WPS7675 Policy Research Working Paper 7675 Trust in Government and Support for Redistribution Joana Silva Matteo Morgandi Victoria Levin Latin America and the Caribbean Region Office of the Chief Economist & Social Protection and Labor Global Practice Group May 2016 Policy Research Working Paper 7675 Abstract In many countries safety nets consist predominantly of makes middle-class citizens (particularly among the youth universal subsidies on food and fuel. A key question for and low-trust individuals) more willing to forgo their own policy makers willing to shift to targeted safety nets is under welfare to benefit the poor. Moreover, increasing transpar- what conditions middle-class citizens would be support- ency enhances the relative support for cash-based safety nets, ive of redistributive programs. Results from a behavioral which have greater impact on poverty compared with in-kind experiment based on a nationally representative sample in transfers, but may be perceived as more prone to elite capture. Jordan reveal that increasing transparency in benefit delivery This paper is a product of the Office of the Chief Economist, Latin America and the Caribbean Region and the Social Protection and Labor Global Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at jsilva@worldbank.org, mmorgandi@ worldbank.org and vlevin@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Trust in Government and Support for Redistribution Joana Silva Matteo Morgandi Victoria Levin Keywords : Redistribution, Altruism, Transparency, Development, Experiments. JEL: C93, D31, H23, I38. Silva, Morgandi, and Levin: The World Bank, 1818 H St NW, Washington, DC 20433 (email: jsilva@worldbank.org, vlevin@worldbank.org, mmorgandi@worldbank.org). We are especially grateful to Karla Ho¤ for her help on the design of the experiment and helpful suggestions on previous versions of this paper. We thank the sta¤ of the Center for Strategic Studies in Jordan and particularly: Musa Shteiwi, Yasmina Suleyman, and Walid Alkhatib, for guidance and e¤ective implementation of the …eld work. We are also grateful to Gallup, and in particular Krista Ho¤, Cynthia English and Joe Daly, for guidance on sampling issues and questions phrasing. We thank our colleagues Carole Chartouni, Rania Atieh, Mo- hammed Alloush, and Anne Hilger for their contributions to the …eldwork and assistance with the data; and those who provided feedback at various stages and during various seminars, including Harold Alder- man, Jean Louis Arcand, Benedicte de la Briere, Matias Busso, Gustavo-Javier Canavire, Laura Chioda, Augusto de la Torre, Roberta Gatti, Margaret Grosh, Steen Jorgensen, Adriana Kugler, Daniel Lederman, Julian Messina, Ezequiel Molina, Carlos Parra, Nadine Poupart, Viajenda Rao, Haneen Sayed, Martina Viarengo and Ruslan Yemtsov. We thank Alejandra Martinez for excellent research assistance. 1 Introduction Trust is an essential ingredient of altruistic behavior. A growing body of literature em- phasizes its importance in shaping individual decisions to support the provision of public goods. In a prominent example, a social experiment found that charitable provision was signi…cantly lower if the solicitors of contributions belonged to a minority group that in- spired a lower level of trust among potential donors (List and Price 2009). This behavior can be explained by a principal-agent problem: the principal (i.e. potential donor) can choose, under no obligation, whether to give money to bene…t a group on whose behalf the agent (i.e. donation solicitor) is purportedly acting. The agent’ s optimal strategy is to provide the principal with credible signals that the transferred resources will be used as the principal intends. Mechanisms that enhance the credibility of principal-agent transac- tions can be particularly important in securing support for public goods, including those that bene…t only a subset of the population— such as redistributive policies.1 Obtaining citizens’support for redistributive policies may be especially important in countries where resources are scarce and governments historically enjoy little public trust in their capacity to deliver goods and services fairly and e¢ ciently.2 In fact, even in nondemocratic regimes, citizens have ways to retaliate against unpopular use of public funds or to try to control rent-seeking (Acemoglu, Hassan, and Tahoun 2014). But while there are high expectations about the positive e¤ects of increased transparency on citizens’ trust in — and therefore their support for — redistributive policies, there is a dearth of rigorous evidence on this topic. In this paper, we o¤er experimental evidence on the e¤ect of trust-enhancing mea- sures on public support for redistributive policies. We conducted a behavioral experi- ment on a nationally representative sample of the middle class in Jordan— the Jordan Gives experiment— to identify the e¤ect of enhanced transparency on the support for, and the preferred design of, social safety nets.3 The experiment involved a nationally- representative sample of 420 participants recruited from 21 middle-class primary sampling units (PSUs) (ie localities) across Jordan; within each PSU, participants were randomly assigned to treatment and control groups of 10 subjects. Each participant received a fuel voucher roughly comparable to the daily minimum wage.4 Participants in the control 1 For example, the literature has established that trust in the goverment is an important determinant of compliance with taxation obligations of citizens and …rms (Barone and Mocetti 2011; Friedman, Johnson, and Kaufmann 2000; Johnson et al. 2000; Silverman, Slemrod, and Uler 2014). 2 A government’ s failure to provide such credibility could a¤ect the provision of public goods in di¤erent forms: in more mature democracies through voting outcomes, and in other cases, through tax evasion, exercise of corruption, or public protest. 3 Social safety nets (SSNs), also known as social assistance or welfare schemes, are de…ned as noncon- tributory transfers targeted to the poor or vulnerable. They include income or in-kind support and can be made conditional on certain behaviors of recipients’households (e.g. conditional cash transfers, CCTs) or provided without any conditions (e.g. unconditional cash transfers, UCTs) (Grosh et al. 2008, World Bank 2009). 4 The experiment was followed by a Becker, DeGroot, and Marschak (1964) auction to establish valuation 2 group had to decide whether to give to people in need, without expecting anything in return. Participants in the treatment group were given the opportunity to verify whether their transfers actually reached poor individuals. Our experimental design simulates at the micro level the choice faced by middle-class citizens on whether to support a shift of public resources toward targeted interventions that bene…t the poor, under di¤erent designs of those interventions and di¤erent degrees of certainty about the delivery of the bene…ts to the intended recipients. The experiment thus allows us to provide rigorous evidence on: (1) the propensity for redistribution and relative support for di¤erent redistributive methods; and (2) the underlying impact of transparency and trust on redistributive preferences. In doing so, this paper bridges the gap between evidence obtained from traditional opinion surveys and the behavioral literature. We have two main sets of results. First, although the e¤ect of transparency on altruism is not statistically signi…cant for the whole sample, the transparency-enhancing treatment caused signi…cant increases in the support for redistribution among two groups of partic- ipants: “low-trust” individuals and “youth.” The …rst group consists of individuals who needed a credible signal of trustworthiness in order to exhibit altruism— people who were ex ante suspicious about the implementation of social safety nets. In the control group, such low-trust individuals were signi…cantly less likely to give to the poor than individuals who trusted that social safety nets reach the intended bene…ciaries. In the treatment group, on the other hand, the giving rate of low-trust individuals matched that of high-trust participants, suggesting that the transparency-enhancing treatment mitigated the e¤ect of their initial mistrust on altruistic behavior. The second group— youth, de…ned with di¤erent age thresholds— represents those least likely to give to the poor in the absence of the material guarantees of delivery. With increased transparency, this group experienced the highest increase in the rate of giving. Second, we provide evidence that the transparency-enhancing treatment particularly increased redistribution to the poor through unconditional cash transfers as opposed to in-kind or conditional cash transfers. In the control group, unconditional in-kind transfers were equally preferred to unconditional cash transfers. The treatment group, however, had higher rates of giving through unconditional cash transfers, which became the most popular bene…t delivery option. In sum, enhancing transparency of delivery increased the support for the delivery option— cash— that is generally considered most e¢ cient in reducing poverty relative to in-kind bene…ts. In fact, recent opinion surveys in four Arab countries, including Jordan, showed that the majority of the poor prefer cash-based trans- fers to in-kind bene…ts (Silva, Levin, and Morgandi 2013).5 However, in the absence of of the fuel voucher. This type of auction is a mechanisms commonly used in the literature to induce individuals to reveal their willingness to pay for a given good (Noussair, Robin , and Ru- ieux 2004). It showed that 95 percent of the participants considered the fuel vouchers to be equivalent to cash. 5 The other countries surveyed were the Arab Republic of Egypt, Lebanon, and Tunisia. 3 a credible signal of trustworthiness from the state, cash transfers are also likely to be perceived as carrying the highest risk of capture. Our paper contributes to several strands of research. Much of the existing literature on preferences for redistribution is based on opinion survey data (for example, Alesina and Angeletos 2005; Alesina, Di Tella, and MacCulloch 2004; Alesina and La Ferrara 2002). We add to this literature by providing rigorous evidence from a …eld experiment that elicits preferences on redistribution in a setting where participants face real trade- o¤s, while maintaining the national representativeness of the results. In doing so, we also complement and extend the existing behavioral economics literature that investigates altruistic behavior using samples of higher-education students in laboratory settings (e.g., Charness and Rabin 2002; Fehr and Schmidt 2003). While also using real trade-o¤s to evaluate subjects’preferences, inference from laboratory studies explores human behavior through selected samples that may not be representative of the population of interest. Field behavioral experiments testing altruistic behavior have been less common, have never been based on nationally representative samples, and have not tested the e¤ects of enhanced transparency (Parra 2011; Johansson-Stenman, Mahmud, and Martinsson 2009). Finally, the causal estimates we o¤er based on a nationally representative sample con- tribute to inform pressing policy issues in developing countries, where many governments are deliberating whether and how to shift resources away from costly universal subsidies, which bene…t the middle and upper classes the most, and toward more e¢ cient forms of social safety nets (Silva, Levin, and Morgandi 2013, World Bank 2015).6 Besides the re- distributive implications, one of the main challenges of a policy shift away from subsidies is the high de…cit of trust that governments face as they attempt to replace an easy- to-monitor price subsidy for everyone with a targeted social safety net that can deliver better impacts for the poor and vulnerable at a lower cost. Indeed, important episodes of civil unrest in Jordan in the past decade were linked to attempted reforms of utility or consumption subsidies (Atamanov, Jellema, and Serajuddin 2015).7 6 The Jordan Gives experiment was accompanied by a survey module added to the the Gallup World Poll on citizens’ willingness to reform fuel subsidies, preferred design of reformed safety nets, and willingness to support the reform if preferred safety nets were implemented. The survey was conducted on nationally representative samples in Egypt, Jordan, Lebanon and Tunisia. Most respondents in all four countries preferred savings from fuel or diesel subsidy reform to be distributed to the poor and spent on improving social services. Half of the respondents in Jordan preferred the savings to be distributed only to poor families. More than 20 percent of Jordanians who initially opposed the idea of any subsidy reform would support fuel subsidy reform if the savings were to be distributed to the poor, alone or combined with education and healthcare spending. 7 A similar situation occured during energy subsidy reform episodes in many other countries (see IMF 2013). As in those cases, mitigating measures were considered as part of the reform in an attempt to generate public support for the reform and o¤set adverse e¤ects on the poor. In November 2012, a few months after the Jordan Gives experiment, the petroleum subsidy was removed and an unconditional cash transfer was created to compensate the poor and vulnerable. To ensure transparent administration 4 The remainder of the paper proceeds as follows. Section 2 describes the experiment. Section 3 presents the empirical model. Section 4 reports the main results, highlighting how much participants decided to give, which delivery option encouraged more giving, how the transparency-enhancing treatment a¤ected preferences for giving, and how trust and age a¤ected the impact of the treatment and the relative preferences across the dif- ferent program designs. Section 5 discusses alternative explanations for the increase in giving because of a transparency shock and checks the robustness of the estimates of the interaction between treatment and trust. Section 6 concludes. 2 Research design and data 2.1 Sample design and selection The Jordan Gives experiment was carried out with 420 participants in 21 PSUs in Jordan, on a nationally representative sample of the Jordanian middle class. Participants were identi…ed through a three-stage process: (1) 21 PSUs were drawn from a sampling frame of middle-class enumeration areas in Jordan based on the 2004 census; (2) within each PSU, households were selected using a random walk method; and (3) adults were recruited (one per household) to participate in the experiment using a Kish (1949) table. Based on extensive piloting, a protocol was devised to ensure that two groups of 10 randomly assigned individuals each could be constituted in each PSU (10 for treatment, 10 for control) at the same time and place (see Annex 1 for more details). At the recruitment stage, the invitation letter explained that all participants who appear at the speci…ed place and time (usually a local public school the day after recruit- ment) would receive a fuel voucher of JD 5 (about US$7.50) as a show-up fee and that there would be a chance to keep JD 10 more in such vouchers, depending on the outcome of the meeting. These vouchers were issued by Jordan Petroleum Re…nery Company. They were widely known in Jordan and could be exchanged for gasoline in petrol stations throughout the country. The value of the JD 10 voucher was equivalent to slightly more than the daily minimum wage, or about …ve days of participants’self-reported mean per capita household expenditure. Each invitee who agreed to participate was left with two receipts (one for the JD 5 voucher and the other for the JD 10 vouchers), which they were of the transfer, the government also decided to set up a National Uni…ed Registry (NUR) of the poor and vulnerable as a common platform for eligibility for social assistance programs, with several checks of living standards. Similar instruments (e.g. Cadastro Unico in Brasil, Ficha de proteccion social/Registro social de hogares in Chile) have been the backbone of transparent social assistance programs around the world (see Silva, Levin, and Morgandi 2013; Lindert et al. 2007; and Ministry of Social Development of Chile 2015). Many other developing countries have, in recent years, accompanied subsidy reforms by measures to enhance transparency. This includes, for example, earmarking increased funding for education and infrastructure linked to …scal savings from subsidy reform, creating a website where each person could compute his/her score or enter their national ID and verify his/her (in)eligibility/information, or displaying the list of social assitance bene…ciaries in a public place (IMF 2013). 5 encouraged to bring to the meeting to exchange for real vouchers.8 The sampling frame was based on the Government of Jordan’ s de…nition of “middle class”: middle-class PSUs were identi…ed as those whose households’average annual per capita expenditure was between twice and four times the poverty line (ESC 2008). Annex 1 provides a detailed explanation of the sample design and the selection protocols. The decision to focus the experiment on middle-class behavior was driven by the need to study a population of highest relevance to the policy makers considering safety net reform. Such a reform would imply a redistribution of public funds away from the wealthy and the middle class, as both groups were capturing most bene…ts from universal food and fuel subsidies (Silva, Levin, and Morgandi 2013). The reform under consideration would bene…t the poor, who would gain from the increased magnitude of transfers due to targeting and potentially from attaining a more optimal consumption basket, depending on the design of the new safety net.9 The Jordanian middle class was the group that was likely to loose the most in relative terms from shifts of resources away from universal subsidies towards targeted social safety nets, and that could assemble a su¢ ciently large interest group to thwart the reform. 2.2 The redistributive proposals The experiment was conducted on 20 participants in each randomly selected PSU. Upon arrival to the location of the experiment, participants were randomly allocated to either the control or treatment group and invited to enter a corresponding room. At the start of the experiment, each participant received the two vouchers that had been promised at recruitment stage: a JD 5 voucher as a show-up fee and a JD 10 voucher to use in the experiment.10 The experiment asked participants to make a series of decisions concerning whether to keep their JD 10 fuel voucher or to give it up in exchange for di¤erent scenarios (“pro- posals” ) of assistance to the poor. The exact wording of the proposals, intended to mimic the design of social safety net programs, was as follows: P1 (Unconditional cash transfer): “You give up your JD 10 voucher. Our team gives JD 20 cash per family to 5 poor families in this community.” P2 (Unconditional food transfer): “You give up your JD 10 voucher. Our team gives a food basket worth JD 20 per family to 5 poor families in this community.” 8 To approximate the experience of subsidy reform, which entails the removal of what is often perceived as a citizen’s right, the experiment activated an endowment e¤ect (Kahneman, Knetch, and Thaler 1991) for the fuel vouchers by creating a sense of ownership using receipts with speci…ed voucher values at the time of recruitment. 9 For example, cash transfers would allow poor households to buy goods and services in the amounts providing the highest utility, whereas in-kind transfers (and price subsidies) distort such consumption patterns toward the provided or subsidized goods. 10 To strengthen the endowment e¤ect, initiated at recruitment via receipts, the vouchers were handed out at the very beginning of the experiment. 6 P3 (Unconditional cash transfer and school): “You give up your JD 10 voucher. Our team gives JD 20 cash per family to 2 poor families in this community and JD 60 cash goes to the local public school.” P4 (Cash transfer conditional on training): “You give up your JD 10 voucher. Our team gives JD 20 cash per family to 5 poor families in this community conditional on one family member completing a free training program on work-related skills.” 11 Proposals were revenue-neutral, since the amount to be disbursed in the proposals was equivalent to the total value of fuel vouchers in the room (i.e. 10 participants’ JD 10 vouchers, a total of JD 100). After each proposal was presented, participants were asked to …ll their individual decision cards in silence and con…dentially, marking whether they “accept” the proposal (indicating a preference to see the proposal implemented) or “reject” it (preference to keep the JD 10 voucher).12 Participants were asked to write down their decision on each proposal before being presented with the next proposal. After all four proposals were presented and all four decisions were marked on decision cards, participants were asked to rank the four proposals in the order of preference. At the end, all decision cards were collected by the facilitator, who placed them in a glass bowl. This decision selection process was chosen to ensure that participants had a clear incentive to consider each proposal independently of what they had decided in preceding proposals. A second glass bowl contained numbers 1 through 4, corresponding to the proposal numbers. After all the cards were submitted, the facilitators drew one decision card from the …rst bowl and one number from the second bowl. The decision made regarding the selected proposal number on the selected decision card was implemented on the whole group. If the selected decision 11 The di¤erent proposals correspond to the most common types of safety net schemes. There is an intense debate in the literature over the relative merits of each of these designs. Recent empirical evidence …nds that cash transfers are generally as e¤ective as food transfers in improving nutritional outcomes (Cunha 2012; Attanasio, Battistin, and Mesnard 2012; Hoddinott, Sandström, and Upton 2014), but they are more e¢ cient when markets function well (that is, are not plagued by hyperin‡ ation, con‡ icts, or supply constraints) (Busso and Galiani 2014). Several recent papers have discussed the marginal impact of attaching conditions to cash transfer programs. Although their administration is costlier relative to uncoditional transfers, they intend to address market failures that lead to underinvestment in education or health by imposing certain behaviors on recipient households (Hanlon, Barrientos, and Hulme 2010). Recent studies found that such schemes generally improve the conditioned-on outcome, but pose trade-o¤s with respect to gains in overall welfare, which can be particularly large in the presence of low quality (or accessibility) of conditioned services (Baird, McIntosh, and Ozler 2011; Attanasio, Veruska, and Marcos 2015; Blattman, Fiala, and Martinez 2015; Benhassine et al. 2013). On the other hand, transfers condi- tional on educational outcomes usually also provide a valuable mechanism to improve parents’monitoring over children’ s school attendance (Bursztyn and Co¤man 2012). Finally, the literature discusses that ac- companying cash transfers to the poor with …nancing of public goods with a broader user base (such as schools) promotes acceptance of public social assistance, and thus makes …rst-best redistribution (targeted safety nets) possible (Gahvari and Mattos 2007). 12 To prevent peer pressure from biasing their decisions, the participants were not allowed to discuss their decisions with each other. 7 was “accept,”then the JD 10 voucher was collected from each participant and the selected proposal would be later implemented in the local community.13 If the selected decision was "reject," all participants would keep their vouchers. The experiment was followed by a Becker, DeGroot, and Marschak (BDM) (1964) auction and the collection of basic demographic, socio-economic, and attitudinal characteristics of participants (via a written questionnaire).14 To better understand the reasoning of study participants, debrie…ng focus groups discussions were also conducted. They showed that the proposals, decision mechanisms and consequences of the decisions in the context of the experiment were well understood by participants. 2.3 Audiovisual implementation of the experiment Results of behavioral experiments can be biased by the heterogeneity of implementation. Biases can include accidental priming to values or anchoring to certain numbers, and they can also originate from the identity of the facilitator, see Brewer and Chapman (2002) and Furnham and Boo (2011) for a survey. To ensure that the messages conveyed to participants were homogeneous, the experi- ment was implemented through an 18-minute video that featured a Jordanian woman with a neutral background explaining the purpose of the experiment and giving directions to participants at each stage. The video presented the decision cards and the proposals and il- lustrated graphically the proposal selection mechanics.15 Participants were presented with 13 In each selected PSU, on the day of the experiment, the facilitators arrived equipped to implement any of the potential outcomes of the experiment. Facilitators’ cars contained the food baskets, training vouchers, and cash. Contact information of poor families that could be the recipients of these bene…ts was provided by the local community leader. 14 In the BDM auction participants were told that they had the possibility of exchanging for cash the JD 5 fuel vouchers that they had received as a show-up fee. The video explained and illustrated the auction mechanism to ensure that all participants understood that their dominant strategy was to reveal their true preferences. They were then asked to write down the minimum cash amount, in denominations of JD 1, that they would need to receive in order to “sell back” their vouchers. Cards displaying di¤erent cash amounts (1 through 5) were then placed in a bowl, and one of them was randomly drawn. If the drawn value was above the value written by the participant, the participant would retain his or her voucher. If it was equal or lower, he or she would exchange the voucher for the cash amount drawn. The auction revealed that more than 95 percent of the participants considered the voucher to be equivalent to cash: that is, they wrote “5” , as they were not ready to exchange their JD 5 voucher for a lower monetary value than its nominal value. This is understandable given that 57 percent of respondents had a car in their households, and those who did not could also have had motorcycles or readily exchanged the voucher. 15 During the pilot, the use of pictures emerged as important to improve participants’ understanding of the experiment. To further ensure that participants had a good understanding of the experiment’ s mechanics, particularly of the fact that their decision, if selected, would a¤ect everyone’ s payo¤s, the experiment was preceeded by a mock trial (…rst part of the video). Participants were given a chocolate as an endowment, and they wrote down their preference between keeping their chocolate or getting a postcard (proposal 1), and between keeping their chocolate and having one of the facilitators recite a poem about Jordan (proposal 2). As with the actual experiment, one decision was randomly drawn and implemented on the whole group. 8 a sequence of four proposals. The order in which the the four proposals were presented was randomized at the PSU level by producing multiple versions of the same video. This procedure aimed to avoid any systematic anchoring e¤ect due to a particular proposal or- der. The facilitator’s role was to distribute and collect decision cards and questionnaires, answer questions according to a pre-developed answer script, and lead the focus group discussion that followed the experiment. 2.4 The transparency-enhancing treatment Treatment status was assigned at the PSU level and the sample of 20 individuals in each PSU was randomly divided into two groups of equal size: treatment and control. After the random assignment, the experiment was started simultaneously in two separate rooms, one room with individuals in the control group and another room with individuals in the treatment group. In each PSU we implemented the experiment only once. Hence, the total sample contained 420 individuals in 21 PSUs: 210 individuals in the control group and 210 individuals in the treatment group. The video for the treatment groups contained all the features of the video for the control groups, but it included additional information that would make the transfer delivery to the poor and content more transparent for participants. In particular, individuals in the treatment group were o¤ered the option to accompany the facilitator after the experiment to witness the actual implementation of the proposal among poor families, if the randomly- selected decision was an acceptance of the proposal. To reinforce this message, right before participants were asked to make their decisions on each proposal, they were told that the facilitator would wait after the conclusion of the experiment for anyone who wanted to follow and witness the implementation of the proposal.16 In addition, participants in the treatment groups were shown in the video a basket of essential supplies worth JD 20, as in proposal 2 (unconditional in-kind transfer). Thus, the treatment increased transparency of the redistributive proposals by alleviating par- ticipants’uncertainty about the delivery of the transfer to the intended bene…ciaries and about the value of the JD 20 in the case of an unconditional food transfer. The treatment was chosen as a result of a focus group on the perceived barriers to redistribution as well as consultations with Jordanian experts to o¤er concrete recommendations on measures to implement as part of a fuel subsidy reform. 2.5 Data The data used in this paper were collected between late May and June 2012. The quan- titative data from Jordan Gives includes decisions made by each individual participant 16 Indeed, in some cases, participants in the treatment group did decide to follow the facilitator and, as highlighted by participants in the focus group discussion, the mere availability of this option sent a credible commitment signal of trustworthiness. 9 during the experiment, their valuation of fuel vouchers obtained via a post-experiment BDM auction, as well as basic demographic, socioeconomic, and attitudinal information on each participant collected via a short written survey administered after the experiment. Finally, we collected a rich qualitative dataset from structured in-depth focus groups that were conducted by facilitators after all quantitative data was collected. 3 Empirical model We estimate by ordinary least squares (OLS) a set of treatment-e¤ects models of the following form: Yi = + Ti + Xi + ei (1) where: Yi is the an outcome variable (mean giving rate in models using information from all the proposals, or the binary decision to accept or reject a speci…c proposal in models using information from a speci…c proposal) for individual i ; Ti is an indicator variable equal to one if the individual was assigned to the treatment group and 0 otherwise; Xi is the vector of baseline characteristics; and ei is the error term. The parameter of interest is the average treatment e¤ect. Estimates are computed with a linear regression model even for binary dependent variables, such as decisions on speci…c proposals, as the coe¢ cients are nearly identical to the marginal e¤ects of a probit model (as discussed in Miguel, Satyanath, and Sergenti 2004). The advantage of using OLS is the availability of an established procedure to compute clustered wild bootstrap-t standard errors, which are more suitable for estimations with a small number of clusters. The standard errors are clustered at the PSU level, which accounts for the design e¤ect of our PSU level treatment and for heteroscedasticity inherent in the linear probability model.17 We estimate results of equation (1) for six outcomes, two aggregating individual i’ s decisions across the four proposals (as described below) and four using decisions on each proposal at a time. “Mean giving rate” is the share of accepted proposals to give up the fuel voucher (out of possible four); “Frequent giver”is an indicator variable equal to one if individual i’s mean giving rate is greater than 0.5, i.e. if the individual gave up the voucher in the majority of the proposals. The other four outcomes are binary indicator variables equal to one if individual i indicated that he or she would give up his/her voucher for that speci…c proposal (unconditional cash transfer, unconditional food transfer, unconditional cash transfer and school, or cash transfer conditional on training). 17 Clustering at the PSU level was used to account for the …rst stage of the sampling strategy, which picked PSUs from the census sampling frame. Clustering thus adjusts the standard errors to take into account intra-cluster correlation, which could be relatively high for outcomes related to redistribution pref- erences. Given that Huber-White heteroskedastic-standard errors (commonly known as “cluster-robust” ) are potentially underestimated when the number of clusters is small (as discussed in Bertrand, Du‡ o, and Mullainathan 2004; Cameron, Gelbach, and Miller 2008; Cameron and Miller 2015), in this paper we re- computed all standard errors with wild cluster bootstrap-t statistics following the procedure by Cameron, Gelbach, and Miller (2008), which avoids standard error underestimation in the presence of few clusters. 10 In all regressions, we include the following variables as controls: gender, education level, residence in the capital city, and the number of cars in the household (the latter to proxy for relative wealth).18 These variables were chosen because they are strongly predictive of outcomes and, as a result, they improve the precision of the impact estimates.19 4 Main Results 4.1 Sample balance Sample balance statistics are presented in Table 1, testing the outcome of the random- ization process at the PSU level, and thus ensuring that observable characteristics of participants in the treatment groups were similar to those in the control groups. The standard errors of the mean di¤erence between treatment and control groups are cor- rected for intra-cluster correlation at the level of the 21 PSUs. Panel A shows balance on individual characteristics, Panel B shows balance on household attributes and Panel C shows balance on baseline giving behavior. Overall, the experiment appears well balanced between the treatment and control groups over a broad range of outcomes (see column 4). 4.2 E¤ect of the treatment Table 2 describes the e¤ect of the treatment on the probability of giving (i.e. choosing to accept a proposal and thus give up the fuel voucher), obtained via a treatment-e¤ect regression controlling for participants’basic demographic and socioeconomic characteris- tics. The constant term in these regressions represents the mean in control groups, while the coe¢ cient on the indicator variable for individuals’ assignment to treatment groups represents the impact of the treatment. Table 2 reports the di¤erence between treatment and control means in the giving rates (both aggregate and for each of the four proposals) as well as in the share of participants who gave up the voucher in more than two proposals. Column 5 shows full sample averages. Results indicate that mean giving (at the partici- pant level) was 67 percent and that more than half of all participants in the experiment (61 percent) were frequent givers (i.e. opted to accept more than two proposals).20 Across pro- posals, the unconditional food transfer proposal attained the highest acceptance rate (70 percent), closely followed by the unconditional cash transfer proposal (69 percent). Given the monetary value that the voucher represented for the subjects, the average giving rate of 67 percent was remarkable, compared with the giving rates found in other experiments, which ranged between 20 percent and 37 percent (DellaVigna, List, and Malmendier 2012; 18 Appendix Table A.1 presents the summary statistics. 19 Appendix Table A.2 presents more parsimonious results without control variables. 20 About half of all participants in the experiment opted to accept all four proposals (i.e. to give up their vouchers to the poor in each of the presented scenarios), 17 percent decided to reject all the proposals (i.e. never to give up their vouchers), while the remaining one-third decide to give up their vouchers for some but not all the proposals. 11 List and Price 2009; Parra 2011). However, the design of the present experiment was unique in the literature. In contrast to Jordan Gives, classic dictator games allow the principal to de…ne transfer size.21 Fundraising experiments, on the other hand, do not provide participants with any endowment, while Jordan Gives took measures to enhance the endowment e¤ect for the voucher.22 Another potentially important distinctive feature of the present experiment is the identi…cation of direct recipients of the transfer as local poor families rather than more abstract notions of giving to charities.23 Columns 1 to 4 of Table 2 show the results from equation (1). The point estimates of giving rates suggest that individuals in the treatment groups are slightly more likely to give up their vouchers than individuals in the control groups. Nonetheless, none of the average treatment e¤ects reaches statistical signi…cance at conventional levels (with the lowest p-value of 0.11 for the unconditional cash transfer proposal). 4.3 Heterogeneity of e¤ects Although the results for the full sample are not signi…cant at the conventional levels, the treatment appeared to have signi…cant impacts on two speci…c subgroups of participants: individuals with low trust in the delivery of safety nets and young people. In fact, as shown in Table 3, the level of trust in the delivery of social safety nets, as measured with a post- experiment attitudinal question, appears to mediate the e¤ect of treatment, particularly with regard to the two unconditional cash transfer proposals (i.e. unconditional cash transfer and cash transfer with school …nancing).24 The results are obtained by estimating equation (1) above separately for individuals that reported being completely or somewhat 21 In classic “dictator games” participants can determine the share of the received amount to distribute in a single-shot game. Parra (2011) found that Ghanaian participants shared 37 percent of the endowment in the baseline scenario. Forsythe et al. (1994) found a 25 percent giving rate in a dictator experiment where the donor knew the identity of the receiver. However, in our case participants had a discrete choice between giving up or retaining their vouchers, for a repeated number of proposals that were heterogeneous by design. We also had a full loss or full retention of the endowment in each proposal, approximating the experience of a subsidy reform which is also one shot. The design of both "whole versus part" and "one shot versus repeated" was adopted to approximate the experience of subsidy reform. 22 In fundraising experiments List and Price (2009) and DellaVigna, List, and Malmendier (2012) found, respectively, a 20 percent and a 25 percent giving rate in the United States. Our …ndings would also be consistent with individuals being more generous when their endowmente depends solely on a random shock (Cappelen et al. 2007 and Cherry, Frykblom, and Shogren 2002). 23 Interestingly, the proposal in which individuals could give up their voucher to both help the poor and to …nance a public good (the local school) proved to be the least popular proposal among participants. Although other experiments have suggested that altruism could be enhanced by introducing a chance of personal gain (for instance, lotteries, as in Landry et al. 2006), in this case individuals may have thought that contributing such a limited amount of funding to the school was neither bene…cial to themselves nor as impactful as a charitable transfer given directly to the poor. 24 The question was “How con…dent are you that the public funds allocated for social assistance reach the poor?” The response scale had four options: “completely con…dent,” “somewhat con…dent,” “not very con…dent,” and “not con…dent at all.” Low-trust individuals are de…ned as those who responded with the two latter options. 12 con…dent that public funds for social assistance reach the poor (de…ned as “high trust” ) and those who were not con…dent about this (de…ned as “low trust” ). The treatment e¤ect is always higher among low-trust individuals and is statistically signi…cant for three outcomes: the mean giving, the propensity to be a frequent giver and the unconditional cash transfer. The results point to two other important …ndings. First, the comparison of columns 1 and 4 reveals that among individuals in the control group, the mean rate of giving was 18 percentage points higher for high-trust individuals than for low-trust individuals, which is a statistically signi…cant di¤erence. It is also striking that this di¤erence was driven essentially by the proposals involving unconditional cash transfers: in the control group, high-trust individuals were 31 and 23 percentage points more likely than low-trust individuals to give up their vouchers for, respectively, the unconditional cash transfers and cash transfer with school …nancing. Second, in the treatment group, we observe no statistically signi…cant di¤erences in mean giving rate between high-trust and low-trust participants. These relatively smaller di¤erences are due to higher giving rates among low- trust individuals in the treatment groups compared to the control groups. For instance, in the case of unconditional cash transfer, 44 percent of low-trust participants gave up their vouchers in the control group, while 60 percent did so in the treatment group (implying a 16 percentage point treatment e¤ect), whereas the treatment e¤ect for high-trust participants was minimal (less than 2 percentage points). One exception was the proposal of giving cash both to the poor and to the local school; in this case the transparency treatment enhanced the overall giving rate for all participants, but did not reduce the gap in giving between low- and high-trust individuals. The treatment e¤ect was also heterogeneous based on participants’age. Figure 1 sum- marizes the average treatment e¤ect estimates for youth and older adults’ subsamples. Compared to older adults, young individuals (aged 18– 29) were far more susceptible to changing their behavior as a result of the transparency-enhancing treatment. The treat- ment e¤ects are always higher for young individuals and are statistically signi…cant in two outcomes: mean giving and the unconditional cash transfer proposal. Exploring the data further reveals a certain level of overlap between low trust and age, which explains why treatment impacts are highly heterogeneous on both dimensions. Figure 2 summarizes the local e¤ect of age, as a continuous variable, on average giving behavior in both treatment and control groups according to the participants’ level of trust that public funds for social assistance reach the poor. Panel A presents results for aggregate/all proposals while Panel B presents results for each individual proposal. There is an obvious upward-sloping relationship in the control groups between age and giving rate, implying that youth are less likely to redistribute their endowment. However, for low-trust youth, the transparency-enhancing treatment ‡ attens the age-giving curve, at least until around age 50, and makes these youth about as likely to give up their fuel vouchers as middle-aged individuals who have high trust in the provision of safety nets. 13 Panel B con…rms this pattern at the level of speci…c proposals. Taken together, Figure 3 shows that while young individuals are clearly those who are most a¤ected by the treatment in three out of the four proposals, treatment increases the giving rate the most among those who are both low-trust and young. This result is for- mally con…rmed in Table 4. Column 1 shows that trust is an important mediator for the transparency-enhancing treatment, with the interaction term of trust and treatment ob- taining a signi…cant negative coe¢ cient in almost all model speci…cations. In other words, providing a signal that the redistributive transfer would reach the intended bene…ciaries is most e¤ective for “low-trust”individuals, i.e. those who ex ante have low con…dence in the functioning of redistribution ‡ ows. The heterogeneity of treatment e¤ects is demonstrated further by controlling for par- ticipants’ age. As mentioned above, young participants were less likely to give up their fuel vouchers; this is con…rmed in column 2 in Table 4 by a consistently signi…cant neg- ative coe¢ cient on the indicator for youth. The regressions were also repeated on the subsamples of young (ages 18– 29) and older individuals separately, with results presented in columns 3 and 4 of Table 4. This analysis reveals that the treatment had a signi…cant and larger impact among the youth, con…rming the earlier …ndings of Figure 3. 4.4 E¤ects on preferred transfer modality: Cash versus in-kind assis- tance In addition to the di¤erences in the giving rates, treatment and control groups also di¤ered in terms of the proposal that they favored the most.25 Table 5 reports the favorite proposal among participants who gave up their voucher in response to at least one proposal. Because the control groups contained more individuals who never gave up their vouchers, the two samples are not of identical size, so we compare the distribution of responses. In the control group, the unconditional food transfer option was the most frequent favorite, with a third of respondents picking it as their most-preferred proposal. With food transfers being more visible than cash transfers, food could be considered to be more tractable and of lesser interest for capture by the better-o¤ in the context of high potential fraud and corruption. However, after getting a transparency-enhancing signal, participants in the treatment group were much less likely (by nearly 13 percentage points) to pick the food transfer as their most-favored proposal. Thus the treatment appears to have enhanced the attractiveness of cash-based delivery options that may be perceived as more prone to capture but also commonly considered to be more e¢ cient in poverty reduction than in-kind food transfers (Currie and Gahvari 2008). 25 As a reminder, after all decisions were marked, participants were asked to rank the proposals in the order of preference. In order to focus on true preferences, the analysis that follows uses only responses by participants who chose to give up the voucher for at least one proposal, and for those who gave up the voucher for strictly one proposal, it imposes that proposal as the revealed favorite. 14 5 Alternative explanations and robustness checks Although the trust channel discussed in the introduction is, in our view, the most plausible explanation for the increase in giving because of a transparency shock, it is not the only possible one. As discussed in section 2.4, increased awareness about the value of giving o¤ers a plausible alternative mechanism linking treatment to giving rates. It is possible that in addition to increasing the ability to monitor delivery, the treatment has also corrected some informational asymmetries on the bene…ts of the transfer for poor families. In particular, better-o¤ or more educated participants could be less familiar with the consumption basket of the poor and unaware that JD 20 could buy as many essential supplies. This could happen because participants’consumption basket di¤ers from that of the poor either in terms of the products or their quality. Thus, the fact that participants in the treatment groups were shown in the video a basket of essential supplies worth JD 20, as in proposal 2 (unconditional in-kind transfer), could have increased awareness of the value of giving among the better-o¤ participants. However, in this case, we would expect the treatment e¤ect to vary according to the level of education or income. We check this formally in Table 6. Panel A describes the e¤ect of the interaction between treatment and being skilled (de…ned as having completed high school or more) on giving, controlling for participants’basic demographic and socioeconomic characteristics (gender, location in the capital city and number of cars in the household).26 Each column reports a regression on participants’ decisions on that speci…c proposal. In all speci…cations, the interaction term between treatment and skilled is not signi…cant. Panel B describes the e¤ect of the interaction between treatment and being high income (a variable equal to one if the participant reported an above mean value of per capita income) controlling for gender, location in the capital city and education.27 In all speci…cations, the interaction term is not signi…cant. Panel C considers an alternative de…nition of high income using participants’ responses on their subjective position on the income distribution of Jordan (self-identi…ed income quartile). In particular, self-identi…ed high income is a dummy variable equal to one if the participant declared to be in income quartiles three or four. Also in this case, the interaction term between treatment and high income is not signi…cant in all model speci…cations. This evidence indicates that trust rather than information/awareness on the value of giving explains the increase in giving because of a transparency shock. To provide additional evidence for the robustness of the trust-based channel, we con- sider how the e¤ect of the interaction between treatment and trust on giving varies across groups of the population with di¤erent scope for giving. We consider two dimensions: economic distance from the poor (measured using education and income) and how much the participant values redistribution (measured using social norms that have been found 26 Results without controlling for number of cars in the household are similar. These are available from the authors upon request. 27 The results obtained controlling for number of cars in the household, or excluding the education variables are similar to the ones reported and are available from the authors upon request. 15 to be correlated with redistributive behavior in the literature).28 To construct indicators of the second dimension we use information from three survey questions that participants answered after the experiment, asking about participants’ agreement with the following statements: (1) People are poor in Jordan because of bad luck or injustice (rather than laziness or lack of willpower); (2) Successful careers are a matter of luck and connections (rather than hard work)29 ; and (3) A just society should make people’ s incomes more equal. Tables 7 and 8 present the results. Table 7 is similar to Table 4 but focuses on distance from the poor rather than age. Results show that distance from the poor does not appear to mediate the e¤ect of the transparency-enhancing treatment, with the interaction term of distance from the poor and treatment being not signi…cant in most speci…cations. Table 8 focuses on social norms and adds to our baseline speci…cation controls for the scope for giving using the three attitudinal questions described above (Columns 1 to 3) and a composite index, produced with polychoric principal-components analysis, for the importance of redistributing (Column 4). When controls for indicators of being in favor of redistribution are included, results on the e¤ect of the interaction term between treatment and trust on giving remain largely unchanged. This suggests that the relationship of interest is not being driven by di¤erences in these indicators. Finally, in Table 9 we check the robustness of our results to a di¤erent age threshold for youth. Table 9 is similar to Table 4 but considers youth to be those aged 18 to 34. Results on the interaction term between trust and treatment are robust to this alterantive de…nition. 6 Conclusion This paper analyzes income redistribution preferences and the e¤ect of program design and enhanced transparency on willingness to give up a personal endowment. It uses data from a behavioral experiment conducted on a nationally representative sample of the Jordanian middle class. In contrast to opinion surveys, the experiment evaluated preferences using real trade-o¤s. It contributes to the literature on redistributive preferences by o¤ering 28 The literature has shown that the demand for redistribution varies according to personal beliefs about the causes of poverty and success, with those who believe that people are poor because of bad luck or injustice or that success is the result of individual e¤ort rather than luck being more prone to redistribution (Alesina, Glaeser, and Sacerdote 2001; Alesina and Glaeser 2004; Alesina and Angeletos 2005; Alesina and La Ferrara 2005; Alesina and Giuliano 2011; Charness and Rabin 2002; Konow 2010). 29 66 percent of the participants expressed a general belief that poverty is the result of bad luck or injustice rather than laziness, 37 percent believed that hard work usually brings success and 80 percent agreed that society should make people’ s incomes more equal. These perceptions are in line with those in Latin America and Western Europe but in stark contrast with the United States, where government redistribution from the rich to the poor is less extensive. This might also be a factor behind the higher giving rates in our experiment compared to the classic dictator games or fundraising experiments in the United States. 16 experimental evidence and results on the in‡ uence of transparency in bene…t delivery on the overall support for targeted social assistance and on program design preferences. The paper shows that support for redistributive programs is sensitive to the level of trust in the system’ s ability to deliver bene…ts to intended bene…ciaries. In fact, although transparency does not appear to signi…cantly a¤ect the overall rates of giving (controlling for program design), it does have a signi…cant positive e¤ect on the giving by youth and by those people who exhibit low trust in the existing delivery of safety nets, especially in the case of unconditional cash transfers. Moreover, among those low-trust individuals, enhanced transparency makes cash-based transfers more attractive than in-kind transfers. Because the latter are generally less e¢ cient but may be perceived as less prone to elite capture, transparency could thus also enhance program e¢ ciency by allowing policy mak- ers to switch from in-kind to cash transfers without losing the support of their middle-class citizens. Annex 1: Sample design and selection protocols The experiment’ s sampling strategy adopted the de…nition developed by the Government of Jordan’ s study of the middle class (ESC 2008), which de…ned the middle class as those households that have per capita incomes between twice and four times the Jordan’ s national poverty line. This de…nition corresponded to the population between the 4th and the 8th income decile according to the 2004 Jordanian census, the latest available at the time. For this study, middle-class primary sampling units (PSUs) in the census were selected by a three-step process: 1) constructing a proxy means test regression using Jordan’ s 2010 Household Expenditure and Income Survey;30 2) applying coe¢ cients from that regression to the 2004 census data; and 3) choosing PSUs with resulting average scores between 4th and 8th income deciles. Within the population of middle-class PSUs, 21 sampling units were selected for the experiment via random selection with probability proportionate to size. Within each sampling unit, the following protocol was used to recruit the needed 20 individuals (10 for treatment, 10 for control) at the same time and place. The day before the experiment in a PSU, a team of enumerators would visit the selected PSU, and the team leader would use a random walk method to select households for recruitment. Enumerators then visited this sample of households, introducing themselves with an invitation letter from the Center for Strategic Studies in Jordan (CSS), and used a Kish (1949) table to identify one eligible person who was at least 18 years old to be invited to a meeting the next day at a the reserved location (usually a nearby public school). The purpose of 30 The regression included the following variables, which appear in both the Household Expenditure and Income Survey and the census: average household size, owning a …xed phone, a computer, internet connection, central heating, microwave, home ownership, and having at least one family member with university education 17 the meeting was not directly explained to the invitees except to say that they have been randomly selected in their community to participate in a research study by the CSS, and that it is not related to market research. To compensate participants for their time, the invitation letter explained that all par- ticipants would receive a fuel voucher of JD 5 (equivalent to about US$7.50) as a show-up fee and that there would be a chance to keep JD 10 more in such vouchers, depending on the outcome of the meeting. Each invitee who agreed to participate was left with two re- ceipts, which they were encouraged to bring to the meeting to exchange for real vouchers: one for the show-up fee of JD 5, and the other for JD 10. If the person selected by the Kish table was not present at the time of enumerators’ …rst visit, enumerators would schedule an appointment and visit the household again in the evening to make the invitation in person. Based on extensive piloting, protocols were designed to replace households whose members refused the invitation and to ensure that two groups of 10 randomly assigned individuals could be constituted in each PSU. To ensure that 20 participants would show up at the set time to the next day’ s meeting, enumerators invited 30 individuals per PSU, emphasizing that it is very important to show up on time. At the start of the meeting, all present participants signed a consent form and were randomly assigned to either the treatment group classroom or the control group classroom. Despite the appeal to show up on time, the team had to delay the start of the experiment virtually every time to …ll the quota of 20 participants. After 20 participants had assembled, randomization into control and treatment groups occurred. Any latecomers were assembled into a mock group, which received the show-up fee and completed a questionnaire. References Acemoglu, D., T. Hassan, and A. Tahoun (2014): "The Power of the Street: Evidence from Egypt’s Arab Spring." National Bureau of Economic Research, Working Paper 20665, Cambridge, MA. Alesina, A. and P. Giuliano (2011): Preferences for Redistribution. In Handbook of Social Economics, A Bisin and J. Benhabib (eds.), pp. 93-132, North Holland. Alesina, A., and E. Glaeser (2004): Fighting poverty in the US and Europe: A world of di¤ erence, Oxford: Oxford University Press. Alesina, A., and E. La Ferrara (2002): "Who Trusts Others?" Journal of Public Eco- nomics, 85 (2), pp. 207–34. — — — (2005): "Preferences for Redistribution in the Land of Opportunities," Journal of Public Economics, 89 (5), pp. 897–931. 18 Alesina, A., and G. Angeletos (2005): "Fairness and Redistribution," American Economic Review, 95 (4), pp. 960–80. t the US have a European- Alesina, A., E. Glaeser, and B. Sacerdote (2001): "Why doesn’ style welfare system?" Brookings Papers on Economic Activity, 2, pp. 187-277. Alesina, A., R. Di Tella, and R. MacCulloch (2004): "Inequality and Happiness: Are Europeans and Americans Di¤erent?" Journal of Public Economics, 88 (9–10), pp. 2009– 42. Atamanov, A., J. Jellema, and U. Serajuddin. (2015): "Energy Subsidies Reform in Jor- dan: Welfare Implications of Di¤erent Scenarios," Policy Research Working Paper 7313, World Bank, Washington, DC. Attanasio O., E. Battistin, and A. Mesnard (2012): "Food and Cash Transfers: Evidence from Colombia," Economic Journal, 122, pp. 92– 124. Attanasio, O., O. Veruska, and V. Marcos (2015): "Should Cash Transfers Be Conditional? Conditionality, Preventive Care, and Health Outcomes." American Economic Journal: Applied Economics, 7 (2), pp. 35-52. Baird, S., C. McIntosh, and B. Ozler (2011): "Cash or Condition? Evidence from a Cash Transfer Experiment." Quarterly Journal of Economics, 126 (4), pp. 1709–53. Barone, G., and S. Mocetti (2011): "Tax Morale and Public Spending Ine¢ ciency." In- ternational Tax and Public Finance, 18 (6), pp. 724–49. Becker, G., M. DeGroot, and J. Marschak (1964): "Measuring utility by a single-response sequential method," Behavioral Science, 9 (3), pp. 226–32. Benhassine, N., F. Devoto, E. Du‡o, P. Dupas, and V. Pouliquen (2013): "Turning a shove into a nudge? A" labeled cash transfer" for education," National Bureau of Economic Research, Working Paper 19227, Cambridge, MA. Bertrand, M., E. Du‡ o, and S. Mullainathan (2004): "How Much Should We Trust Di¤erences-in-Di¤erences Estimates?" Quarterly Journal of Economics, 119, pp. 249-275. Blattman, C., N. Fiala, and S. Martinez (2015): "Generating skilled self-employment in developing countries: Experimental evidence from Uganda." Quarterly Journal of Eco- nomics. Brewer, N., and G. Chapman (2002): "The Fragile Basic Anchoring E¤ect." Journal of Behavioral Decision Making, 15 (1), pp. 65–77. Bursztyn, L. and L. Co¤man (2012): "The Schooling Decision: Family Preferences, In- tergenerational Con‡ict, and Moral Hazard in the Brazilian Favelas." Journal of Political Economy, 120 (3), pp. 359-397. 19 Busso, M., and S. Galiani (2014): "The Causal E¤ect of Competition on Prices and Quality: Evidence from a Field Experiment", Working Paper 20054, National Bureau of Economic Research, Cambridge, MA. s Guide to Cluster-Robust Inference." Cameron, A., and D. Miller (2015): "A Practitioner’ Journal of Human Resources, 50 (2), pp. 317– 72. Cameron, A., J. Gelbach, and D. Miller (2008): "Bootstrap-based improvements for in- ference with clustered errors." Review of Economics and Statistics, 90, pp. 414–427. Cappelen, A., A. Hole, E. Sorensen, and B. Tungodden (2007): "The pluralism of fairnes- sideals: An experimental approach." American Economics Review, 97, pp. 818-827. Charness, G., and M. Rabin (2002): "Understanding social preferences with simple tests." Quarterly Journal of Economics, 117 (3), pp. 817–69. Cherry, T., P. Frykblom, and J. Shogren (2002): "Hardnose the dictator." American Eco- nomic Review, 92, pp. 1218-21. Cunha, J. (2012): "Testing Paternalism: Cash Versus In-kind Transfers in Rural Mex- ico." Research paper, Calhoun Institutional Archive of the Naval Postgraduate School, Monterey, CA. DellaVigna, S., J. List, and U. Malmendier (2012): "Testing for Altruism and Social Pressure in Charitable Giving." Quarterly Journal of Economics, 127 (1), pp. 1–56. ESC (Economic & Social Council) (2008): “Assessing the Middle Class in Jordan.”Policy paper, ESC, Amman. Fehr, E., and K. Schmidt (2003): "Theories of Fairness and Reciprocity: Evidence and Economic Applications." In Advances in Economics and Econometrics: Theory and Ap- plications, Eighth World Congress (Econometric Society Monographs) (Volume 1), edited by M. Dewatripont, L. Hansen, and S. Turnovsky, pp. 208– 57, Cambridge: Cambridge University Press. Forsythe, R., J. Horowitz, N. Savin, and M. Sefton (1994): "Fairness in Simple Bargaining Experiments." Games and Economic Behavior, 6 (3), pp. 346– 69. Friedman, E., S. Johnson, and D. Kaufmann (2000): "Dodging the Grabbing Hand: The Determinants of Uno¢ cial Activity in 69 Countries." Journal of Public Economics, 76 (3), pp. 459–93. Furnham, A., and H. Boo (2011): "A Literature Review of the Anchoring E¤ect." The Journal of Socio-Economics, 40 (1), pp. 35–42. 20 Gahvari, F., and E. Mattos (2007): “Conditional Cash Transfers, Public Provision of Private Goods, and Income Redistribution.” American Economic Review, 97 (1), pp. 491– 502. Grosh, M., C. Del Ninno, E. Tesliuc, and A. Ouerghi (2008): For Protection & Promotion: The Design and Implementation of E¤ ective Safety Nets, Washington DC: World Bank. Hanlon, J., A. Barrientos, and D. Hulme (2010): Just Give Money to the Poor: The Development Revolution from the South, West Hartford, CT: Kumarian Press. Hoddinott, J., S. Sandström, and J. Upton (2014): "The Impact of Cash and Food Trans- fers: Evidence from a Randomized Intervention in Niger." Discussion Paper, International Food Policy Research Institute (IFPRI), Washington, DC. IMF (International Monetary Fund) (2013): Case Studies on energy subsidy reform: lessons and implications. Washington, DC: IMF. Johansson-Stenman, O., M. Mahmud, and P. Martinsson (2009): "Trust and Religion: Experimental Evidence from Rural Bangladesh." Economica, 76 (303), pp. 462–85. Johnson, S., D. Kaufmann, J. McMillan, and C. Woodru¤ (2000): "Why Do Firms Hide? Bribes and Uno¢ cial Activity after Communism." Journal of Public Economics, 76 (3), pp. 495–520. Kahneman, D., J. Knetsch, and R. Thaler (1991): "Anomalies: The endowment e¤ects, loss aversion, and status quo bias." Journal of Economic Perspectives, 5(1): 193-206. Kish, L. (1949): "A Procedure for Objective Respondent Selection within the Household." Journal of the American Statistical Association, 44 (247), pp. 380–87. Konow, J. (2010): "Mixed Feelings: Theories of and Evidence on Giving." Journal of Public Economics, 94 (3–4), pp. 279–97. Landry, C., A.. Andreas, J. List, M. Price, and N. Rupp (2006): "Toward an understanding of the economics of charity: evidence from a …eld experiment", The Quarterly Journal of Economics, 121 (2), pp. 747– 782. Lindert, K., A. Linder, J. Hobbs, and B. de la Brière (2007): "The Nuts and Bolts of Brazil’s Bolsa Familia Program: Implementing Conditional Cash Transfers in a Decen- tralized Context". Social Protection Discussion Paper No.709, World Bank, Washington, DC. List, J., and M. Price (2009): "The Role of Social Connections in Charitable Fundraising: Evidence from a Natural Field Experiment." Journal of Economic Behavior & Organiza- tion, 69 (2), pp. 160–69. 21 Miguel, E., S. Satyanath, and E. Sergenti (2004): "Economic Shocks and Civil Con‡ ict: An Instrumental Variables Approach." Journal of Political Economy, 112 (4), pp. 725–53. Ministry of Social Development of Chile (2015): Guia del cidadano: validacion y actu- alizacion de datos del hogar atraves de la pagina web del municipio, Chile: Govierno de Chile. Noussair, C., S. Robin, and B. Ru¢ eux (2004): "Revealing Consumers’ Willingness-to- Pay: A Comparison of the BDM Mechanism and the Vickrey Auction." Journal of Eco- nomic Psychology, 25 (6), pp. 725–41. Parra, J. (2011): "The E¤ects of Religion on Social Cooperation: Results from a Field Experiment in Ghana." Mimeo, Georgetown University, Washington, DC. Ringold, D., A. Holla, M. Koziol, and S. Srinivasan (2011): Citizens and Service Delivery: Assessing the Use of Social Accountability Approaches in Human Development Sectors, Directions in Development Series, Washington, DC: World Bank. Silva, J., V. Levin, and M. Morgandi (2013): Inclusion and Resilience: The Way Forward for Social Safety Nets in the Middle East and North Africa. MENA Development Report, Washington, DC: World Bank. Silverman, D., J. Slemrod, and N. Uler (2014): "Distinguishing the Role of Authority ’in’ and Authority ’ to.”’ Journal of Public Economics, 113, pp. 32-42. World Bank (2009): Conditional Cash Transfers: Reducing Present and Future Poverty, A. Fiszbein and N. Schady (Eds.), Washington, DC: World Bank. World Bank. (2015): The State of Social Safety Nets 2015, Washington DC: World Bank. 22 Table 1: Sample balance statistics Mean [s.d.] Difference Number of Control Treatment Full (C-T) obs. (C) (T) sample [p -value] (1) (2) (3) (4) (5) Panel A: Individual characteristics % male 0.42 0.49 0.45 0.20 420 [0.495] [0.501] [0.498] % with primary education 0.06 0.06 0.06 0.85 420 [0.242] [0.233] [0.236] % with secondary education 0.61 0.57 0.59 0.15 420 [0.488] [0.497] [0.492] % with tertiary education 0.32 0.38 0.35 0.17 420 [0.469] [0.486] [0.477] % young (18-25 year old) 0.23 0.20 0.21 0.34 420 [ 0.420] [0.397] [ 0.409] % young (18-29 year old) 0.31 0.25 0.28 0.11 420 [0.465] [0.435] [0.451] % young (18-34 year old) 0.41 0.35 0.38 0.16 420 [0.493] [0.478] [0.486] % currently employed 0.34 0.33 0.34 0.67 418 [0.476] [0.471] [0.473] Panel B: Household caracteristics Per capita expenditure (JD per 63.02 63.37 63.19 0.90 406 month) [44.67] [44.810] [44.682] % that has cars in the household 0.71 0.67 0.69 0.69 420 [0.691] [0.832] [0.763] Household size 6.07 6.11 6.09 0.84 416 [2.236] [2.464] [2.350] % with low ”subjective” income 0.18 0.23 0.21 0.15 412 [0.388] [0.421] [0.405] % with middle "subjective" income 0.783 0.751 0.77 0.38 412 [0.413] [0 .433] [0.423] % with high "subjective" income 0.03 0.02 0.03 0.48 412 [0.181] [0.139] [0.161] Panel C: Giving behaviour % that gave to charity in the last 0.65 0.58 0.62 0.14 413 3 months? [0.477] [0.495] [0.487] % that can rely on kins’ help if 0.16 0.19 0.17 0.37 412 needed? [0.366] [0.394] [0.380] Notes: Columns 1 and 2 report the mean and standard deviation (in square brackets) of each variable for the control and treatment groups. Column 3 reports the mean and standard deviation (in square brackets) of each variable for the full sample (i.e. control + treatment groups). Column 4 reports the p-value of the t-test of the difference between the control and treatment groups (using clustered wild bootstrap-t statistics at the PSU level). Column 5 shows the number of observations used. 23    Table 2: Average treatment effect on giving rates Control Treatment ATE Difference Full Number (C) (T) (C-T) Sample of obs. [p -value] (1) (2) (3) (4) (5) (6) Panel A: Aggregate/all proposals Mean giving 0.58 0.63 0.05 0.22 0.67 420 [0.396] [0.377] [0.387] Frequent giver 0.51 0.57 0.06 0.26 0.61 420 [0.495] [0.481] [0.488] Panel B: Individual proposals Unconditional cash transfer 0.60 0.68 0.08 0.11 0.69 420 [0.479] [0.443] [0.462] Unconditional food transfer 0.62 0.64 0.02 0.63 0.70 420 [0.465] [0.453] [0.458] Unconditional cash transfer 0.59 0.66 0.07 0.15 0.63 420 and school [0.493] [0.473] [0.483] Cash transfer conditional on 0.59 0.62 0.03 0.52 0.65 420 training [0.483] [0.471] [0.476] Notes: Panel A uses information from all the proposals, while panel B uses information from each proposal at a time. Each line reports the results of a regression on giving in that specific proposal controlling for gender, three education levels, location in the capital city, and number of cars in the household. Mean giving is computed at the participant level and is the share total proposals in which the participant indicated he would give up his voucher. Columns 1 and 2 report the mean and standard deviation (in square brackets) of each variable for the control and treatment groups. Column 3 reports the average treatment effect and column 4 reports the p-value of the t-test of the difference between the two samples (using clustered wild bootstrap-t at the PSU level). Column 5 reports the mean and standard deviation (in square brackets) of each variable for full sample. Column 6 shows the number of observations used. 24    Table 3: Average treatment effect on giving rates among low- and high-trust participants   Low trust (LT) High trust (HT) Difference (p -value) Control Treatment p -value Control Treatment p -value (C in LT (T in LT (C) (T) (C-T) (C) (T) (C-T) - - C in HT) T in HT) (1) (2) (3) (4) (5) (6) (7) (8) Panel A: Aggregate/all proposals Mean giving 0.48 0.57 0.08 0.66 0.70 0.45 0.02 0.15 [0.403] [0.371] [0.378] [0.367] Frequent giver 0.37 0.52 0.06 0.61 0.63 0.79 0.04 0.59 [0.502] [0.483] [0.479] [0.473] Panel B: Individual proposals Unconditional cash 0.44 0.60 0.04 0.75 0.77 0.63 0.01 0.21 transfer [0.497] [0.456] [0.439] [0.424] Unconditional food 0.56 0.63 0.16 0.66 0.67 0.79 0.44 0.91 transfer [0.473] [0.444] [0.455] [0.444] Unconditional cash 0.39 0.49 0.18 0.62 0.67 0.35 0.06 0.10 transfer and school [0.501] [0.489] [0.472] [0.444] Cash transfer 0.51 0.56 0.44 0.63 0.67 0.48 0.46 0.27 conditional on training [0.489] [0.479] [0.476] [0.456] Notes: Columns 1 and 2 report the mean and standard deviation (in square brackets) of each variable for the control and treatment group among low trust individuals. Column 3 reports the p-value of the t-test of the difference between the two samples among low trust individuals (using clustered wild bootstrap-t at the PSU level). Columns 4 and 5 report the mean and standard deviation (in square brackets) of each variable for the control and treatment group for high trust individuals. Column 6 reports the average treatment effect and column [8] reports the p-value of the t-test of the difference between the two samples among high trust individuals (using clustered wild bootstrap-t at the PSU level). Column 7 reports the p-value of the t-test of the difference between giving among low- and high-trust participants in the control group. Column 8 reports the p-value of t-tests of the difference between giving among low- and high- trust participants in the treatment group. All regressions include controls for gender, three education levels, location in the capital city, and number of cars in the household. Observations are 194 in the low trust group and 217 individuals in the high-trust group for a total of 411 participants (due to 9 missing responses to the trust question). 25    Table 4: Effect of the interaction between treatment and trust on giving for youth and adults All Young (18-29) Adults (30+) (1) (2) (3) (4) (A) Aggregate/all proposals Mean giving Treatment 0.08 0.07 0.19 0.02 [0.054] [0.050] [0.093]** [0.050] Treatment*Trust -0.05 -0.05 -0.10 -0.03 [0.020]** [0.020]** [0.047]** [0.031] Trust 0.11 0.11 0.11 0.10 [0.038]*** [0.039]*** [0.063] [0.042]** Young (18-29) -0.10 [0.047]** Frequent giver Treatment 0.13 0.11 0.21 0.07 [0.077] [0.074] [0.132] [0.082] Treatment*Trust -0.12 -0.11 -0.20 -0.08 [0.043]*** [0.042]*** [0.085]** [0.050]* Trust 0.14 0.14 0.16 0.13 [0.050]*** [0.051]*** [0.094] [0.061]** Young (18-29) -0.15 [0.063]** (B) Individual proposals Unconditional cash Treatment 0.14 0.13 0.25 0.07 transfer [0.075]* [0.075]* [0.134]* [0.085] Treatment*Trust -0.12 -0.11 -0.16 -0.11 [0.038]*** [0.037]*** [0.071]** [0.050]** Trust 0.18 0.18 0.17 0.18 [0.000]*** [0.000]*** [0.088]* [0.058]*** Young (18-29) -0.13 [0.047]*** Unconditional food Treatment 0.07 0.06 0.21 0.00 transfer [0.050] [0.048] [0.000]*** [0.042] Treatment*Trust -0.05 -0.05 -0.14 -0.02 [0.027]* [0.028]* [0.057]** [0.076] Trust 0.05 0.05 0.06 0.03 [0.036] [0.037] [0.053] [0.068] Young (18-29) -0.03 [0.050] Unconditional cash Treatment 0.09 0.08 0.13 0.05 transfer and school [0.072] [0.068] [0.141] [0.076] Treatment*Trust -0.04 -0.03 -0.03 -0.03 [0.027] [0.027] [0.038] [0.039] Trust 0.15 0.15 0.08 0.17 [0.063]** [0.063]** [0.097] [0.078]** Young (18-29) -0.12 [0.060]** Cash transfer Treatment 0.03 0.02 0.19 -0.05 conditional on [0.064] [0.058] [0.170] [0.090] training Treatment*Trust 0.01 0.01 -0.09 0.06 [0.020] [0.047] [0.054] [0.056] Trust 0.05 0.06 0.11 0.02 [0.055] [0.059] [0.084] [3.495] Young (18-29) -0.12 [0.056]** Number of observations 411 411 116 295 Notes: The estimation method is a linear probability model. All regressions include controls for gender, three education levels, location in the capital city, and number of cars in the household. The dependent variable is the giving rate. Standard errors clustered at the PSU level using wild bootstrap-t are reported in brackets. Trust is a dummy variable equal to one if the answer to the question “How confident are you that the public funds allocated for social assistance reach the poor?” is “completely confident” or “somewhat confident”, and zero otherwise. ***, **, * significant at 1%, 5%, and 10% level. 26    Table 5: Distribution of the preferred proposal, by treatment status Control Treatment ATE Difference (C) (T) (C-T) [p -value] [1] [2] [3] [4] Prefered proposal Unconditional cash transfer 0.21 0.24 0.03 0.56 [0.409] [0.429] Unconditional food transfer 0.34 0.21 -0.12 0.07 [0.473] [0.410] Unconditional cash transfer and school 0.13 0.20 0.07 0.18 [0.337] [0.397] Cash transfer conditional on training 0.32 0.35 0.03 0.62 [0.469] [0.478] Total 1 1 Number of observations 161 174 Notes: Columns 1 and 2 report the mean and standard deviation (in square brackets) of each variable for the control and treatment groups. Column 3 reports the average treatment effect and column 4 reports the p-value based on wild bootstrap-t cluster-robust standard errors. Results based on reported preferred proposal among those actually chosen by participants. Results control for gender, three education levels, location in the capital city, and number of cars in the household. For individuals who only decided to give up their voucher once, the preferred proposal is assumed to be the delivery method actually chosen. 27    Table 6: Effect of the interaction between treatment and education and between treatment and income A) Aggregate/all proposals B) Individual proposals Mean Frequent Uncond. Uncond. Uncond. cash Cash transfer giving giver cash food transfer and conditional on transfer transfer school training (1) (2) (3) (4) (5) (6) Panel A: Interaction with education Treatment 0.02 0.01 0.04 0.00 0.04 0.01 [0.048] [0.051] [0.061] [0.017] [0.059] [0.055] Treatment*Skilled 0.09 0.14 0.12 0.06 0.11 0.06 [0.093] [0.101] [0.129] [0.083] [0.151] [0.147] Skilled 0.00 -0.04 -0.07 0.03 -0.02 0.06 [0.003] [0.091] [0.094] [0.054] [0.190] [0.065] Number of observations 420 420 420 420 420 420 Panel B: Interaction with income Treatment 0.06 0.04 0.08 0.07 0.05 0.02 [0.041] [0.055] [0.058] [0.050] [0.050] [0.052] Treatment*High income 0.02 0.09 0.03 -0.10 0.08 0.05 [0.045] [0.164] [0.070] [0.063] [0.228] [0.124] High income 0.04 0.04 0.03 0.10 0.01 0.03 [0.033] [0.046] [0.023] [0.062] [0.019] [0.052] Number of observations 406 406 406 406 406 406 Panel C: Interaction with " subjective " income Treatment 0.04 0.03 0.05 0.00 0.08 0.01 [0.050] [0.058] [0.051] [0.027] [0.057] [0.036] Treatment*High 0.11 0.13 0.17 0.11 0.04 0.13 "subjective" income [0.144] [0.159] [0.101]* [0.116] [0.080] [0.165] -0.04 -0.07 -0.06 -0.05 -0.02 -0.04 High "subjective" income [0.219] [0.120] [0.194] [0.103] [0.048] [0.146] Number of observations 404 404 404 404 404 404 Notes: The estimation method is a linear probability model. Each column reports regression on participants’ decisions on all (columns 1 and 2) or specific proposal (columns 3-6) controlling for gender, location in the capital city, and number of cars in the household (Panel A), and for gender, three education levels, location in the capital city and number of cars in the household (Panels B and C). Results are maintained if excluding education levels and number of cars in the household as control variables. Standard deviation reported below (in square brackets). Standard errors clustered at the PSU level using wild bootstrap-t are reported in brackets. Skilled is a dummy variable equal to one if the participant has completed high school or more. High income is a dummy variable equal to one if the participant lives in a household with an income per capita level above the sample mean. High "subjective" income is a dummy variable equal to one if the participant reports that his relative position on an income scale from one (lowest) to four (highest) is three or four. ***, **, * significant at 1%, 5%, and 10% level. 28    Table 7: Robustness of the interaction between treatment and trust to the inclusion of distance from the poor Measure of distance from the Education Per capita income " Subjective " income poor used: All Skilled Unskilled All Above mean Below mean All High Low (1) (2) (3) (4) (5) (6) (7) (8) (9) Mean Treatment 0.08 0.10 0.07 0.10 0.13 0.09 0.09 0.37 0.05 giving [0.055] [0.079] [0.073] [0.057]* [0.093] [0.067] [0.055]* [0.177]** [0.058] Treatment*Trust -0.05 0.03 -0.07 -0.07 -0.12 -0.06 -0.06 -0.27 -0.03 [0.021]** [0.049] [0.054] [0.025]*** [0.048]** [0.033]* [0.022]*** [0.088]*** [0.023] Distance from the 0.04 0.05 -0.00 poor [0.044] [0.034] [0.031] Frequent Treatment 0.12 0.14 0.11 0.15 0.23 0.11 0.13 0.48 0.09 giver [0.076] [0.099] [0.102] [0.082]* [0.133]* [0.090] [0.077]* [0.244]* [0.079] Treatment*Trust -0.11 0.03 -0.16 -0.15 -0.20 -0.14 -0.12 -0.35 -0.11 [0.042]*** [0.030] [0.088]* [0.049]*** [0.065]*** [0.064]** [0.046]*** [0.112]*** [0.043]** Distance from the 0.03 0.08 -0.02 poor [0.062] [0.040]** [0.210] Uncond. Treatment 0.14 0.22 0.09 0.16 0.21 0.14 0.15 0.49 0.10 cash [0.077]* [0.084]** [0.098] [0.078]** [0.112]* [0.102] [0.077]* [0.227]** [0.078] transfer Treatment* Trust -0.12 -0.13 -0.09 -0.14 -0.19 -0.12 -0.13 -0.37 -0.09 [0.038]*** [0.049]** [0.063] [0.045]*** [0.062]*** [0.054]** [0.041]*** [0.118]*** [0.050]* Distance from the -0.00 0.06 -0.00 poor [0.012] [0.032]* [0.009] Uncond. Treatment 0.06 0.02 0.08 0.08 0.06 0.11 0.07 0.36 0.02 food [0.049] [0.083] [0.052] [0.050]* [0.094] [0.073] [0.049] [0.197]* [0.059] transfer Treatment*Trust -0.05 0.11 -0.11 -0.08 -0.15 -0.06 -0.06 -0.29 -0.02 [0.027]* [0.180] [0.048]** [0.028]*** [0.094] [0.023]** [0.027]** [0.092]*** [0.039] Distance from the 0.05 0.04 -0.01 poor [0.053] [0.056] [0.231] Uncond. Treatment 0.09 0.08 0.09 0.11 0.15 0.10 0.11 0.31 0.09 cash [0.072] [0.124] [0.105] [0.079] [0.141] [0.090] [0.076] [0.233] [0.084] transfer Treatment*Trust -0.03 0.10 -0.09 -0.06 -0.05 -0.08 -0.06 -0.24 -0.03 [0.026] [1.772] [0.076] [0.033]* [0.043] [0.060] [0.033]* [0.101]** [0.031] and Distance from the 0.04 0.05 -0.02 school poor [0.051] [0.037] [0.129] Cash Treatment 0.03 0.06 0.01 0.04 0.12 0.00 0.04 0.31 0.00 transfer [0.067] [0.084] [0.056] [0.070] [0.107] [0.029] [0.067] [0.225] [0.003] cond. on Treatment*Trust 0.01 0.05 0.00 -0.00 -0.08 0.03 -0.00 -0.19 0.02 [0.030] [0.268] [0.054] [0.004] [0.047] [0.259] [0.001] [0.104]* [0.126] training Distance from the 0.09 0.06 0.01 poor [0.050]* [0.040] [0.035] Number of observations 411 146 265 399 150 249 404 100 304   Notes: The estimation method is a linear probability model. All regressions include controls for trust, gender, location in capital city, and number of cars in household. Standard errors clustered at PSU level using wild bootstrap-t reported in brackets.***,**,* significant at 1%, 5% and 10% level. 29    Table 8: Robustness of the interaction between treatment and trust to the inclusion of social norms on redistribution Mea sure o f "redi stri buti o n va l ues " People are poor Success is a Society should Composite because of bad matter of hard make incomes index luck or insjustice work more equal (not laziness) (1) (2) (3) (4) (A) Aggregate/all proposals Mean giving Treatment 0.12 0.09 0.08 0.11 [0.065]* [0.055] [0.056] [0.066]* Treatment*Trust -0.10 -0.06 -0.03 -0.09 [0.033]*** [0.022]** [0.017]* [0.036]** Redistribution values 0.05 0.11 -0.02 -0.03 [0.044] [0.037]*** [0.039] [0.050] Frequent giver Treatment 0.18 0.13 0.13 0.17 [0.091]* [0.077]* [0.078] [0.097]* Treatment*Trust -0.18 -0.12 -0.10 -0.17 [0.060]*** [0.046]*** [0.038]*** [0.057]*** Values redistribution 0.08 0.12 -0.07 -0.04 [0.064] [0.049]** [0.058] [0.099] (B) Individual proposals Unconditional Treatment 0.17 0.14 0.14 0.16 cash transfer [0.089]* [0.076]* [0.080]* [0.089]* Treatment*Trust -0.17 -0.12 -0.10 -0.16 [0.054]*** [0.041]*** [0.034]*** [0.051]*** Redistribution values 0.07 0.10 -0.06 -0.03 [0.045] [0.041]** [0.044] [0.048] Unconditional Treatment 0.10 0.07 0.06 0.09 food transfer [0.057]* [0.049] [0.050] [0.056] Treatment*Trust -0.10 -0.06 -0.04 -0.09 [0.037]*** [0.028]** [0.025] [0.034]*** Redistribution values 0.05 0.12 -0.06 -0.05 [0.055] [0.050]** [0.053] [0.056] Unconditional Treatment 0.15 0.10 0.09 0.14 cash transfer and [0.084]* [0.076] [0.071] [0.092] Treatment*Trust -0.11 -0.04 -0.01 -0.10 school [0.040]*** [0.029] [0.013] [0.043]** Redistribution values 0.08 0.10 0.04 -0.01 [0.051] [0.049]** [0.053] [0.196] Cash transfer Treatment 0.06 0.04 0.03 0.06 conditional on [0.083] [0.068] [0.068] [0.089] Treatment*Trust -0.03 0.00 0.02 -0.02 training [0.030] [0.004] [0.201] [0.027] Redistribution values 0.01 0.10 -0.01 -0.03 [0.038] [0.044]** [0.056] [0.061] Number of observations 346 405 407 340 Notes: The estimation method is a linear probability model. All regressions include controls for trust, gender, three education levels, location in the capital city, and number of cars in the household. The dependent variable is the giving rate. Results in column 1 control for believing that people are poor because of bad luck or injustice rather than laziness. Results in column 2 control for believing that success is a matter of hard work rather than luck or connections. Results in column 3 control for believing that society should redistribute income. Results in column 4 control for the composite index made of all three values constructed via polychoric principal-components model. Standard errors clustered at the PSU level using wild bootstrap-t are reported in brackets. ***, **, * significant at 1%, 5%, and 10% level. 30    Table 9: Robustness of interplay between treatment, trust and age to a different definition of youth All Young (18-34) Adults (35+) (1) (2) (3) (A) Aggregate/all proposals Mean giving Treatment 0.07 0.22 -0.01 [0.051] [0.088]** [0.141] Treatment*Trust -0.04 -0.19 0.03 [0.018]** [0.067]*** [0.054] Trust 0.11 0.17 0.08 [0.039]*** [0.060]*** [0.064] Young (18-34) -0.12 [0.045]** Frequent giver Treatment 0.11 0.28 0.02 [0.073] [0.138]** [0.063] Treatment*Trust -0.10 -0.29 -0.01 [0.038]*** [0.103]*** [0.034] Trust 0.14 0.21 0.11 [0.051]*** [0.069]*** [0.078] Young (18-34) -0.16 [0.064]** (B) Individual proposals Unconditional cash Treatment 0.12 0.31 0.02 [0.076] [0.136]** [0.069] Treatment*Trust -0.11 -0.29 -0.02 [0.037]*** [0.102]*** [0.038] Trust 0.18 0.29 0.12 [0.000]*** [0.000]*** [0.069]* Young (18-34) -0.13 [0.042]*** Unconditional food Treatment 0.06 0.23 -0.03 transfer [0.049] [0.076]*** [0.071] Treatment*Trust -0.05 -0.22 0.03 [0.027]* [0.071]*** [0.045] Trust 0.05 0.12 0.01 [0.039] [0.062]* [0.043] Young (18-34) -0.06 [0.051] Unconditional cash Treatment 0.07 0.14 0.03 transfer and school [0.067] [0.108] [0.066] Treatment*Trust -0.02 -0.04 -0.03 [0.022] [0.036] [0.051] Trust 0.16 0.11 0.19 [0.064]** [0.087] [0.092]** Young (18-34) -0.14 [0.051]*** Cash transfer Treatment 0.02 0.21 -0.08 conditional on [0.050] [0.147] [0.101] training Treatment*Trust 0.02 -0.22 0.14 [0.451] [0.086]** [0.069]** Trust 0.06 0.16 -0.01 [0.063] [0.086]* [0.017] Young (18-34) -0.14 [0.056]** Number of observations 411 157 254 Notes: The estimation method is a linear probability model. All regressions include controls for gender, three education levels, location in the capital city, and number of cars in the household. The dependent variable is the giving rate. Standard errors clustered at the PSU level using wild bootstrap-t are reported in brackets. ***, **, *significant at 1%, 5%, and 10% level. 31    Table A.1: Summary statistics Mean s.d. Number of [s.d.] observations (1) (2) (3) Mean giving 0.67 0.39 420 % frequent giver 0.61 0.49 420 % ”accepted” unconditional cash transfer (UCT) 0.69 0.46 420 % ”accepted” unconditional food transfer (UFT) 0.70 0.46 420 % ”accepted” unconditional cash transfer and school (UCT+School) 0.63 0.48 420 % ”accepted” cash transfer conditional on training (CCT) 0.65 0.48 420 % high trust 0.53 0.50 411 % young (18-25 year old) 0.21 0.41 420 % young (18-29 year old) 0.28 0.45 420 % young (18-34 year old) 0.38 0.49 420 % skilled 0.35 0.48 420 Per capita expenditure (JD per month) 63.19 44.68 406 % high income 0.37 0.48 406 % that self-identified as high income 0.25 0.43 412 % that reported prefered proposal UCT 0.23 0.42 414 % that reported prefered proposal UFT 0.28 0.45 414 % that reported prefered proposal UCT+School 0.16 0.37 414 % that reported prefered proposal CCT 0.33 0.47 414 People are poor because of bad luck or insjustice (not laziness) 0.60 0.49 350 Success is a matter of hard work 0.63 0.48 408 Agrees that society should make incomes more equal 0.80 0.40 410 32    Table A.2: Average treatment effect on giving rates without control variables Mean [s.d.] Difference Number of Control Treatment Full (C-T) obs. Sample (p -value) (1) (2) (3) (4) (5) Panel A: Aggregate/all proposals Mean giving 0.65 0.71 0.67 0.19 420 [0.395] [0.377] [0.387] Frequent giver 0.58 0.64 0.61 0.26 420 [0.495] [0.481] [0.488] Panel B: Individual proposals Unconditional cash transfer 0.65 0.73 0.69 0.11 420 [0.478] [0.443] [0.462] Unconditional food transfer 0.69 0.72 0.70 0.55 420 [0.465] [0.452] [0.458] Unconditional cash transfer and school 0.59 0.67 0.63 0.13 420 [0.492] [0.472] [0.483] Cash transfer conditional on training 0.63 0.67 0.65 0.47 420 [0.483] [0.476] [0.476] Notes: Panel A uses information from all the proposals, while panel B uses information from each proposal at a time. Each line reports regression on participants’ decisions on that specific proposal without any controls. Mean giving is computed at the participant level and is the share total proposals in which the participant indicated he would give up his voucher. Columns 1, 2 and 3 report the mean and standard deviation (in square brackets) of each variable for the control groups, treatment groups and full sample. Column 4 reports the p-value of the t-test of the difference between the two samples (using clustered wild bootstrap-t at the PSU level). Column 5 shows the number of observations used. 33    Figure 1: Average Treatment Effect on Giving Rates, by Age Group Notes: Linear effect of participating in treatment group on giving decisions in age group subsamples. P-values reported are based on wild-t bootstrap standard errors, N=420. “Freq. giver” means gave up the voucher in more than 2 proposals. Adults in the sample range up to 80 years old. Figure 2: Treatment Impact, by Age and Trust Level Panel A: Aggregate proposals 34   Panel B: Individual proposals Notes: Lines produced with local polynomial smooth function on mediating variable age. Mean giving is the number of times a participant decided to donate his voucher out of 4. Vertical line represents the median of age in each sub-group. At the bottom of the graph the histogram of age is plotted. Bandwidth of 0.8. Figure 3: Treatment Impact, by Age Group and Trust Level Note: Figure shows the average treatment effect (difference between treatment and control groups’ donation rates) for the nested subsamples of youth and adult with either high or low trust in the delivery of social safety nets. For the formal econometric results of the interaction between trust level and treatment for different age groups, see table 4. 35