Does Africa Need a Rotten Kin Theorem?

Transcription

1 Pubic Discosure Authorized Pubic Discosure Authorized Pubic Discosure Authorized Pubic Discosure Authorized Poicy Research Working Paper 6085 Impact Evauation Series No. 58 Does Africa Need a Rotten Kin Theorem? Experimenta Evidence from Viage Economies Pamea Jakiea Owen Ozier The Word Bank Deveopment Research Group Human Deveopment and Pubic Services Team June 2012 WPS6085

2 Poicy Research Working Paper 6085 Abstract This paper measures the economic impact of socia pressure to share income with kin and neighbors in rura Kenyan viages. The authors conduct a ab experiment in which they randomy vary the observabiity of investment returns. The goa is to test whether subjects reduce their income in order to keep it hidden. The anaysis finds that women adopt an investment strategy that conceas the size of their initia endowment in the experiment, athough that strategy reduces their expected earnings. This effect is argest among women with reatives attending the experiment. Parameter estimates suggest that women behave as though they expect to be pressured to share four percent of their observabe income with others, and substantiay more when cose kin can observe income directy. Athough this paper provides experimenta evidence from a singe African country, observationa studies suggest that simiar pressure from kin may be prevaent in many rura areas throughout Sub-Saharan Africa. This paper is a product of the Human Deveopment and Pubic Services Team, Deveopment Research Group. It is part of a arger effort by the Word Bank to provide open access to its research and make a contribution to deveopment poicy discussions around the word. Poicy Research Working Papers are aso posted on the Web at The authors may be contacted at pjakiea@arec.umd.edu and oozier@wordbank.org. The Impact Evauation Series has been estabished in recognition of the importance of impact evauation studies for Word Bank operations and for deveopment in genera. The series serves as a vehice for the dissemination of findings of those studies. Papers in this series are part of the Bank s Poicy Research Working Paper Series. The papers carry the names of the authors and shoud be cited accordingy. The findings, interpretations, and concusions expressed in this paper are entirey those of the authors. They do not necessariy represent the views of the Internationa Bank for Reconstruction and Deveopment/Word Bank and its affiiated organizations, or those of the Executive Directors of the Word Bank or the governments they represent. Produced by the Research Support Team

3 Does Africa Need a Rotten Kin Theorem? Experimenta Evidence from Viage Economies Pamea Jakiea and Owen Ozier May 28, 2012 JEL codes: C91, C93, D81, O12 Jakiea: University of Maryand, pjakiea@arec.umd.edu; Ozier: Deveopment Research Group, The Word Bank, oozier@wordbank.org. We are gratefu to Feipe Dizon for exceent research assistance, to the staff at IPA-Kenya for their assistance and support in the fied, and to the Weidenbaum Center and the Center for Research in Economics and Strategy at Washington University in St. Louis for funding. We aso thank Key Bishop, Esther Dufo, Garance Genicot, Jess Godberg, Shachar Kariv, Cynthia Kinnan, Ashey Langer, Karen Levy, Maggie McConne, Justin McCrary, Sendhi Muainathan, Avin Murphy, Caro Nekesa, Sivia Prina, Gi Shapira, Erik Sørensen, Sergio Urzúa, Dean Yang, Andrew Zeitin, and numerous conference and seminar participants for hepfu comments. A errors are our own. The findings, interpretations and concusions expressed in this paper are entirey those of the authors, and do not necessariy represent the views of the Word Bank, its Executive Directors, or the governments of the countries they represent.

4 Whoever has a more mobie occupation, and ess respect for tradition, tries to cover his tracks. In Dodoma, I once ran into a street vendor hawking oranges who used to bring these fruits to my house in Dar es Saaam. I was happy to see him, and asked him what he was doing here, five hundred kiometers from the capita. He had had to fee from his cousins, he expained. He had shared his meager profits with them for a ong time, but finay had had enough, and ran. I wi have a few cents for a whie, he said happiy. Unti they find me again! Kapuscinski (2002) 1 Introduction Risk is a pervasive aspect of the ives of individuas in many deveoping economies, and informa risk-pooing arrangements which hep househods cope with shocks can have substantia wefare impacts when credit and insurance markets are incompete. A substantia body of empirica evidence documents the existence of mutua insurance arrangements throughout the word, demonstrating that informa mechanisms typicay fai to competey insure househods against idiosyncratic shocks (cf. Townsend 1994, Coate and Ravaion 1993, Fafchamps and Lund 2003). Much of the iterature focuses on mutua insurance arrangements which are efficient given constraints, characterizing the conditions under which sef-interested househods wi enter risk-pooing schemes vountariy ex ante and the participation constraints which keep househods from defecting ex post. 1 However, the expectation of future transfers is ony one of many reasons househods offer assistance to those worse off: atruism, guit, and socia pressure to share income may aso pay a roe (Scott 1976, Foster and Rosenzweig 2001, Ager and Weibu 2010). In fact, severa recent studies suggest that individuas iving in poor, rura communities often fee obigated to make transfers to reatives and neighbors, and that successfu famiies who do not make sufficient transfers to others can face harsh socia sanctions (Patteau 2000, Hoff and Sen 2006, Di Faco and Bute 2011, Comoa and Fafchamps 2011). For exampe, Barr and Stein (2008) argue that Zimbabwean viagers punish househods who are becoming better off than their neighbors by re- 1 See Ligon, Thomas, and Worra (2002), Abarran and Attanasio (2003), and Kinnan (2011) for exampes. Foster and Rosenzweig (2001), which expores the impact of atruism on the set of sef-enforcing insurance arrangements, is an important exception. 2

5 fusing to attend the funeras of members of those famiies. When socia pressures to assist kin, and the sanctions against those who vioate sharing norms, are strong enough, they can reduce incentives to make profitabe investments and drive savings into ower-return technoogies which are ess observabe to famiy members. Baand, Guirkinger, and Mai (2011) provide evidence of this type of behavior in Cameroon, where members of credit cooperatives take out oans to signa that they are iquidity constrained even when they aso hod substantia savings in order to avoid sharing accumuated weath with reatives. In this paper, we report the resuts of an experiment designed to measure socia pressure to share income with reatives and neighbors in rura viages in Sub-Saharan Africa. We use a controed aboratory environment to expore behaviors which are difficut or impossibe to document using survey data: the wiingness to forgo profitabe investment opportunities to keep income secret. We conduct economic experiments in 26 rura, agricutura communities in western Kenya. Within the experiment, subjects receive an endowment which they divided between a risk-free savings account and a risky but profitabe investment. The size of the endowment varies across subjects, and the distribution of endowment sizes is common knowedge. Whie the amount saved is aways private information, we randomy vary whether the amount invested in the risky security can be observed by other subjects, creating an incentive for those receiving the arge endowment to invest no more than the amount of the sma endowment, thereby keeping their endowment size hidden. We aso offer a subset of subjects the option of paying to keep their investment returns secret, aowing us to directy measure the wiingness-to-pay to hide income. In ight of evidence that women and men have different risk preferences (cf. Gneezy 2009) and that women in poor communities may have more troube accumuating savings (cf. Dupas and Robinson forthcoming), we stratify our experiment by gender, and report resuts for men and women separatey. 2 We find convincing evidence that women are wiing to reduce their expected income to avoid making investment returns observabe. Women receiving the arge endowment, who may wish to hide this fact from their reatives and neighbors, are 25.4 percent (9.6 percentage 2 Dupas and Robinson (forthcoming) find evidence that femae daiy wage earners in western Kenya are more savings constrained than men in simiar occupations. In a simiar vein, De Me, McKenzie, and Woodruff (2008), De Me, McKenzie, and Woodruff (2009), and Fafchamps, McKenzie, Quinn, and Woodruff (2011) find ower returns to capita for microenterprises operated by women than for those operated by men. 3

6 points) more ikey to invest an amount no arger than the sma endowment when returns are observabe; this is equivaent to a 5.4 percentage point decine in investment eve. We find no simiar tendency to hide income among men. The effect we observe among women appears to be driven primariy by the behavior of those with reatives attending the experiment, who woud be abe to observe their payoffs directy. Estimates suggest that these women invest 22.1 percent ess when investment income is observabe than when it is hidden; they are 38.9 percentage points more ikey to invest no more than the amount of the sma endowment, suggesting that their strategy is designed to keep the size of their endowment hidden. Consistent with a simpe mode of decisions in the experiment when subjects face pressure to share income and risk preferences are heterogeneous, we find that women receiving the arge endowment are more ikey to invest exacty the amount of the sma endowment when investment returns are observabe, and that this tendency is aso more pronounced among women with reatives present. Impacts are unikey to be driven by in-aws providing information to husbands: there is no direct impact of having one s husband present, and choice patterns are simiar in the sub-sampe of unmarried women. Among subjects given the opportunity to pay a randomy-assigned price to keep income hidden, 30 percent of those abe to afford the cost of hiding income choose to do so. These subjects pay an average of 15 percent of their gross payout from the experiment. Though women are no more ikey than men to pay to avoid the pubic announcement, their wiingness to do so is strongy reated to the amount of observabe income they receive; this is not true for men. At the viage eve, women s tendency to hide income within the experiment is negativey associated with durabe asset accumuation by househods, skied and forma sector empoyment, and the probabiity of using fertiizer on crops, suggesting that socia pressure to share may hinder growth and deveopment. After presenting our reduced form resuts, we estimate the magnitude of the kin tax parameter via maximum simuated ikeihood in a mixed ogit framework. In our setting, this extension is important because the size of the treatment effect of observabiity depends on individua risk preferences. Decisions in our private information treatments, in which investment returns are unobservabe, suggest substantia heterogeneity in risk aversion across subjects. 3 After controing 3 See Hey and Orme (1994); Choi, Fisman, Gae, and Kariv (2007); Andersen, Harrison, Lau, and Rutström 4

7 for unobserved heterogeneity in risk preferences, we find a statisticay significant tax for women, averaging four percent in genera. For women whose kin attend the experimenta session, consistent with the reduced form pattern, the tax appears to be twice as arge. The main aim of this paper is to document the importance of socia pressure in interhousehod transfer reationships within kin networks in poor communities. Though the tite is meant to be tongue-in-cheek, it is intended to highight the fact that, in many viage economies, the ine separating the househod from the rest of the extended famiy can be quite burry (cf. Randa, Coast, and Leone 2011), and something akin to (potentiay inefficient) househod bargaining may occur between househods within the same kin network. This paper is a first step toward exporing the dynamics of socia pressure within these compex reationships empiricay. We make severa contributions to the existing iterature. First, we introduce a nove ab experiment designed to measure socia pressure to share income in fied settings. The experiment is simpe to understand, but provides subjects with a rich menu of investment options and mutipe mechanisms for hiding their income. Our design aows us to effectivey rue out severa aternative expanations: behavior is not consistent with modes of attention aversion or a desire to avoid socia sanctions against risk-taking, and the ack of a direct effect of having one s spouse present suggests that kin networks are not just passing information on to husbands. Second, treatment assignments within the experiment were randomized within viages, aowing us to expore the association between community outcomes and income hiding in the ab. Third, we ink decisions in the experiment to a mode of individua investment choices when risk preferences are heterogeneous, and estimate this mode via maximum simuated ikeihood. This aows us to recover an estimate of the kin tax parameter. The rest of this paper is organized as foows: Section 2 describes our experimenta design and procedures; Section 3 presents a simpe theoretica framework for interpreting our resuts; Section 4 presents our main reduced-form empirica resuts; Section 5 presents our structura framework and estimates; and Section 6 concudes. (2008); Choi, Kariv, Müer, and Siverman (2011), and Von Gaudecker, van Soest, and Wengström (2011) for further experimenta evidence on risk preference heterogeneity. 5

8 2 Experimenta Design and Procedures 2.1 Structure of the Experiment The experiment was designed to introduce exogenous variation in the observabiity of investment returns. Within the experiment, each participant was given an initia endowment, either 80 or 180 Kenyan shiings. 4 Each subject divided her endowment between a zero-risk, zero-interest savings account and an investment which was risky but profitabe in expectation. The subject received five times the amount that she chose to invest in the risky prospect with probabiity one haf, and ost the amount invested otherwise. A coin was fipped to determine whether each risky investment was successfu. Thus, the main decision subjects faced was how much of their endowment to invest in the risky security and how much to aocate to the secure, zero-profit aternative. Within the experiment, payers were randomy assigned to one of six treatments. First, payers were aocated either the smaer endowment of 80 shiings or the arger endowment of 180 shiings. Endowment sizes were aways private information experimenters never identified which subjects received the arge endowment. However, the distribution of endowments was common knowedge, so a subjects were aware that haf the participants received an extra 100 shiings. Every payer was aso assigned to either the private treatment or one of two pubic information treatments, the pubic treatment or the price treatment. Participants assigned to the private treatment were abe to keep their investment income secret: the decisions they made in the experiment were never discosed to other participants. In contrast, those assigned to the pubic treatment were required to make an announcement reveaing how much they had invested in the risky security, and whether their investment was successfu, to a of the other participants at the end of the experiment. 5 The amounts that subjects invested in the zero-interest savings technoogy were never reveaed, so those who received the arger amount coud choose whether to invest 4 Experimenta sessions were conducted between August 3 and October 1, Over that period, the vaue of the US doar reative to the Kenyan shiing fuctuated between 74.5 and shiings to the doar. We report doar amounts using the exchange rate 75.9 shiings to the doar, which is the average over that period. The two endowments were equivaent to 1.05 and 2.35 U.S. doars, respectivey. Among our subjects, the median monthy wage for individuas in fu-time, unskied empoyment is 2000 shiings (26.35 U.S. doars), and the median monthy wage for subjects in fu-time work in the skied sector is 4750 shiings (62.58 U.S. doars). 5 Subjects were informed up front that they were aowed to deegate the task of making the pubic announcement to a member of the research team if they wished to avoid the pubic speaking aspect of the announcement process. 6

9 80 shiings or ess so as to obscure their endowment size. Finay, those assigned to the price treatment were obiged to make the pubic announcement reveaing their investment returns uness they preferred to pay a price, p, to avoid making the announcement. Prices ranged from 10 to 60 shiings, and were randomy assigned to subjects in the price treatment. Subjects were informed what price they faced before making their investment decisions, but decided whether to pay the price after investment returns were reaized. Hence, subjects in the price treatment were not aways abe to afford to buy out: those who invested and ost a arge fraction of their endowment did not aways have enough experimenta income eft over to pay the exit price, p. 6 Random assignment to treatment generated exogenous variation in the observabiity of investment returns and created costy opportunities to hide income. Assignment to the pubic information treatments meant that outcomes were verifiabe, and might therefore faciitate risk-pooing and, consequenty, risk-taking. On the other hand, if subjects face socia pressure to share income with neighbors and kin, they might be wiing to pay for obscurity when returns are visibe. 7 In particuar, the experiment creates two mechanisms through which subjects coud incur a cost to hide income from others. First, those receiving the arger endowment coud keep their endowment size secret by investing no more than 80 shiings. Second, subjects in the price treatment coud pay the randomy assigned price, p, to concea their income entirey. 2.2 Experimenta Procedures Experiments were conducted in 26 rura, predominanty agricutura communities in western Kenya. 8 One day prior to each experimenta session, the survey team conducted a door-to-door recruitment campaign, visiting as many househods within the viage as possibe. 9 A househods within each viage were invited to send members to participate in the experimenta economic game session the 6 Subjects were never aowed to use money from outside the experiment to pay to avoid the pubic announcement. 7 Our design is reated to those of Barr and Genicot (2008), Attanasio, Barr, Cardenas, Genicot, and Meghir (2012), and Ligon and Schechter (forthcoming), who aso use experiments which capture the desire to avoid socia sanctions outside the ab. 8 Communities were seected to be at east five kiometers apart from one another, to prevent overap in subject popuations, and to avoid areas where IPA Kenya had ongoing projects. 9 When the recruitment team was unabe to contact and survey a househod prior to the experiment, viage eders were asked to invite them to attend the subsequent experimenta session. 7

10 foowing day percent of househods contacted prior to the sessions chose to participate. 10 Experimenta sessions were conducted in empty cassrooms at oca primary schoos. Sessions incuded an average of 83 subjects; no session incuded fewer than 65 or more than 100 subjects. Each session asted approximatey three hours. Within each session, participants were stratified by gender and education eve (an indicator for having done any post-primary education). There were six experimenta treatments, corresponding to the three information conditions (private, pubic, price) interacted with the two endowment sizes. Within each stratum, payers were randomy assigned to each of the six treatments with equa probabiity. 11 Payers assigned to the price treatments were subsequenty assigned a random price from the set of mutipes of ten between 10 and Experimenta sessions were structured as foows. After a brief introduction, enumerators read the instructions and answered participant questions, iustrating the decisions that a subject might face with a series of wa posters. 13 Subjects were then caed outside one at a time, by ID number, to make their investment decisions. Since some participants had imited iteracy skis, decisions were recorded by members of the research team. To ensure that earnings not announced pubicy remained private information, each enumerator sat at a desk in an otherwise empty section of the schooyard. Enumerators began by asking a series of questions designed to make sure that subjects understood the experiment. Subjects were then informed whether they had received the arge or sma endowment and whether they were assigned to the private, pubic, or price treatment. Those who were assigned to the price treatment were aso tod what price they woud need to pay if they wished to avoid the pubic announcement. Subjects then made their investment decisions: each subject was handed a number of 10 shiing coins equivaent to her endowment; the participant 10 In Tabe of Appendix III (onine), we examine the correates of choosing to send a househod member to the experimenta session. Participants were broady representative of the overa popuation, though as expected arger househods were more ikey to send a member to the experiment. Women from poorer househods and those with reativey higher math skis were more ikey to attend. Men were more ikey to attend if they attended church the previous week. Subjects in more isoated communities as measured by distance from a paved road were more ikey to participate, though this cannot bias our resuts since assignment to treatment occurs within viages. 11 The first experimenta session did not incude the two price treatments, so payers were assigned to each of the other four treatments with equa probabiity. 12 Note that each session incuded subjects assigned to a three information conditions, so it was impossibe to infer which individuas chose to pay to avoid the pubic announcement, since these subjects coud not be distinguished from those randomy assigned to the private treatments. 13 Detaied instructions are incuded in Appendix II (onine). 8

11 divided these coins between a savings cup and a business cup. 14 After recording a subject s investment decision, the enumerator woud give the subject a one shiing coin to fip. 15 The outcome of the coin fip determined whether the money paced in the business cup was mutipied by five or removed from the subject s fina payout. Subjects assigned to the price treatment were then asked whether they wanted to pay the fee to avoid announcing their investment resuts. If they had enough money eft to pay the fee, and they chose to do so, it was deducted from their payoff. After a decisions had been recorded, pubic announcements were made. At the end of the session, subjects were caed outside one at a time to receive their payouts. Each subject received her payout in private, and was aowed to eave immediatey after receiving her money. Figure 1 summarizes the progression of activities within the experiment. 2.3 Experimenta Subjects Sessions were conducted in Kenya s Western Province, in three adjoining districts: Bunyaa, Samia, and Butua. 16 A three districts are predominanty smahoder farming communities, though Samia and Bunyaa aso have ports on Lake Victoria. Summary statistics on experimenta subjects are presented in Tabe percent of subjects were femae. Respondents ranged in age from eighteen to percent of subjects had no forma schooing, whie 12.2 percent had finished secondary schoo. The median participant was a 34 year od married woman with seven years of education, iving in a six-person househod with her husband and her four chidren. The median participant s househod owns one bicyce, one ce phone, four chickens, and two mosquito nets, but does not own a teevision, any catte or goats, or a motor vehice percent of respondents ive in househods with at east one empoyed househod member; most (64.6 percent) empoyed subjects do agricutura work or other unskied abor. The median monthy wage among participants with fu-time empoyment was 2, 050 Kenyan shiings, or 1.35 USD per day (assuming twenty work 14 The swahii words for savings ang business are, respectivey, akiba and biashara. The savings cup was marked with a arge etter A and the business cup was marked with a arge etter B. The two cups were made of pastic and were identica in size and coor. 15 To imit the possibiity of infuencing the outcome of the coin fip, each subject paced the coin into a seaed, opaque container which she shook vigorousy before opening it to revea the outcome of the coin toss. 16 Kenya s recent redistricting carved these three former administrative divisions of Busia District off as new districts of their own. One of the districts, Butua, was decared a new district during the course of this project. 9

12 days per month) percent of subjects operate their own business enterprise; among these businesses, ony 11.6 percent have any empoyees percent of participants have bank savings accounts, 17 and 52.8 percent are members of rotating savings and credit associations (ROSCAs). 18 Most experimenta subjects in our sampe ive amongst their kin, and are embedded in interhousehod transfer networks. The median participant has seven reatives iving outside her househod but in the same viage: two cose reatives (parents, parents-in-aw, adut chidren, grandparents, sibings, aunts, and unces) and another five more distant reatives percent of subjects had received a transfer in the ast three months, whie 89.8 percent report making a transfer to another househod over the same period. 19 The median househod making a transfer had given 363 shiings, whie the median househod receiving a transfer had been given 600 shiings. In the three months prior to being surveyed, 42.4 percent of subjects househods had been asked for a gift or oan, and 89.7 percent of househods had contributed money to a harambee, a oca fundraising drive. 20 Tabe 1 reports tests of baance across our six experimenta treatments. 21 The randomization was successfu, generating minima differences in observabes across treatments. Of 68 tests (34 for each gender) reported, we observe ony five significant differences in observabes across treatments. 17 Dupas and Robinson (forthcoming) found that ess than 3 percent of the daiy wage earners samped in Bumaa, Kenya, had savings accounts. Whie Bumaa is just a few kiometers from the region where the present study took pace, their data were coected over two years before our househod survey. The daiy wage earners (primariy market vendors and bicyce taxi drivers) incuded in their study may aso be somewhat worse off than our subjects. 18 Gugerty (2007) surveys ROSCA participants in Busia and Teso Districts in western Kenya; she argues that the socia component of ROSCA participation heps individuas overcome savings constraints. Anderson and Baand (2002) show income-earning women iving in Nairobi sums use ROSCAs to protect their savings from their husbands. Dupas and Robinson (forthcoming) show that femae daiy income earners make more productive investments when given access to even a costy savings account. 19 There are severa reasons for the asymmetry between the probabiity of making a transfer and the probabiity or receiving a transfer. First, househods comprising ony chidren and/or the i or handicapped coud not participate (because ony those aduts who coud be physicay present for the session and were abe to hear the ora instructions coud take part), but are ikey to be net recipients of transfers. Another reason is that most househods in any viage wi make a sma transfer toward the funera expenses of their neighbors; since funeras are reativey rare events, we woud expect to see an asymmetry between the ikeihood of giving and receiving in the short-term. Finay, transfer data are quite noisy, and househods are more ikey to report transfers made than transfers received (Comoa and Fafchamps 2011). 20 A harambee is a sef-hep effort in which community members contribute money or resources to assist a particuar person in need. They may be for sending a chid to schoo, paying for a wedding, or any number of other purposes. The concept existed within a number of different triba groups in Kenya, but was made into a nationa raying cry by Kenya s first president, Jomo Kenyatta (Ngau 1987). 21 We show that the randomization of the price of avoiding making an announcement was successfu, though the randomization was not stratified, in Tabe 2 of Appendix III (onine). 10

13 Though the number of distant famiy members iving in one s viage differs significanty across treatments in the sampe of men, this variation is driven by outiers: the maximum number (within a treatment) of distant reatives reported to ive in one s viage ranges from 72 to 154. A quantie regression of the median number of distant famiy members on the set of treatment dummies does not find significant differences across experimenta treatments (resuts not shown). Among women attending the experiment, there is a significant difference in the number of teevisions owned across treatments, and a marginay significant difference in the number of cows, but no difference in the tota vaue of durabe househod assets. 3 Theoretica Framework In this section, we outine a simpe theoretica mode of individua decisions within the experiment. Subjects faced a finite set of investment options nine in the sma endowment treatments and nineteen in the arge endowment treatments. However, we present decisions in a continuous, nonstochastic framework in order to derive simpe comparative statics that yied testabe predictions in reduced-form regressions. In Section 5, we present a discrete version of the mode that aigns precisey with the choices subjects faced, and that aows us to recover the structura parameters of interest through maximum ikeihood estimation. 3.1 Individua Decisions in Private Information Treatments In a treatments, individua i divides her budget of m i {m s, m } between the business cup (a risky security) and the savings cup (a risk-free but zero interest savings technoogy). When income is private, i seeks to maximize E [u i (b i, s i m i )] = 1 2 u i (s i + c i0 ) u i (s i + 5b i + c i0 ) = 1 2 u i (m i b i + c i0 ) u i (m i + 4b i + c i0 ), (1) where b i is the amount invested in the risky security, s i = m i b i is the amount saved, and c i0 is background consumption or permanent income. Foowing Ashraf (2009) and Godberg (2010), we 11

14 assume income is observabe whenever an individua is known to have received it with probabiity one; income is unobservabe whenever a person can pausiby deny having received it. In the private information treatments, a subject can caim to have invested her income and ost it, imiting the potentia for socia pressure to share payouts. Thus, individua i s optima interior soution in the private treatment, b pri i, soves u i ( ) ( ) m i b pri i + c i0 = 4u i m i + 4b pri i + c i0. (2) We assume that individua preferences can be represented by a utiity function of the CRRA form with parameter ρ i 0 and that background consumption, c i0, is equa to zero. The scae invariance property of the CRRA utiity function is consistent with aggregate data from the private information treatments: subjects aocate an average of 51.8 percent of their budgets to the business cup when they receive the smaer endowment, and an average of 51.9 percent of their budgets to the business cup when they receive the arger endowment. 22 Given these two assumptions, the amount invested in the business cup by individua i in the private information treatment is given by b pri i = ( ) 4 1/ρ i 1 4 1/ρ m i. (3) i + 4 Thus, when individua risk preferences can be represented by a utiity function of the CRRA form, the proportion of the budget invested depends on ρ i, but not on the size of the budget. When ρ i is cose to zero, an individua wi invest amost a of her budget in the risky prospect; the proportion of m i invested decreases with ρ i. 22 A Mann-Whitney test fais to reject the nu hypothesis (p-vaue 0.865) that the fraction of the budget invested does not depend on the budget size. There is, however, suggestive evidence that women invest a sighty arger fraction of their budgets when they receive the arger endowment (53.5 versus 50.7 percent invested) whie men invest sighty ess (49.3 versus 53.6 percent). Since these differences are quite sma in magnitude, we restrict attention to the CRRA case. 12

15 3.2 Individua Decisions in Pubic Treatments Next, we consider how individua i s optimization probem changes when she is obiged to stand up and announce her investment income. After making the pubic announcement, individuas may face socia pressure to share observabe income. Foowing Ashraf (2009) and Godberg (2010), we assume that individua i is obiged to transfer a proportion, τ i, of observabe income to members of her socia network for exampe, her spouse or her reatives and that income is observabe when an individua is known to have it with probabiity one, whether it is announced or not. 23 For exampe, if it is common knowedge that each viager receives a grant of a certain size, then that income is observabe even if it is not distributed pubicy. Within the experiment, the impication is that a subject assigned to the pubic treatment who receives the smaer endowment, m s, cannot hide any of her income, since every subject is known to have received at east m s and she is forced to announce whether her investment succeeded. In contrast, an individua assigned to the private treatment can pausiby caim to have invested and ost a or most of her endowment, and a subject receiving m = m s + d can choose to invest b m s, thereby making d shiings of income unobservabe. Consider the decision probem facing a subject assigned to the pubic treatment receiving the smaer endowment, m s. She chooses b pub i,s m s such that i,s = argmax 1 [(1 τ i ) (m s b i,s )] 1 ρ i b i,s m s ρ i 2 b pub [(1 τ i ) (m s + 4b i,s )] 1 ρ i 1 ρ i (4) Since the 1 τ i term drops out of the first-order condition characterizing the optima interior soution, individuas receiving the sma endowment make the same aocation decisions regardess of whether investment returns are pubic or private. Proposition 1 characterizes individua behavior and wefare for subjects receiving the sma endowment in the pubic treatment. Proposition 1. If individua i receives the sma endowment, m s, then her optima investment in 23 See De Me, McKenzie, and Woodruff (2009) for a cosey reated modeing approach. 13

16 the business cup in the pubic treatment is equa to her optima investment in the private treatment: b pub i,s = bpri i,s = ( ) 4 1/ρ i 1 4 1/ρ m s. i + 4 Expected utiity is ower in the pubic treatment than in the private treatment when τ i > 0. Expected utiity is decreasing in τ i for a τ i [0, 1]. The proof foows directy from the soution to i s expected utiity maximization probem. Since b pub i,s = bpri i,s, whenever τ i > 0, expected utiity must be ower in the pubic treatment because consumption is ower than in the private treatment in both outcome states. Subjects who receive the arger endowment have the option of investing an amount ess than or equa to the smaer endowment, thereby creating pausibe deniabiity and making themseves indistinguishabe from those who received the sma endowment. The discrete change in observabe income means that both utiity and margina utiity are discontinuous at b = m s. Reca that d = m m s. An individua who invests b > m s in the pubic treatment wi choose b i, = argmax 1 [(1 τ i ) (m s b i, + d)] 1 ρ i b i, >m s + 1 [(1 τ i ) (m s + 4b i, + d)] 1 ρ i 2 1 ρ i 2 1 ρ ( ) i 4 1/ρ i 1 = 4 1/ρ (m s + d), i + 4 (5) the same budget share she woud have invested in the private treatment. However, since utiity is ower in the presence of socia pressure to share observabe income, such individuas may prefer to set b i, = m s. Proposition 2 characterizes this pattern. Proposition 2. There exists a threshod risk aversion parameter, ρ, such that if ρ i ρ, then: 1. i s optima investment in the private, arge endowment treatment, b pri i,, exceeds m s; 2. i s optima investment in the pubic, arge endowment treatment takes one of two vaues: b pub i, { } m s, b pri i, ; (6) 3. and, finay, there exists τ (ρ i ) (0, 1) such that b pub i, = m s if and ony if τ i τ (ρ i ). 14

17 Proposition 2 demonstrates that individuas with sufficienty ow eves of risk aversion, who woud invest more than m s in the private, arge endowment treatment, wi either do the same in the pubic treatment or wi invest exacty m s, the highest eve of investment which keeps endowment size hidden. Whether the atter strategy is optima depends on the eve of socia pressure to share one s winnings. In contrast, an individua who invests b m s in the pubic, arge endowment treatment wi choose b i, = argmax 1 [(1 τ i ) (m s b i, ) + d] 1 ρ i b i, <m s + 1 [(1 τ i ) (m s + 4b i, ) + d] 1 ρ i κ i 2 1 ρ i 2 1 ρ ( ) i 4 1/ρ i ( 1 = 4 1/ρ m s + d ). i τ i (7) Since 0 < 1 τ < 1, ( ) ( ) 4 1/ρ i 1 4 1/ρ i ( 1 4 1/ρ (m s + d) < i /ρ m s + d ) ; (8) i τ i the impication is that sufficienty risk averse individuas, who woud invest ess than m s in the private treatment, wi invest more when investment returns are observabe than when they are private. Proposition 3 characterizes this pattern of behavior. Proposition 3. For a ρ i > ρ, there exists a threshod socia pressure parameter, τ (ρ i ) such that: 1. i s optima investment in the pubic, arge endowment treatment, b pub i,m, fas in the interva: b pub i, ( ) b pri i,, m s (9) if and ony if τ i < τ (ρ i ); 2. otherwise, for a ρ i > ρ and τ i τ (ρ i ), b pub i, = m s. Propositions 2 and 3 partition (ρ i, τ i ) space into four cases which are depicted in Figure 2. Whether individuas receiving the arge endowment invest above or beow m s in the private treatment depends on risk aversion. The more risk averse individuas invest beow m s in the private 15

18 treatment (Regions A and B in Figure 2), whie those who are ess risk averse invest above m s in the private treatment (Regions C and D in Figure 2). Regions A and B are separated from Regions C and D at ρ = ρ. The formua for ρ is given in Appendix Equation 40; ρ equas n(4)/ n(5) for the endowment sizes used in this experiment. As described in Proposition 2, ess risk averse types facing significant socia pressure to share income (Region C) invest ess in the pubic, arge endowment treatment than they do in the private, arge endowment treatment: they reduce their investment to exacty m s. The behavior of ess risk averse types who face itte socia pressure (Region D) is not affected by assignment to the pubic treatment. 24 However, more risk averse types facing any nonzero socia pressure (Regions A and B) increase investment when returns are announced pubicy. This is because the first-order condition characterizing optima investment changes when ony a portion of income is taxed, as shown in Inequaity 8 and described in Proposition 3. How much risk averse types increase investment depends on the eves of socia pressure and risk aversion: those facing ower socia pressure (Region A) invest more when investment returns are announced pubicy, but sti ess than m s ; those facing higher socia pressure (Region B) shift up to exacty m s in the pubic treatment. 25 Figure 3 provides additiona intuition, depicting expected utiity in the pubic and private treatments as a function of the eve of investment. 3.3 Individua Decisions in Price Treatments We now consider individua decisions in the price treatments, where subjects can choose to pay a randomy chosen price, p > 0, to avoid making the pubic announcement. Observe that subject i receiving gross payout x i > 0 in the sma endowment treatment prefers to pay p whenever (x i p) 1 ρ i 1 ρ i (1 τ i) 1 ρi x 1 ρi 1 ρ i, (10) 24 The boundary between Regions C and D is given by the reationship between τ and ρ. Aong this boundary, the utiity-maximizing investment above m s, given in Equation 5, yieds the same utiity as investing exacty m s (thereby hiding d). This boundary does not appear anayticay tractabe, but can be cacuated numericay. 25 Aong the boundary between A and B, τ can be expressed anayticay by soving for τ as a function of ρ (for ρ > ρ) for which the interior soution maximizing utiity conditiona on investing at most m s, shown in Equation 7, is to invest exacty m s. The formua for τ is given in Appendix Equation

19 or p τ i x i. Simiary, a subject in the arge endowment treatment prefers to pay p to avoid the pubic announcement whenever (x i p) 1 ρ i 1 ρ i [(1 τ i)x + τ i d 1{b i m s }] 1 ρi 1 ρ i, (11) or p τ i x i τ i d 1{b i m s }. It is apparent that, for any fixed 0 p < x that makes i exacty indifferent between paying p and making the pubic announcement, i stricty prefers to pay p for a p < p and stricty prefers not to for a p > p. Simiary, for any fixed x which makes i precisey indifferent, she stricty prefers to pay p whenever her payout is above x and stricty prefers to make the pubic announcement whenever her gross payout is beow x. Hence, we need ony consider three strategies for subjects in the price treatments: aways paying p to avoid the pubic announcement, never paying p, and paying p when the investment is successfu, but not when it is unsuccessfu. 26 When p is sufficienty high, never paying p is the optima strategy, and individua investment decisions wi be identica to those in the pubic treatments. For exampe, consider a subject in the sma endowment treatment facing a τ of 0.10 who invests haf her endowment in the business cup. If her investment is successfu, her gross payout is 240 shiings, so she woud be indifferent between announcing her investment return and paying a price of 24 shiings to avoid the announcement. For higher vaues of p, her investment decisions wi be the same as in the pubic treatments. When τ i > 0 and p is sufficienty ow, aways paying p to avoid the pubic announcement is optima. In this case, the optima investment eve is given by b aways i = ( ) 4 1/ρ 1 4 1/ρ (m i p). (12) + 4 Hence, investment eves are equivaent to those in the private information treatments, but with a smaer budget of m i p, accounting for the fact that subjects aways pan to pay p to avoid making the announcement. Foowing directy from Equation (12), we can see that for those subjects who pan to aways pay to avoid making the pubic announcement, optima investment 26 In other words, it wi never be optima for a subject to pay p when her investment is unsuccessfu but not when it is successfu. In Section 5, we refer to these strategies as aways, never, and heads, respectivey. 17

20 eves are decreasing in p. For intermediate vaues of p and τ > 0, it is optima to foow a strategy of paying to avoid the announcement ony when the investment succeeds. Predictions in this setting are ess cear-cut, and we defer consideration of this case to Section Extensions to the Mode In this section, we expore the possibiity that individua behavior in the pubic information treatments might be infuenced by factors other than socia pressure to share income. We consider two such aternative expanations. First, we discuss the hypothesis that subjects might be wiing to pay to avoid the pubic announcement because they are averse to attention or pubic scrutiny. We refer to this as attention aversion. We then expore the possibiity that risky behaviors incuding investing in the business cup might be sociay sanctioned Attention Aversion A straightforward mode of participants desire not to draw attention to themseves without regard to any effective taxation is to incorporate an additive cost parameter, κ, into the utiity function: u attention i (x) = u i (x) u i (x) κ if no announcement is made if pubic announcement occurs (13) This parameter, κ, does not affect the first order condition characterizing the optima aocation decision. It wi therefore have no impact on the ikeihood of investing exacty 80 shiings in either the pubic treatment or the price treatment (when the price of giving is high). In ight of this, our reduced-form resuts on investment decisions are evidence of socia pressure, and cannot be driven by the desire to avoid pubic attention. However, within the price treatment, κ > 0 woud infuence the wiingness to pay to avoid making the pubic announcement. We return to this point in Section 5, when we expicity estimate the κ parameter to test whether it is a significant factor driving decisions in the price treatments. 18

21 3.4.2 Socia Sanctions Against Risk-Taking Another aternative hypothesis is that differences in investment choices between the private and pubic information treatments stem from the desire to avoid potentia socia sanctions against risktaking for exampe, if investing in the business cup were perceived as simiar to gambing. We made it cear to participants that there was no way for them to reaize an actua oss during the experiment, and used the business cup framing to push this point more subty. However, we can aso examine the data for patterns that woud suggest that participants were concerned about hiding their risky investments from others. We consider two possibe strategies for modeing socia sanctions against risk-taking. First, if investing any positive amount were seen as gambing and subjects wished to avoid being observed taking such risks, we shoud see a arger fraction of participants choosing to invest zero in the pubic information treatments than in the private treatments. Aternativey, if the size of the sanction against gambing were increasing in the amount invested in the business cup, this woud add a downward soping cost to the expected utiity function, thereby reducing the optima investment amount, b. This woud ead to an overa reduction in investment eves in the pubic information treatments reative to the private treatments, and a particuary strong move away from arge investment amounts if the function were convex; however, this woud not ead to an increase in the proportion of subjects investing exacty the amount of the sma endowment. 4 Resuts In this section, we characterize individua choices within the experiment and estimate the impacts of observabiity on investment decisions and fina payoffs. We report resuts reating to investment decisions in Section 4.1, and describe decisions regarding whether to pay to avoid the pubic announcement in Section 4.2. Summary statistics on outcomes in the experiment are presented in Tabe 2. Averaging across a treatments, subjects earned approximatey 240 Kenyan shiings (3.16 US doars) in the experiment, which is equivaent to eight percent of mean monthy wages among subjects reporting paid 19

22 empoyment. Thus, stakes were arge, but not ife-atering. There is substantia variation in payoffs across across treatments: subjects in the pubic, sma endowment treatment received the owest payoffs, averaging 139 Kenyan shiings (1.83 US doars); those in the private, arge endowment treatment earned the most, on average 355 Kenyan shiings (4.68 US doars). 4.1 Individua Investment Decisions On average, subjects chose to invest just over haf their endowment in the business cup (Tabe 2). The fraction of the endowment invested is simiar in the arge and sma endowment treatments. Among those receiving the arger endowment, the amount invested is sighty ower in the two pubic information treatments (the pubic and price treatments) than in the private treatment. When aotted 180 shiings, subjects coud avoid reveaing that their endowment exceeded that of others by investing 80 shiings or ess; the probabiity of doing so is higher in the pubic information treatments than in the private treatment. We present histograms of investment decisions in Figure 4. Dark bars represent the distribution of investment decisions in the private treatments; ighter bars represent investment decisions in the two pubic information treatments. As predicted by the mode, decisions are simiar in the pubic and private sma endowment treatments. However, women receiving the arge endowment are more ikey to invest exacty 80 shiings in the pubic information treatments than in the private, arge endowment treatment. A Komogorov-Smirnov test rejects the hypothesis that the two distributions are equa (p-vaue 0.068). There is no difference in behavior across treatments among men receiving the arge endowment. We now examine the impact of observabiity on investment decisions in a regression framework, estimating the effects of assignment to one of the pubic information treatments on the amount of money invested in the business cup. We estimate the OLS regression: Investment itv = α + β 1 P ubic i + β 2 Large i + β 3 P ubic Large i + η v + X iζ + ε itv (14) where Investment itv is the amount invested in the business cup by subject i assigned to treatment 20

23 t in viage v, P ubic i is an indicator for assignment to either the pubic or the price treatment, Large i is an indicator for receiving the arge endowment, P ubic Large i is an interaction between P ubic i and Large i, η v is a viage fixed effect, and X i is a vector of individua contros incuding dummies for age and education categories, marita status, househod size, and the og vaue of househod durabe assets. 27 Standard errors are custered at the viage eve in a specifications. Resuts, disaggregated by gender, are reported in Tabe 3. Odd-numbered coumns report regression resuts without contros; even-numbered coumns incude them. The P ubic Large i interaction is negative and significant among women (Coumns 1 and 2), but positive and insignificant among men (Coumns 3 and 4). The coefficient estimates indicate that women receiving the arge budget invest approximatey 6.3 shiings, or 6.5 percent, ess when returns are observabe than when they are hidden. Adding viage fixed effects and additiona contros has amost no impact on estimated coefficients and significance eves. Next, we test whether the observed changes in investment eves are consistent with women attempting to obscure the size of their endowments. We test the hypothesis, predicted by the mode, that subjects are more ikey to invest exacty 80 shiings when investment returns are observabe by estimating probit regressions of the form Pr [Investment = 80] = Φ ( α + βp ubic i + X iζ ) (15) among the sampe of subjects randomy assigned to the arge endowment treatments, as we as a inear probabiity mode (LPM) where the dependent variabe is an indicator for investing exacty 80 shiings (Tabe 4, Pane A). In a specifications, women are significanty more ikey to invest exacty 80 shiings in the pubic information treatments. For exampe, the LPM estimates indicate that women are 6.2 to 7.0 percentage points more ikey to invest 80 shiings when investment returns are observabe. The coefficient on P ubic in the sampe of men is not significant in any specification, in ine with our previous resuts. 27 As discussed in Section 3, decisions in the price treatments wi be identica to decisions in the pubic treatments when p is sufficienty high, so we poo data from both pubic information treatments to maximize power. Coefficient estimates are simiar when data from the price treatments is omitted. 21

24 Moving beyond the precise predictions of the mode presented in Section 3, we aso estimate simiar specifications using indicators for investing either 70 or 80 shiings (Tabe 4, Pane B) and for investing any amount ess than or equa to 80 shiings (Tabe 4, Pane C). Since ony 2.3 percent of women and 6.0 percent of men in the private, sma endowment treatment actuay invest 80 shiings, subjects might fee that investing 70 shiings better obscures the size of their endowment. 28 In more genera terms, any investment eve ess than or equa to 80 shiings obscures the size of the subject s endowment, and may be seen as consistent with a desire to hide a portion of the arge endowment from others. In a specifications, the P ubic variabe is positive and significant in the sampe of women, but insignificant in the sampe of men. Women receiving the arge endowment are between 7.0 and 7.6 percentage points more ikey to invest 70 or 80 shiings and between 9.6 and 10.5 percentage points more ikey to invest 80 shiings or ess than those in the private, arge endowment treatment Treatment Effect Heterogeneity Our hypothesis is that women face socia pressure to share income, and this creates an incentive to hide investment returns when possibe, even if it is costy to do so. Under this hypothesis, the extent of income hiding shoud be associated with factors predicting the eve of socia pressure an individua is ikey to face after the experiment. We focus on cose kin outside the househod as the group most ikey to pressure individuas into sharing income. 29 We test this by interacting the P ubic variabe with indicators for whether or not a subject had any cose kin attending the experiment, thereby disaggregating the impact of the pubic information treatments. 30 We estimate 28 Investing ess than, but not exacty 80 shiings may refect eve-k reasoning, aong the ines first discussed by Nage (1995), Stah and Wison (1994), and Stah and Wison (1995). 29 Hoff and Sen (2006) highight the roe payed by kin networks in extracting surpus from successfu reatives, whie Baand, Guirkinger, and Mai (2011) provide evidence that individuas seek to hide income from famiy members. 30 We define cose kin as parents, grandparents, sibings, grown chidren, and aunts and unces. Note that we were unabe to stratify treatment assignment by kin presence because the variabe was constructed after experimenta sessions took pace using a name-matching agorithm. 22

25 the OLS regression equation Y itv = α + β 1 KinP resent i + β 2 P ubic KinP resent i + β 3 P ubic NoKinP resent i + η v + X iζ + ε itv (16) where Y itv is one of three outcomes of interest: investment eve, an indicator for investing exacty 80 shiings, or an indicator for investing no more than 80 shiings. 31 We restrict the sampe to subjects assigned to the arge endowment treatments, who coud obscure the size of their endowment by investing 80 shiings or ess. We incude viage fixed effects (η v ) and contros for age and education categories, HH size, marita status, and househod durabe asset hodings in a specifications. Resuts are reported in (Tabe 5, Pane A). Women with cose kin attending the experiment invest ess in the pubic information treatments and are more ikey to invest exacty 80 shiings and no more than 80 shiings. The coefficient estimates are extremey arge. Women with reatives present invest 24.5 shiings ess (p-vaue ) when investment returns are observabe than when they are hidden, which is equivaent to more than a twenty percent reduction in investment reative to the private information treatment. These women are aso 14.9 percentage points more ikey to invest exacty 80 shiings (p-vaue ) and 41.8 percentage points more ikey to invest 80 shiings or ess (p-vaue ). In contrast, point estimates are smaer and not statisticay significant for women with no kin attending the experiment. As expected, we aso find no statisticay significant resuts for men. An aternative hypothesis is that kin are significant because they pass information about wives incomes to their husbands. 32 This is of particuar concern because 76.5 percent of women in our sampe are married, and the two main ethnic groups in the area the Luhya and the Luo are both patrioca (Brabin 1984, Luke and Munshi 2006), so married women typicay ive near their in-aws rather than their bood reatives. We present two pieces of evidence which suggest that kin are not simpy conveying information to husbands. 31 We estimate inear probabiity modes for the two binary outcomes. Probit resuts are simiar in magnitude and significance, and are omitted to save space. 32 Robinson (2008) and Ashraf (2009) document imited commitment and observabiity effects within the househod, whie Anderson and Baand (2002) argue that Kenyan women use ROSCAs to hide savings from their husbands. 23

26 First, we estimate Equation 16 repacing the kin variabes with indicators for whether or not a subject s spouse attended the experiment (Tabe 5, Pane B). If kin were ony important because they shared information with husbands, we woud expect the direct effect of a husband s presence to be at east as arge as the impact of having kin at the experiment. However, we find that the interaction between P ubic and the indicator for having a spouse present is not significanty associated with the amount invested or the probabiity of investing either exacty 80 shiings or 80 shiings or ess. Moreover, the P ubic N ospousep resent variabe is significanty associated with the probabiity of investing exacty 80 shiings and 80 shiings or ess, suggesting that the main impact of the pubic treatment is on women whose husbands were not present. We aso examine investment decisions among the sma sampe of women receiving the arge endowment who are not married and have cose kin attending the session. These women invest an average of 95.7 shiings in the private treatment versus 84.0 shiings in the pubic information treatments, and they are 51.4 percentage points more ikey to invest 80 shiings or ess in the pubic treatments. These differences are not statisticay significant in this tweve subject subsampe, but it nonetheess appears that women with kin present are concerned about observabiity even in cases where the kin cannot possiby pass information on to husbands Income Hiding across Viages Next, we expore the association between the eve of income hiding, aggregated within a community, and viage-eve outcomes. If pressure to share income does, in fact, act ike a tax on investment returns, we woud expect to see a correation between the eve of socia pressure within a community and a range of deveopment indicators. We expore this by constructing a viage-eve measure of income hiding motivated by the theoretica framework presented in Section 3. Among those receiving the arge endowment, the mode suggests that greater pressure to share income wi ead to an increase in the ikeihood of investing exacty 80 shiings when returns are observabe. In ight of this, we create a viage-eve measure of the extent of income hiding: the difference between the proportion of subjects investing exacty 80 shiings in the two pubic information, arge endowment treatments and the proportion investing exacty 80 shiings in the private, arge 24

27 endowment treatment. 33 For each viage in our sampe, we create a measure of income hiding among women, and a separate measure of income hiding among men. Across a 26 communities, the average eve of income hiding among women is 7.27 percent; the 95 percent confidence interva for the mean eve of income hiding across viages is [ 0.002, 0.147]. Unsurprisingy, the average eve of income hiding among men is substantiay ower, ony 1.14 percent; the 95 percent confidence interva, [ 0.085, 0.108], suggests that the variation in the measure of income hiding among men may be argey attributabe to noise. In ight of this and our earier resuts, we focus on the associations between income hiding among women and viage-eve outcomes. Figure 5 pots the reationship between our measure of income hiding among women and four viage-eve outcome variabes: the og vaue of durabe househod assets, the fraction of subjects in either forma or skied empoyment, the average eve of wages from paid work, and the mean share of farm househods that used fertiizer in the ast year. 34 A estimated reationships are negative, suggesting that more income hiding is associated with ower eves of househod weath and reguar empoyment. We expore these reationships in a regression framework in Tabe 6. In Coumn 1, we estimate the reationship between og househod durabe assets and income hiding among women. In spite of the sma sampe size, the estimated reationship is marginay significant (p-vaue 0.070). In Coumn 2, we incude viage-eve contros for distance from the viage to the nearest paved road, the mean education eve of subjects from that community, the mean number of cose reatives in the viage, and the mean number of community groups in which subjects participate. After adding these contros, the estimated association between income hiding and househod assets becomes sighty arger and more significant (p-vaue ). In the remainder of the tabe, we repicate these specifications using the other three outcome variabes of interest; we find simiar significant, negative reationship in a specifications. Higher 33 A resuts presented in this section are neary identica if we use the difference between the proportion investing exacty 80 shiings in the pubic, arge endowment treatment (ony) and the proportion investing exacty 80 shiings in the private, arge endowment treatment, omitting subjects who participated in the price treatment. 34 We incude the ast outcome in ight of evidence presented in Dufo, Kremer, and Robinson (2011) and Giné, Godberg, Siverman, and Yang (2012) that farmers in Kenya and Maawi, respectivey, often pan to use fertiizer, but are unabe to save the money to purchase it in the ean season which occurs after panting but before the harvest. 25

28 eves of income hiding within the experiment are associated with ower rates of skied and forma empoyment and ower wages, and are aso associated with substantia reductions in the probabiity of using fertiizer. Such cross-community resuts do not merit a causa interpretation. Nonetheess, these findings are of interest for two reasons. First, resuts are consistent with the view that pressure to share income may act as a drag on savings and investment, and may therefore sow deveopment. An aternative, though not mutuay excusive, expanation is that in the poorest areas, enforced sharing norms remain an essentia too for handing risk, and thus the pressure to share is the most acute. In either case, the evidence suggests that socia pressure to share income is greatest in the worst off areas, and shoud be viewed as an important socia factor in modes of poor, rura viage economies. More generay, these resuts go some way toward addressing the concern that individua choices in ab experiments may not be meaningfuy correated with decisions and outcomes outside the ab (cf. Levitt and List 2007); in this case, they ceary are. 4.2 The Wiingness-to-Pay to Hide Income We now examine the wiingness-to-pay to avoid the pubic announcement using data from the price treatments. Subjects assigned to the price treatments were offered the option of paying a randomy chosen price 10, 20, 30, 40, 50, or 60 shiings to avoid announcing their investment income to the other subjects. 35 Of 690 subjects assigned to the price treatments, 627 subjects coud afford to pay p to avoid the pubic announcement, and 190 (30.3 percent of those abe to pay) chose to do so. Subjects who chose to buy out paid an average price of 29.3 shiings equivaent to an average of 15.3 percent of their gross earnings. Figure 6 pots the proportion of subjects paying to avoid the pubic announcement, broken down by gender, endowment size, and exit price. Subjects are generay more ikey to buy out at ower prices, though this is not as evident among men receiving the arge endowment. Tabe 7 reports OLS regressions of the probabiity of paying to avoid the pubic announcement 35 Though this randomization was not stratified, it was argey successfu. Tabe 2 in Appendix III (onine) reports the resuts of a baance check exercise in which subject characteristics were regressed on dummies for prices 20 through 60. The price variabes are jointy significant at the five percent eve in ony four of the 68 F-tests. 26

29 on exogenous factors: the randomy assigned price and indicators for receiving the arge endowment and having the coin and with heads facing up. Women with more observabe income are ceary more ikey to pay to avoid the pubic announcement: the coefficient on Heads is consistenty positive and significant, suggesting that subjects with successfu investments are approximatey 23 percentage points more ikey to pay to avoid announcing their investment return. Men, in contrast, are not significanty more ikey to pay to avoid the announcement when their investments are successfu. As suggested by the figures, both men and women are ess incined to buy out at higher exit prices: the coefficient on the price variabe is negative and significant in a specifications. Anaysis of factors associated with wiingness-to-pay is compicated by the fact that, whie prices are randomy assigned, gross payouts (savings pus investment returns) are not. As the mode suggests, the reationship between pressure to share income, individua risk preferences, and the unconditiona probabiity of paying to avoid the pubic announcement is compicated; and we observe whether a subject pays the exit price after a particuar reaization of the coin fip, but not whether she panned to never buy out, buy out if her investment was successfu, or aways buy out. In ight of this, we defer consideration of the expicit wiingness to pay to Section 5. Tabe 8 reports the resuts of regressions of amount invested within the experiment on the price of exit, restricting the sampe to those randomy assigned to the price treatments. The mode predicts that for subjects panning to pay to avoid the pubic announcement whether or not their investment was successfu, the amount invested wi be a decreasing function of the price avoiding the pubic announcement. Consistent with this prediction, we find that random assignment to a higher price of avoiding the pubic announcement is associated with significanty ower eves of investment. In contrast to our previous resuts, this appears is true for both women (Coumns 1 and 2) and men (Coumns 3 and 4). 5 Estimating the Socia Pressure Parameter In this section, we estimate the magnitude of the socia pressure parameter, τ, whie controing for unobserved heterogeneity in individua risk aversion. As discussed above, the treatment effects 27

30 reported in Section 4 demonstrate that among women τ is not equa to zero: making investment returns observabe impacts individua choices. However, the magnitude of the reduced form impact does not identify τ because the treatment effect depends on an individua s eve of risk aversion. Fortunatey, random assignment shoud insure that the distribution of individua risk parameters in each of the six treatments is representative of the popuation distribution. We can therefore estimate the parameters of that distribution using data from the private treatments, and estimate τ whie controing for heterogeneity in risk aversion in a structura framework. We mode decisions in a discrete choice setting, where each subject decides how many ten shiing coins to aocate to the business cup, and aocates the remainder to the savings cup. Subjects in the sma endowment treatments have nine options (0, 10, 20,..., 80), whie those assigned to the arge endowment treatments have nineteen (0, 10, 20,..., 180). In what foows, we index experimenta treatments with t, individua subjects with i, and investment options with j and, when necessary, k. We assume that subject i s expected utiity of investing b j is given by: EU ij = EV ij + ε ij (17) where EV ij denotes the expicity-modeed expected utiity of investing b j (what Train (2003) terms representative utiity ) and ε ij is a Gumbe-distributed preference shock. 36 Behavior in the private treatments suggests that individuas have heterogeneous risk preferences. As in Section 3, we assume these can be represented by a CRRA utiity function, v i ( ), parameterized by ρ i, and that individua ρ i parameters are normay distributed with mean µ ρ and variance σρ. 2 The probabiity that subject i chooses to invest b j can then be written in the form of a mixed ogit mode: ( ) e EV ij/σ ε P ij = k=1,...,j t e EV f (ρ) dρ. (18) ik/σ ε where σ 2 ε is proportiona to the variance of ε ij ε ik. 37 Let y ij be an indicator function equa to one 36 Loomes (2005) refers to such error terms as Fechner errors. See Hey and Orme (1994) and Von Gaudecker, van Soest, and Wengström (2011) for exampes of their use in modeing stochastic choices in individua decision-making experiments. 37 When V ij = X β, β and σ ε are not separatey identified; σ ε is identified in our framework because EV ij is a non-inear function of parameters. 28

31 if subject i chooses to invest amount b j. The og-ikeihood function for treatment t can be written as: LL t = i I t j J t y ij n [ ( e EV ij/σ ε k=1,...,j t e EV ik/σ ε ) f (ρ) dρ ]. (19) The og-ikeihoods can then be summed across treatments. Foowing the procedures outined in Train (2003), we simuate the og-ikeihood by taking one thousand random draws from the standard norma distribution, and maximize the simuated og-ikeihood numericay Heterogeneity in Individua Risk Preferences We express the CRRA utiity function as v (x ρ i ) = 1 η i x 1 ρ i, (20) where η i = ρ i 10 1 ρ i, the difference between the utiity of the highest possibe payout in the experiment and the utiity of the owest stricty positive payout. 39 Since VNM expected utiity functions represent preference orderings over otteries and are robust to positive, affine transformations, this utiity function represents the same preferences as the standard CRRA formuation, u (x ρ i ) = x1 ρ i 1 ρ i. (21) However, in a mixed ogit framework, the probabiity of choosing investment option b j depends on the magnitude of the difference between EV ij and the utiities associated with other options, and not just the position of b j in the preference ordering. Hence, the scae of EV ij is directy reated to the ikeihood of choosing an investment option, b k, that is ess-preferred in the sense that EV ik < EV ij. 40 The standard normaization of the CRRA utiity function eads to very different 38 We impement this using the MATLAB command fminunc using the Broyden-Fetcher-Godfarb-Shanno (BFGS) updating procedure. Standard errors are cacuated using the outer product of the gradient. 39 See Goeree, Hot, and Pafrey (2003) for a simiar approach. Using the owest possibe payout, zero, is not feasibe because u(x) as x 0 for subjects with ρ 1. Resuts are amost identica when 10 1 ρ is repaced with 1 1 ρ or ρ. 40 We acknowedge the sight abuse of the term ess-preferred in this context since, by construction, the chosen option is aways the most-preferred once the unobserved preference shock has been taken into account. 29

32 scaings of the utiity function across the range of feasibe ρ i vaues. As a resut, for a fixed vaue of σ ε, it forces individuas with ow vaues of ρ i to make choices that are cose to deterministic, whie individuas with high enough ρ i parameters make choices which approach a uniform distribution. For exampe, consider investment decisions in the private treatments. Using our scaing, the expected utiity of investing b j is given by: EU ij = 1 (m i b j ) 1 ρ i + 1 (m i + 4b j ) 1 ρ i +ε ij, (22) 2η i 2η }{{ i } EV ij whie η i woud be repaced with (1 ρ i ) if we instead used the scaing in Equation 21. When the conventiona scaing is used, as in Equation 21, investing any amount between 0 and 70 shiings in the private, sma endowment treatment eads to EV ij vaues between 26 and 38 for an agent with ρ i = 0.35, but EV ij vaues between 0.37 and 0.20 for an agent with ρ i = 1.5. The range of EV vaues is substantiay smaer for the more risk averse agent. As a consequence, when σ ε = 0.3 the agent with ρ i = 0.35 woud choose the EV -maximizing amount, 70 shiings, more than 85 percent of the time, but the agent with ρ i = 1.5 woud choose the EV -maximizing investment of 20 shiings ess than 14 percent of the time, and woud choose a of the options ess than 70 shiings with probabiities between 0.11 and Our proposed utiity range (UR) scaing of the CRRA utiity function addresses this issue. In the exampe considered above, UR scaing impies that, given σ ε = 0.3, a subject with ρ i = 0.35 woud choose the EV -maximizing amount, 70 shiings, with probabiity 0.125, whie the subject with ρ i = 1.5 woud choose the EV -maximizing investment of 20 shiings with probabiity If the noise parameter, σ ε, were reduced to 0.01, the ess risk averse subject woud chose the EVmaximizing amount with probabiity 0.419, whie the more risk averse subject woud chose the EV-maximizing amount with probabiity Though simpe to impement, UR scaing generates resuts which are simiar to those generated by the contextua utiity mode of Wicox (2008), in which utiity is scaed by the difference in the utiities faced by an individua decision maker within a specific choice probem, and when the expected utiities are repaced with their certainty equivaents as in Von Gaudecker, van Soest, 30

33 and Wengström (2011). We expore the reationship between the form of scaing used and the estimated parameters µ ρ and σ ρ in Tabe 9. We report the parameter estimates using UR scaing in Coumn 1, parameters estimated using the utiity function defined in Equation 21 in Coumn 2, parameters estimated using the certainty equivaent in pace of EV ij in Coumn 3, and parameters estimated using the contextua utiity mode in Coumn 4. We incude data from both arge and sma endowment private treatments; the expected utiity expression in both treatments is given in Equation 22. The contextua utiity mode in Coumn 4 uses different scaings for the arge and sma endowment treatment; and the certainty equivaent mode in Coumn 3 raises EV ij to the 1/(1 ρ i ) power to convert utiity back into monetary terms. UR scaing generates parameter estimates for µ ρ and σ ρ which are neary identica to those produced using either the certainty equivaent or the contextua utiity procedures. The estimated µ ρ is between and in a three modes, whie the estimated σ ρ ranges from to (Tabe 9). 41 Estimated eves of risk aversion are higher than those typicay reported for undergraduate subjects (cf. Hot and Laury 2002, Goeree, Hot, and Pafrey 2003) and sighty higher than typica estimates of risk aversion in non-student popuations (cf. Andersen, Harrison, Lau, and Rutström 2008, Harrison, Humphrey, and Verschoor 2010, Tanaka, Camerer, and Nguyen 2010). 42 As the tabe demonstrates, though the utiity range scaing, certainty equivaent, and contextua utiity modes a ead to comparabe parameter estimates, using the standard CRRA utiity function, in which x 1 ρ i is divided by 1 ρ i, eads to sighty different parameter estimates (Tabe 9, Coumn 2). In Figure 7, we compare actua investment decisions in the experiments to the predictions of the mode given the estimated µ ρ and σ ρ parameters (using UR scaing). The top panes present in-sampe predictions: decisions in the private, sma endowment treatment, which were incuded in the data set used for estimation. The bottom panes present actua and simuated investment decisions in the pubic, sma endowment treatment, which was not incuded in the estimation 41 We aso estimate σ ε, but omit it from the tabe to save space. As expected given the different utiity scaings, the modes generate different estimates of σ ε. 42 Andersen, Harrison, Lau, and Rutström (2008) report mean CRRA parameters of 0.74 in a representative sampe of the Danish popuation. Harrison, Humphrey, and Verschoor (2010) estimate a mean CRRA parameter of in a sampe of subjects drawn from Ethiopia, India, and Uganda. Tanaka, Camerer, and Nguyen (2010) estimate the average CRRA parameter to be between 0.59 and 0.63 among their Vietnamese subjects. 31

34 sampe. In both cases, the mode appears to predict the distribution of actua investment decisions successfuy. 5.2 Individua Decisions when Investment Returns Are Observabe Whie individua i s investment decision in either of the private treatments depends ony on the size of i s endowment and her eve of risk aversion, choices in the pubic treatments aso depend on τ i, and choices in the price treatments depend on both τ i and p. In the sma endowment, pubic treatment, the expected utiity of investing b j is given by: EU ij = 1 [(1 τ i )(m s b j )] 1 ρ i + 1 [(1 τ i )(m s + 4b j )] 1 ρ i +ε ij, (23) 2η i 2η }{{ i } EV ij where, again, η i = ρ i 10 1 ρ i. Given this expression for EV ij, the og-ikeihood function is defined by Equations 18 and 19. Note that, referring to Equation 22, we coud re-write this as EU ij = (1 τ i ) 1 ρ i EV private ij + ε ij. (24) As discussed earier, the EV -maximizing investment choice is identica in the sma endowment pubic and private treatments. However, scaing down the EV vaues by (1 τ i ) 1 ρ shoud make choices in the pubic treatment sighty noisier, and this is exacty what we find: the standard deviation the amount invested is 15.8 in the private, sma endowment treatment, and it is 16.6 in the anaogous pubic treatment. In the pubic, arge endowment treatment, subjects have the option of investing 80 shiings (m s ) or ess, thereby making 100 shiings (m m s ) unobservabe to other participants. We indicate this potentiay hidden quantity as: H ij = (m m s ) 1{b j m s } (25) 32

35 where 1{ } is the indicator function. Thus, the expected utiity of investing b j is given by: EU ij = 1 [(1 τ i )(m b j ) + τ i H ij ] 1 ρ i + 1 [(1 τ i )(m + 4b j ) + τ i H ij ] 1 ρ i +ε ij (26) 2η i 2η }{{ i } EV ij Again, given the EV ij vaues, the og-ikeihood function is defined by Equations 18 and 19. In the price treatments, subjects first decide how much to invest, then earn their investment return, and finay decide whether to pay the price, p, to avoid the pubic announcement (if they have enough money to do so after investment outcomes are reaized). The probabiity of any observed sequence of choices investing b j, and then deciding whether to pay p to avoid the announcement is the product of the probabiity of making the observed investment decision and the conditiona probabiity of making the observed buyout decision. As discussed in Section 3, there are three possibe strategies which may be optima for subjects in the price treatment: never paying p, aways paying p, and paying p ony when the investment is successfu. We refer to these strategies as never, aways, and heads, respectivey. For some vaues of b j and p, aways may not be a viabe strategy for exampe, participant i cannot pan to pay p if she invests her entire endowment and that investment fais. The expected utiity of the never strategy is defined by Equation 23 for the sma endowment treatment and by Equation 26 for the arge endowment treatment. The representative utiities of of the aways and heads strategies are defined as foows: EV aways ij = 1 2η i (m i b j p) 1 ρ i + 1 2η i (m i + 4b j p) 1 ρ i (27) 1 EVij heads 2η i [(1 τ i )(m s b j )] 1 ρ i + 1 2η i (m s + 4b j p) 1 ρ i if m i = m s = (28) 1 2η i [(1 τ i )(m b j ) + τ i H ij ] 1 ρ i + 1 2η i (m + 4b j p) 1 ρ i if m i = m We assume subjects make investment decisions without knowing the reaization of the preference shock which wi infuence their buyout decisions. Hence, subject i in a price treatment chooses 33

36 investment amount b j to maximize: { EU ij = max EV never ij, EV heads ij }, EV aways ij + ε ij. (29) Whenever m i b j < p, EV aways ij is omitted from Equation 29. After the investment return is reaized, if i has a gross payout greater than p, she decides whether to pay p to avoid the pubic announcement. Given a reaized gross payout of x i, the utiity of paying to avoid the announcement is: 1 η i (x i p) 1 ρ i + ζ i0 (30) This incudes ζ i0, participant i s idiosyncratic utiity from avoiding announcement given this payout. Meanwhie, the utiity of making the announcement is: 1 η i ((1 τ i )x i + τ i H ij ) 1 ρ i + ζ i1 (31) where, anaogousy, ζ i1 is idiosyncratic utiity associated with announcing. We assume that ζ i0 and ζ i1 are independent draws from a Gumbe distribution. Let e i equa one if subject i chooses to pay p and zero otherwise. Now define U exit i ( = ( 1) e i 1 (x i p) 1 ρ i 1 ) ((1 τ i )x i + τ i H ij ) 1 ρ i +( 1) e i (ζ i0 ζ i1 ) (32) η i η i }{{} Vi exit so that V exit i is the difference between the utiity of the action that was not chosen (paying p to avoid the pubic announcement or not) and the utiity of taking the chosen action. The probabiity of i s observed sequence of actions is then given by: P ij P exit i = ( e EV ij/σ ε k=1,...,j eev ik/σ ε e V exit i /γ ) f (ρ) dρ, (33) where γ is proportiona to the variance of the difference between ζ i0 and ζ i1, the Gumbe-distributed 34

37 noise parameters specific to the buy out decision. Before proceeding to our estimates of the kin tax parameter, it is worth reviewing the sources of identification in our mode. Because subjects choose from a rich menu of investment options, the distribution of investment decisions in the two private treatments aow us to identify µ ρ, σ ρ, and σ ε (as demonstrated in Tabe 9). The kin tax parameter, τ, is identified by the differences in the distribution of investment decisions between the private, arge endowment treatment and the two pubic information, arge endowment treatments; the reationship between the wiingness to pay to avoid the pubic announcement and observabe income provides a second source of identification. Finay, the variance of the ogit error term in the buy out decisions, γ, is identified by the excess wiingness to pay to avoid the pubic announcement at very ow gross payouts and high exit prices; our mode attributes this to the stochastic component of individua decisions. 5.3 Parameter Estimates Summing the og ikeihoods in a treatments as shown in Equation 19 with EV ij expressions from Equations 22, 23, 26, and 29, and the conditiona probabiity in Equation 33 we can estimate a parameters of the mode using the fu dataset. We do so separatey for men and women, as the experimenta treatment randomization was stratified on gender to permit separate anaysis. Parameter estimates are reported in Tabe 10. In Coumn 1, we report estimates from a simpified ikeihood function which ony uses data from investment decisions. In Coumn 2, we report parameter estimates based on a the decisions within the experiment. For women, the estimated τ ranges from to , and is significanty different from zero at the 99 percent confidence eve. Among men, the point estimate is smaer, ranging from to , and is not statisticay significant. In Figure 8, we compare the predictions of the mode to the actua frequency with which subjects pay to avoid the pubic announcement. We disaggregate the data according to the impied tax rate facing subjects: the randomy-assigned exit price they faced divided by their gross payout. In a deterministic choice mode, subjects wi pay to avoid the pubic announcement whenever τ i > p i /x i, where x i is their gross payout (or τ i > p i /[x i (m m s )] for subjects assigned to the arge endowment treatment who invested 35

38 ess than m s in the business cup). However, as the figure demonstrates, the stochastic component pays a arge roe in individua decisions within our experiment, particuary when the difference between the expected utiity of paying p and the expected utiity of sharing income with others is sma though the estimated τ parameter is ess than 0.05, the mode predicts the observed high rates of paying to avoid the pubic announcement even when the price is more than five percent of the gross payout. For women, the estimated parameters fit the data reasonaby we, though the mode over-predicts the ikeihood of paying p at impied tax rates between 20 and 30 percent. As discussed in Section 3, one aternative expanation for the reativey high wiingness to pay to avoid the pubic announcement is that subjects are attention averse, and wish to avoid being asked to stand up in front of their neighbors, irrespective of whether their income from the experiment wi be reveaed. We expore this by incorporating an additiona parameter, κ, into the mode, which is simpy subtracted from a EV ij expressions any time a pubic announcement is being made. Intuitivey, κ is identified from excess wiingness to pay to avoid the announcement in the price treatments, as compared to the impied wiingness to pay identified from the increased tendency to invest exacty 80 shiings in the pubic treatments, and by the imperfect correation between tota income and observabe income, since subjects seeking to avoid socia pressure woud ony be wiing to pay to hide observabe income from neighbors and reatives. As shown in Coumn 3, κ is precisey estimated and very cose to zero in the sampe of women, and we cannot reject the nu that κ is zero for both women and men. Our abiity to expore potentia heterogeneity in τ is imited by the reativey sma sampes in different demographic categories, particuary in the price treatment, where subjects faced different randomy-assigned prices and decisions depended on both prices and (not randomy-assigned) gross payouts. However, in ight of our reduced form resuts, we take an initia step in this direction by aowing τ to depend on whether a subject s cose reatives were present at the experiment. In Tabe 11, we report parameter estimates when we repace the singe τ parameter with two separate parameters, τ no kin present and τ kin present. Among women, τ no kin present is sighty smaer than in the fu sampe, at 0.043, but remains statisticay significant. The estimated vaue of τ kin present is substantiay higher, and aso statisticay significant. Among men, as in the fu sampe, 36

39 τ no kin present remans about 2.7 percent, whie the estimated τ kin present is sighty beow zero and imprecisey estimated. Thus, consistent with our reduced form resuts, we find evidence that women with kin present at the experiment are more exposed to socia pressure to share income. To summarize, women in the experiment behave as if they expect to be pressed to share four percent of their cash income with others, and substantiay more if their cose kin can observe their income directy. Determining whether the eve of socia pressure differs across different sources of income cash vs. crop yieds, wages vs. windfa income, etc. is beyond the scope of this paper. However, it is worth noting that a 4 8 percent kin tax may have arge disincentive effects if, for exampe, reatives observe wages or micro-enterprise revenues, but are not abe to easiy separate profits from tota income by accounting for abor time and indirect costs, both of which may be unobserved. This woud be consistent with evidence that many micro-entrepreneurs are unabe to cacuate their own profits, and do not correcty deduct time and indirect costs (cf. Karan, Knight, and Udry 2012). At first gance, our estimated τ parameters may not seem unduy arge; however, if transated into a tax on micro-enterprise revenues, for exampe, they may we be arge enough to discourage the creation or expansion of businesses. 6 Concusions We report the resuts of a nove economic experiment designed to measure socia pressure to share income in Kenyan viages. Our approach is stratified to ensure baance, randomized within viages, and is conducted on a arge sampe. The design permits both reduced-form estimates to find resuts in ine with comparative static predictions, and structura estimates to identify parameters of interest in the presence of heterogeneity. Women who know that the outcome of their investments wi be made pubic make investments that are ess profitabe in expectation. Resuts are strongest for those who have reatives present at the experiment. When we offer some participants the opportunity to pay a fee to avoid making an announcement, they do so at substantia cost: 30 percent of those abe to pay to avoid the pubic announcement choose to do so; these subjects sacrifice 15 percent of their gross payout, on average. 37

40 Structura estimates of the average kin tax are significanty different from zero for women, estimated at roughy 4.3 percent for those whose reatives did not attend the experiment, but at 8.0 percent for those with reatives present. Our mode of stochastic choices in the experiment fits the data we, expaining both investment and exit decisions. We see no evidence that this behavior can be expained by househod bargaining with a spouse, aversion to pubic announcements, or aversion to risk-taking in genera. We hypothesize that the behavior observed in this experiment is a sign that viage sharing norms distort investment incentives towards ess visibe, but potentiay ess profitabe, investments, and may consequenty sow economic growth. The negative correations we observe between the extent of income hiding at the viage eve and the eve of prosperity in the viage, measured severa different ways, are in agreement with this interpretation. However, such resuts shoud be interpreted with caution, since the direction of causaity is uncear. Moreover, the efficiency impacts of socia pressure to share income wi ceary depend on the range of income-hiding technoogies avaiabe, and correation between observabiity and profitabiity. There is sti much to be earned about the range of observabe and unobservabe investment opportunities avaiabe to poor househods, the set of technoogies for hiding or protecting income from pressure to share, and the mechanics of that pressure. Whether the eves of socia pressure we observe are arge or sma wi depend on the ways in which socia pressure is exerted outside of the ab: for exampe, whether individuas are pressed to share a cash income (for exampe, business revenues) or ony their profits after compensating themseves for their indirect costs or abor time. Studies of mutua insurance typicay assume that transfer arrangements are on the efficient frontier, though the anaogous assumption has been caed into question in intrahousehod bargaining contexts. Our work suggests that reationships with cose kin outside the househod may be simiar to within-househod interactions, and that socia sanctions which encourage cooperation and sharing may aso have important disincentive effects. 38

41 References Abarran, P., and O. P. Attanasio (2003): Limited Commitment and Crowding Out of Private Transfers: Evidence from a Randomised Experiment, Economic Journa, 113(486), C77 C85. Ager, I., and J. Weibu (2010): Kinship, Incentives and Evoution, American Economic Review, 100(4), Andersen, S., G. W. Harrison, M. I. Lau, and E. E. Rutström (2008): Eiciting Risk and Time Preferences, Econometrica, 76(3), Anderson, S., and J.-M. Baand (2002): The Economics of Roscas and Intrahousehod Resource Aocation, Quartery Journa of Economics, 117(3), Ashraf, N. (2009): Spousa Contro and Intra-Househod Decision Making: An Experimenta Study in the Phiippines, American Economic Review, 99(4), Attanasio, O., A. Barr, J. C. Cardenas, G. Genicot, and C. Meghir (2012): Risk Pooing, Risk Preferences, and Socia Networks, American Economic Journa: Appied Economics, 4(2), Baand, J.-M., C. Guirkinger, and C. Mai (2011): Pretending to Be Poor: Borrowing to Escape Forced Soidarity in Cameroon, Economic Deveopment and Cutura Change, 60(1), Barr, A., and G. Genicot (2008): Risk Pooing, Commitment and Information: An Experimenta Test, Journa of the European Economic Association, 6(6), Barr, A., and M. Stein (2008): Status and egaitarianism in traditiona communities: an anaysis of funera attendance in six Zimbabwean viages, CSAE Working Paper No. 26. Brabin, L. (1984): Poygyny: an indicator of nutritiona stress in African agricutura societies?, Africa, 54(1), Choi, S., R. Fisman, D. Gae, and S. Kariv (2007): Consistency and Heterogeneity of Individua Behavior under Uncertainty, American Economic Review, 97(5), Choi, S., S. Kariv, W. Müer, and D. Siverman (2011): Who Is (More) Rationa?, working paper. Coate, S., and M. Ravaion (1993): Reciprocity without commitment : Characterization and performance of informa insurance arrangements, Journa of Deveopment Economics, 40(1), Comoa, M., and M. Fafchamps (2011): Are Gifts and Loans between Househods Vountary?, working paper. De Me, S., D. McKenzie, and C. Woodruff (2008): Returns to Capita in Microenterprises: Evidence from a Fied Experiment, Quartery Journa of Economics, 123(4),

42 (2009): Are Women More Credit Constrained? Experimenta Evidence on Gender and Microenterprise Returns, American Economic Journa: Appied Economics, 1(3), Di Faco, S., and E. Bute (2011): A Dark Side of Socia Capita? Kinship, Consumption, and Savings, Journa of Deveopment Studies, 47(8), Dufo, E., M. Kremer, and J. Robinson (2011): Nudging Farmers to Use Fertiizer: Theory and Experimenta Evidence from Kenya, American Economic Review, 101(6), Dupas, P., and J. Robinson (forthcoming): Savings Constraints and Microenterprise Deveopment: Evidence from a Fied Experiment in Kenya, American Economic Journa: Appied Economics. Fafchamps, M., and S. Lund (2003): Risk-sharing networks in rura Phiippines, Journa of Deveopment Economics, 71(2), Fafchamps, M., D. McKenzie, S. Quinn, and C. Woodruff (2011): When is capita enough to get femae microenterprises growing? Evidence from a randomized experiment in Ghana, working paper. Foster, A. D., and M. R. Rosenzweig (2001): Imperfect Commitment, Atruism and the Famiy: Evidence from Transfer Behavior in Low Income Rura Areas, Review of Economics and Statistics, 83(3), Giné, X., J. Godberg, D. Siverman, and D. Yang (2012): Revising Commitments: Fied Evidence on the Adjustment of Prior Choices, working paper. Gneezy, R. C. U. (2009): Gender Differences in Preferences, Journa of Economic Literature, 47(2), Goeree, J. K., C. A. Hot, and T. R. Pafrey (2003): Risk averse behavior in a generaized matching pennies game, Games and Economic Behavior, 45, Godberg, J. (2010): The Lesser of Two Evis: The Roes of Socia Pressure and Impatience in Consumption Decisions, working paper. Gugerty, M. K. (2007): You Can t Save Aone: Commitment in Rotating Savings and Credit Associations in Kenya, Economic Deveopment and Cutura Change, 55(2), Harrison, G. W., S. J. Humphrey, and A. Verschoor (2010): Choice Under Uncertainty: Evidence from Ethiopia, India and Uganda, Economic Journa, 120(543), Hey, J. D., and C. Orme (1994): Investigating Generaizations of Expected Utiity Theory Using Experimenta Data, Econometrica, 62(6), Hoff, K., and A. Sen (2006): The Kin System as a Poverty Trap, in Poverty Traps, ed. by S. Bowes, S. Durauf, and K. Hoff, Chapter 4, pp Princeton University Press, Princeton. Hot, C. A., and S. K. Laury (2002): Risk Aversion and Incentive Effects, American Economic Review, 92(5),

43 Kapuscinski, R. (2002): The Shadow of the Sun. Vintage Books, New York. Karan, D., R. Knight, and C. Udry (2012): Hoping to Win, Expected to Lose: Theory and Lessons on Micro Enterprise Deveopment, working paper. Kinnan, C. (2011): Distinguishing barriers to insurance in Thai viages, working paper. Levitt, S. D., and J. A. List (2007): What Do Laboratory Experiments Measuring Socia Preferences Te Us About the Rea Word?, Journa of Economic Perspectives, 21(2), Ligon, E., and L. Schechter (forthcoming): Motives for Sharing in Socia Networks, Journa of Deveopment Economics. Ligon, E., J. P. Thomas, and T. Worra (2002): Informa Insurance Arrangements with Limited Commitment: Theory and Evidence from Viage Economies, The Review of Economic Studies, 69(1), Loomes, G. (2005): Modeing the Stochastic Component of Behaviour in Experiments: Some Issues for the Interpretation of Data, Experimenta Economics, 8(4), Luke, N., and K. Munshi (2006): Evidence of a New Vioation of the Independence Axiom, Review of Economics and Statistics, 88(2), Nage, R. (1995): Unraveing in Guessing Games: An Experimenta Study, American Economic Review, 85(5), Ngau, P. M. (1987): Tensions in Empowerment: The Experience of the Harambee (Sef Hep) Movement in Kenya, Economic Deveopment and Cutura Change, 35(3), Patteau, J.-P. (2000): Institutions, Socia Norms and Economic Deveopment. Harwood Academic Pubishers, Amsterdam. Randa, S., E. Coast, and T. Leone (2011): Cutura constructions of the concept of househod in sampe surveys, Popuation Studies, 65(2), Robinson, J. (2008): Limited Insurance Within the Househod: Evidence from a Fied Experiment in Western Kenya, working paper. Scott, J. C. (1976): The Mora Economy of the Peasant. Yae University Press, New Haven. Stah, D. O., and P. W. Wison (1994): Experimenta Evidence on Payers Modes of Other Payers, Journa of Economic Behavior and Organization, 25, (1995): On Payers Modes of Other Payers: Theory and Experimenta Evidence, Games and Economic Behavior, 10, Tanaka, T., C. F. Camerer, and Q. Nguyen (2010): Risk and Time Preferences: Linking Experimenta and Househod Survey Data from Vietnam, American Economic Review, 100(1), Townsend, R. M. (1994): Risk and Insurance in Viage India, Econometrica, 62(3),

44 Train, K. (2003): Discrete Choice Methods with Simuation. Cambridge University Press, New York, NY. Von Gaudecker, H.-M., A. van Soest, and E. Wengström (2011): Heterogeneity in Risky Choice Behavior in a Broad Popuation, American Economic Review, 101(2), Wicox, N. T. (2008): Stochastic Modes for Binary Discrete Choice Under Risk: A Critica Primer and Econometric Comparison, Research in Experimenta Economics, 12,

45 Figure 1: Structure of Experiment 1: Subjects arrive at experiment 2: Expanation of game to a subjects 3: Individua decision-making phase Treatment assignment, exit price reveaed Subject makes investment decision Subject fips coin Subject decides whether to buy out 4: Pubic announcements of investment returns 5: Subjects receive individua payouts in private Time Activities in pain text took pace in primary schoo cassrooms, with a subjects seated together. Activities in bod text took pace in one-on-one interactions between individua subjects and enumerators; during these interactions, subjects and members of the research team were seated at desks in private ocations in the schooyard. Figure 2: Behavior in Large Endowment, Pubic Treatment Partitioning arge endowment behavior by ρ and τ ρ D A C B τ Region A: High ρ, ow τ invest beow 80 in both pubic and private Region B: High ρ, high τ invest beow 80 in private, but exacty 80 in pubic Region C: Low ρ, high τ invest above 80 in private, but exacty 80 in pubic Region D: Low ρ, ow τ invest above 80 in both pubic and private Note: vaues above cacuated using arge endowment=180 and sma endowment=80. 43

46 Figure 3: Expected Utiity in Large Endowment, Pubic Treatment Low ρ, ow τ exampe: (ρ=0.6, τ=0.03) High ρ, ow τ exampe: (ρ=0.98, τ=0.03) Expected utiity Amount invested Expected utiity Amount invested 44 Expected utiity in private Expected utiity in pubic Expected utiity in private Expected utiity in pubic Low ρ, high τ exampe: (ρ=0.6, τ=0.1) High ρ, high τ exampe: (ρ=0.98, τ=0.3) Expected utiity Amount invested Expected utiity Amount invested Expected utiity in private Expected utiity in pubic Expected utiity in private Expected utiity in pubic

47 45 Figure 4: Histograms of Investment in the Business Cup by Treatment

48 46 Figure 5: Income Hiding and Viage-Leve Outcomes

49 Figure 6: Fraction of Subjects Paying to Avoid Announcing 47

50 48 Figure 7: Actua vs. Predicted Investment Decisions

51 Figure 8: Actua vs. Predicted Exit Decisions 49