) (-) SSE 86.618 366.620 Notes. Estimates from minimum-distance estimator with moments listed in the Appendix and weights given by inverse of diagonal of variance-covariance matrix. The estimation assumes a normal distribution censored below at 0, with the censored mass assigned to zero altruism (a=0). Notice that the share with zero altruism is not a separate parameter, but it is the share implied by the censoring of a at a=0. In column (2) we estimate the model fixing the social pressure parameters to 0; in addition, since the curvature of the altruism function is very poorly identifed, we fix it at the same value as in column (1). Standard errors are in parentheses. SSE reports the weighted sum of squared errors. share of respondents like doing surveys even for no pay. The value of time for one hour of survey completion is estimated to be $80.66, which corresponds approximately to the value of time TESTING FOR ALTRUISM AND SOCIAL PRESSURE 41 for the households in the wealthy neighborhoods we reached. Finally, the social pressure cost of turning down a survey request, Ss, is estimated to be $6.20, a sizable magnitude. Turning to the charity parameters, we estimate that about 75% of households have no altruism for either charity and that, conditional on a positive altruism, the mean altruism is higher for the La Rabida charity (12.79) than for the ECU charity (9.66). Figure VIIIA plots the implied distribution of the altruism parameter a for a > 0 for La Rabida and ECU. The estimated P is small, implying a highly concave altruism function, consistent with warm-glow rather than with pure altruism. The social pressure parameters are quite precisely estimated, and the data reject the null hypothesis of no social pressure cost. Turning down a door-to-door giving request is associated with a social pressure cost of $3.75 (s.e. $0.58) for La Rabida and $1.44 (s.e. $0.78) for ECU. That is, it is psychologically more costly to say no to a well-liked local nonprofit than to a lesser-known out-of-state charity. The sizable social pressure cost suggests that the welfare implications of door-to-door campaigns can be large, as we explore below. While the model with social pressure does a nice job of fitting the observed moments (Appendix), the same cannot be said of a model with no social pressure. In column (2) of Table IV we estimate the parameters forcing S = 0 (and setting r, which is essentially unidentified, to the value in column (1)). The fit is extremely poor, with the mean squared error 4.5 times larger. Missing social pressure, the model cannot explain opting out nor the decrease in the share answering the door in the flyer treatment. In Table V we explore the robustness of the estimates to alternative model specifications with respect to the distribution of altruism, the cost function, and the heterogeneity of social pressure. We report the weighted sum of squared errors to allow for comparisons among the models. Allowing for negative values of altruism (spite) toward the charity with an untruncated normal distribution (column (2)) yields nearly identical estimates to the benchmark ones (column (l)).19 Given that social pressure is estimated to be substantially lower than $10, any level 18. At an average income of about $100,000 per year, the implied hourly wage is $50. 0- 19. For a < 0, we assume via) = TABLE V Minimum Distance Estimates: Robustness Normal distr. of Exponential distr. Lognormal distr. Benchmark Benchmark Benchmark altruism, no altruism, allows of altruism, asymmetric altruism, altr., allows for estimate mass at 0 of mass at 0 allows mass at 0 sorting cost no social pref. Common parameters (1) (2) (3) (4) (5) (6) Prob. observing flyer (r) 0.341 0.342 0.340 0.345 0.340 0.412 (0.012) (0.012) (0.012) (0.012) (0.012) (0.079) Elasticity of home 0.040 0.040 0.034 0.047 0.340 down 0.055 up 0.033 presence (eta) (0.011) (0.011) (0.008) (0.013) (0.011) (0.051) (0.009) Probability of type with 0.142 no social preferences (p) (0.13) Survey parameters Mean utility (in $) of -26.863 -26.924 -27.771 -26.163 -26.947 -23.049 doing 10-minute survey (4.204) (4.227) (4.332) (4.072) (4.152) (5.072) Std. dev. of utility of doing 29.591 29.662 30.166 29.361 28.947 27.103 10-minute survey (5.129) (5.155) (5.296) (5.039) (4.927) (5.169) Social pressure cost 6.197 6.205 6.990 5.344 6.550 7.836 of saying no to survey (1.492) (1.495) (1.416) (1.333) (1.725) (2.123) TABLE V (CONTINUED) Benchmark Normal distr. of altruism, no Exponential distr. altruism, allows Lognormal distr. of altruism, Benchmark altruism, asymmetric Benchmark altr., allows for no estimates mass at 0 of mass at 0 allows mass at 0 sorting cost social pref. (1) (2) (3) (4) (5) (6) La La La La La La Charity parameters Rabida ECU Rabida ECU Rabida ECU Rabida ECU Rabida ECU Rabida ECU Share with zero altruism 0.753 0.763 — — 0.807 0.846 0.812 0.880 0.745 0.747 0.743 0.744 (0.048) (0.071) (0.026) (0.035) (0.01) (0.012) (0.06) (0.104) (0.052) (0.08) Mean altruism a, 12.786 9.659 12.931 9.789 9.838 7.103 26.482 26.601 13.056 9.868 13.118 10.065 conditional on a>0 (1.444) (1.485) (1.503) (1.551) (0.541) (0.929) (1.314) (1.11) (1.935) (1.777) (1.689) (1.718) Std. dev. of altruism a, 10.545 7.994 10.647 8.080 9.838 7.103 10.710 6.913 10.741 8.122 10.631 8.163 conditional on a>0 (1.038) (1.103) (1.076) (1.141) 0.541 0.929 (0.698) (1.183) (1.394) (1.251) (1.139) (1.177) Curvature of altruism 10.606 11.066 4 .033 20.976 11.612 11.667 function (4.466) (4.675) (2 .116) (0.869) (6.043) (4.956) Social pressure cost 3.751 1.438 3.704 1.413 4.996 2.358 2.420 0.748 3.688 1.376 3.667 1.326 of giving 0 in person (0.581) (0.784) (0.596) (0.772) (0.453) (1.032) (0.19) (0.391) (0.678) (0.803) (0.597) (0.762) SSE 86. .618 86. .623 90.715 86. .653 86.520 86 .233 Welfare and decomposition of giving Average welfare per house. -1.102 -0.442 -1.093 -0.435 -1.330 -0.708 -0.700 -0.213 -1.101 -0.426 -0.874 -0.323 of fund-raiser (in $) (0.145) (0.301) (0.15) (0.297) (0.128) (0.411) (0.082) (0.155) (0.164) (0.3) (0.255) (0.262) Share of givers who seek 0.518 0.528 0.514 0.526 0.543 0.528 0.524 0.628 0.507 0.520 0.507 0.529 the fund-raiser (0.041) (0.095) (0.041) (0.095) (0.056) (0.097) (0.053) (0.105) (0.05) (0.098) (0.04) (0.096) to I I I b to o 8 Notes. Estimates from minimum-distance estimator with moments listed in the Appendix and weights given by inverse of diagonal of variance-covariance matrix. Benchmark estimates in column (1) are from Table IV. Estimates in column (2) assume a (nontruncated) normal distribution for altruism a with no mass at a=0. Estimates in column (3) and (4) assume, respectively, an exponential distribution and a log-normal distribution for altruism a as well as a mass at a=0. Estimates in column (5) allow for a different elasticity eta of the cost function for increases and decreases in home presence. Estimates in column (6) allow for a share p of the population that have no altruism, nor social pressure. These agents do not give to charity and do not complete the survey. In column (6), the share with zero altruism includes the share with no social preferences. Standard errors are in parentheses. SSE reports the Weighted Sum of Squared Errors. CO 44 QUARTERLY JOURNAL OF ECONOMICS of altruism that is sufficiently small, or negative, is associated with identical behavior—turning down the charity and avoidance. Hence, the shape of the left tail of the altruism distribution is not identified and, importantly, does not affect the estimates of the other parameters. Next, we reestimate the model with alternative assumptions for the distributions of altruism: in column (3) we assume a negative exponential distribution of a (with mean j) with additional probability mass z at 0; the results are very similar to the ones in the benchmark model. In column (4) we assume a log-normal distribution (with parameters jjl and a) and an added probability mass z at 0. The results are qualitatively similar, but with some quantitative differences, such as lower social pressure estimates.20 In column (5) we remove the assumption of a symmetric cost function and allow for the cost of avoidance to differ from the cost of sorting in: c{h) = (h-h0y (h-h0)2 for h > ho for h < ho. The estimates of r)+ and rj- are close, even though r)+ is imprecisely estimated given that only the most favorable surveys ($10 payment for 5 minutes) increase on average the presence at home. As a result, the estimates of the other parameters hardly change. In column (6) we remove the assumption of homogeneous social pressure and allow for a simple form of heterogeneity. We assume that a proportion p of households has no social preferences, that is, does not exhibit altruism (a = 0) and does not respond to social pressure (S = 0), and hence neither gives to charity nor completes surveys.21 The remaining portion 1 — p is described by the benchmark model. The data provide some support for this type of heterogeneity, estimating a share p of 0.142 (s.e. 0.13). In turn, this raises somewhat the estimated share of households that observed the flyer r up to 0.41 and lowers the elasticity of the response f). Despite these changes, the 20. The convergence properties for the lognormal distribution are not as good, presumably because of one additional parameter to be estimated in the distribution of altruism (/j,, a, and z), relative to the censored normal (/j, and >S with a > a (S) > a (S) for a given S and a = a (0). The threshold a is always positive, while a(S) and a(S) can be negative for high S. lemma 1A (Giving in Person) For any a,there is a unique optimal donation g* (a,S), conditional on being at home, which is weakly increasing in a and takes the form: (i) g* (a, S) = 0 for a < g(S); (ii) 0 < g* (a,S) < gs for g(S) < a < a (S); (iii) g* (a, S) =gs for a (S) < a < a; (iv) g* (a, S) > gs for a > a. Define am = u'(W)/9v+'(0; G_t), with am > g (S) for all S. lemma IB (Giving via Mail) For any a and provided 0 < 0 < 1, there is a unique optimal donation via mail g*m (a) (conditional on not being at home), which is weakly increasing in a and takes the form: (i) gm(a) = 0 for a < am\ (W gm (a) > 0 for a > am; (iii) for a < a(S), g*(a;S) = g*m (a) = 0 and for a>a(S), g*(a; S) > 9g*m (a). If 9 = 0, g*m (a) = 0 for all a. 52 QUARTERLY JOURNAL OF ECONOMICS Lemma 2 characterizes the solution for h* as a function of the parameters a and S. LEMMA 2 (Presence at Home) For any a, there is a unique optimal probability of being at home h*(a,S) that is weakly increasing in a. For S = 0, h* (a, 0) = ho for a < a (0) and h* (a, 0) > h0 for a > a (0). For S > 0, there is a unique an (S) e (a (S), a) such that h* (a, S) < ho for a < an (S), h* (an (S), S) = ho, and h* (a,S) > h0 for a > a0 (S). The next lemma refers to an (S) from Lemma 2. We break ties by assuming that if the agent is indifferent between h=ho and h = 0, she does not opt out, that is, h = ho. LEMMA 3 (Opting Out). If S = 0, the agent never opts out. If S > 0, the agent opts out for sufficiently low altruism, a < an (S). Probabilities of Presence at Home and of Giving. Lemmas 2 and 3 implies that the probabilities of presence at home are P(H)NF = h0, /oo h*(a,S)dF, -oo /•OO P(H)00 = (1 - r)ho + r / h*(a, S)dF, J a oo where aoo = —oo for S = 0 and aoo = «o (S) for 5 > 0. Lemma lAa implies that the probabilities of giving in person are P(G)NF = [l-F(g(S) )]h0 />OG P(G)F = (l-r)[l-F(a(S))]ho+r / h*(a,S)dF Jg(S) />OG P(G)00 = a-r)[l-F(g(S))]ho+r / h*(a,S)dF. Ja0(S) Appendix Empirical Moments and Estimated Moments Specification: Minimum-distance estimates Charity La Rabida charity ECU charity Empirical Estimated Empirical Estimated moments moments moments moments Panel A: Fund-raising (1) (2) (3) (4) moments P(Home) no flyer 0.4130 0.4137 0.4171 0.4137 P(Home) flyer 0.3733 0.3757 0.3806 0.3989 P(Home) opt-out 0.3070 0.2917 0.3281 0.2835 P(opt out) opt-out 0.1202 0.1295 0.0988 0.1336 P(Giving) no flyer 0.0717 0.0665 0.0455 0.0423 P(Giving) flyer 0.0699 0.0711 0.0461 0.0451 P(Giving) opt-out 0.0515 0.0629 0.0272 0.0389 P(050), no flyer 0.0016 0.0001 -0.0003 0.0000 P(Giving>50), flyer 0.0002 0.0001 -0.0002 0.0000 P(Giving>50), opt-out -0.0006 0.0001 -0.0005 0.0000 n n = 4962 n = 4962 n = 2706 n = 2706 Appendix (continued) P(Home) P(Do survey) P(Opt-out) Empirical Estimated Empirical Estimated Empirical Estimated moments moments moments moments moments moments Panel B: Survey moments (1) (2) (3) (4) (5) (6) Moments 2008 No flyer $0, 10min 0.4138 0.4137 0.1025 0.1003 — — Flyer $0, 10min 0.3576 0.3725 0.1025 0.1230 — — Flyer $0, 5min 0.4118 0.3891 0.1819 0.1646 — — Flyer $10, 10min 0.4033 0.3984 0.1720 0.1873 — — N N = 1865 N = 1865 N = 1865 N = 1865 — — Moments 2009 No flyer $0, 5min 0.4009 0.4135 0.1486 0.1318 — — No flyer $5, 5min 0.3913 0.4135 0.1246 0.1576 — — Flyer $0, 10min 0.3812 0.3723 0.1478 0.1229 — — Flyer $0, 5min 0.3816 0.3889 0.1729 0.1646 — — Flyer $5, 5min 0.4038 0.4033 0.2035 0.1996 — — Flyer $10, 5min 0.4377 0.4193 0.2558 0.2373 — — Opt-out $0, 5min 0.3427 0.3432 0.1460 0.1576 0.1336 0.1061 Opt-out $5, 5min 0.3637 0.3605 0.1745 0.1920 0.0846 0.0982 N N = 10035 N = 10035 N = 10035 N = 10035 N = 2704 N = 2704 Notes. The table presents the empirical moments and the estimated moments from a minimum-distance estimator. The empirical moments are obtained as regression estimates after controlling for the randomization fixed effects and as such can occasionally be negative. The minimum-distance estimates are in Table IV, column (1). TESTING FOR ALTRUISM AND SOCIAL PRESSURE 55 University of California, Berkeley, and National Bureau of Economic Research University of Chicago and National Bureau of Economic Research University of California, Berkeley, and National Bureau of Economic Research Supplementary Material An Online Appendix for this article can be found at QJE online (qj e. oxfordj ournals. org). References Akerlof, George A., and Rachel E. Kranton, "Economics and Identity," Quarterly Journal of Economics, 115 (2000), 715-753. Andreoni, James. "Giving with Impure Altruism: Applications to Charity and Ricardian Equivalence." Journal of Political Economy, 97(1989), 1447-1458. --, "Impure Altruism and Donations to Public Goods: A Theory of Warm-Glow Giving." The Economic Journal, 100(1990), 464-477. --, "Philanthropy." In Handbook of Giving, Altruism, and Reciprocity, Serge- Christophe Kolm and Jean Mercier Ythier, eds., 1201-1265. (Amsterdam: Elsevier/North-Holland, 2006). Ariely, Dan, Anat Bracha, and Stephan Meier, "Doing Good or Doing Well? Image Motivation and Monetary Incentives in Behaving Prosocially." American Economic Review, 99(2009), 544-555. Asch, Solomon E., "Effects of Group Pressure Upon the Modification and Distortion of Judgment." in Groups, leadership, and men. H. Guetzkow, ed. (Pittsburgh: Carnegie Press, 1951). Becker, Gary S., "A Theory of Social Interactions" Journal of Political Economy, 82(1974), 1063-1093. Benabou, Roland, and Jean Tirole, "Incentives and Prosocial Behavior," American Economic Review, 96(2006), 1652-1678. Bodner, R. and Drazen Prelec, "The Diagnostic Value of Actions in a Self-Signaling Model," in The Psychology of Economic Decisions, Vol.1, I. Brocas and J.D. Carrillo, eds., (Oxford: Oxford University Press, 2003). Card, David, Stefano DellaVigna, and Ulrike Malmendier, "The Role of Theory in Field Experiments". Journal of Economic Perspectives, 25(2011), 39-62. Charness, Gary, and Matthew Rabin, "Understanding Social Preferences with Simple Tests," Quarterly Journal of Economics 117(2002), 817-869. Conlin, Michael, Ted O'Donoghue, and Timothy J. Vogelsang, "Projection Bias in Catalog Orders," American Economic Review, 97(2007), 1217-1249. Croson, Rachel, and Jen Shang, "A Field Experiment in Charitable Contribution: The Impact of Social Information on the Voluntary Provision of Public Goods," Economic Journal, 119(2009), 1422-1439. Dana, Jason D., Daylian M. Cain, and Robyn M. Dawes, "What you don't know won't hurt me: Costly (but quiet) exit in dictator games," Organizational Behavior and Human Decision Processes, 100(2006): 193-201. Fehr, Ernst, and Simon Gächter, "Fairness and Retaliation—The Economics of Reciprocity,"Journal of Economic Perspectives, 14(2000), 159-181. Fong, Christina, and Erzo Luttmer, "What Determines Giving to Hurricane Katrina Victims? Experimental Evidence on Racial Group Loyalty," American Economic Journal: Applied Economics, 1(2009), 64-87. Garicano, Luis, Ignacio Palacios-Huerta, and Canice Prendergast, "Favoritism Under Social Pressure," The Review of Economics and Statistics, 87(2005): 208-216. 56 QUARTERLY JOURNAL OF ECONOMICS Gerber, Alan S., Donald P. Green, and Christopher W. Larimer, "Social Pressure and Voter Turnout: The Results of a Large Scale Field Experiment." American Political Science Review, 102(2008), 33-48. Grossman, Zack, "Self-Signaling versus Social Signaling in Giving,"UCSB Working paper, 2010. Harbaugh, William T., "What do donations buy?: A model of philanthropy based on prestige and warm glow," Journal of Public Economics, 67(1998), 269-284. Laibson, David, Andrea Repetto, and Jeremy Tobacman, "Estimating Discount Functions with Consumption Choices over the Lifecycle," Working paper, 2007. Landry, Craig, Andreas Lange, John A. List, Michael K. Price, and Nicholas Rupp, "Towards an understanding of the economics of charity: Evidence from a field experiment," Quarterly Journal of Economics, 121(2006), 747-782. Lazear, Edward, Ulrike Malmendier, and Roberto Weber,"Sorting in Experiments with Application to Social Preferences," American Economic Journal: Applied Economics, Forthcoming. Levine, David K., "Modeling Altruism and Spitefulness in Experiment," Review of Economic Dynamics, 1(1998), 593-622. List, John A., and David Lucking-Reiley, "The Effects of Seed Money and Refunds on Charitable Giving: Experimental Evidence from a University Capital Campaign," Journal of Political Economy, 110(2002), 215-233. Mas, Alexander, and Enrico Moretti, "Peers At Work," American Economic Review, 99(2009), 112-145. Milgram, Stanley., "Behavioral study of obedience," Journal of Abnormal and Social Psychology, 67(1963), 371-378. Powell, M. J. D., "Variable Metric Methods for Constrained Optimization," in Mathematical Programming: The State of the Art, A. Bachem, M. Grotschel and B. Korte, eds., 288-311 (New York: Springer Verlag, 1983). Wooldridge, Jeffrey M., Econometric Analysis of Cross Section and Panel Data (Cambridge, MA: MIT Press, 2002). ©2012 President & Fellows of Harvard University