Social Welfare Programs, Stigma, and Trust: Experimental Evidence for Six Latin American Cities

Alberto Chong and Vanessa Rios1

Abstract

We argue that social welfare programs are linked with the destruction of social capital, as measured by interpersonal trust in laboratory games. To test for this, we employ experimental data for representative samples of individuals in six Latin American capital cities (Bogota, Buenos Aires, Caracas, Lima, Montevideo, and San Jose). In fact, we find that participation in welfare programs damages trust, a result that is robust to the inclusion of individual risk measures and a broad array of controls. Whereas we argue that endogeneity is not a critical issue, we also control for it using an instrumental variables approach. Our findings also support the notion that low take up rates may be due to stigma linked to trust and social capital, rather than to transaction costs.

JEL Classification Code: D01, O12, O10Key Words: Experiments, Surveys, Social Programs, Social Capital, Trust, Stigma, Latin AmericaWord Count: 10,452

1 Chong, Georgia State University, Rios: University of Wisconsin, Madison. We are grateful to Juan Camilo Cardenas, Hugo Nopo, Suzanne Duryea, Jorge Guillen. for comments and suggestions. An earlier draft circulated under the title “Do Welfare Programs Damage Interpersonal Trust? Experimental Evidence from Representative Samples for Four Latin American Cities”. All remaining errors are our own.

1

1. Introduction

Social welfare programs, in particular, those designed and implemented by governments

to help the neediest cope with long term vulnerabilities play a critical role as a policy instrument

in developed and developing countries alike. In fact, the proliferation of these long-term need-

based programs that include nutritional assistance, health assistance, educational support, and

even outright cash transfers have been actively pursued under the premise that they can help

provide a smooth transition towards reaching long-term improved welfare to the neediest. In fact,

it may be difficult to argue against these strategies, in particular, in the case of Latin America

where social welfare programs have become rather ubiquitous, but appear to have delivered on

the promise of improving the welfare of the neediest (Fiszbein and Schady, 2009). In fact, the

abundance of welfare programs in the developing world has spurred a coterie of academic papers

that seek to evaluate their impact and effectiveness, which in turn have generated ever improving

waves of such welfare programs aimed at better targeting the most disadvantaged.

Interestingly, whereas social welfare programs have become widely accepted as a tool to

effectively reach and positively impact the neediest, there is an on going debate on whether or

not they may also have undesirable effects on the society. In particular, several sociologists and

psychologists have argued that stigma in social programs may be linked with disapproval or

discontent with oneself and the society-at-large, as the very same characteristics that make them

eligible to such programs may also distinguish them, which may have negative connotations in

the society. After all, stigma is defined as a sign of disgrace or discredit, which sets a person

apart from others as a marker of adverse experiences reflected in a deep sense of shame, and thus

secrecy and withdrawal. Individuals who try to pursue a secrecy strategy and withdraw have a

more insular support network (Byrne, 2000). Furthermore, family and friends may also endure

2

stigma by association, the so-called “courtesy stigma”. In fact, negative cultural sanction and

myths may combine to ensure scapegoating in the wider community, which translates in an

increase in social distance of those stigmatized by social programs with the rest of the

community (Goffman, 1963; Byrne, 2000).

In the context above, this paper provides empirical evidence on the possible link between

participation in social welfare programs and social capital, as measured by one of its key

components, interpersonal trust. Unlike typical studies on the latter, we do not employ

perception data, but offer representative new data at the city level from experimental games

collected from the following six Latin American capital cities, Bogota, Colombia; Buenos Aires,

Argentina; Caracas, Venezuela; Lima, Peru; Montevideo, Uruguay, and San Jose, Costa Rica. In

particular, we apply a standard, well-known, trust games with a “tried and true” protocol and

methodology that have been widely applied in the literature as a proxy for interpersonal trust

(Burks et al., 2003; Berg et al., 2005; Levitt and List, 2007). In addition, we also collected

representative information at the city level on the type and extent of social welfare program

participation by individuals, along with their corresponding basic socio-economic information.

We ask individuals about participation on mean-tested programs in education, health, nutrition,

and child care specific programs.

It is important to mention that since our focus is on welfare programs that may elicit

stigma, for instance, the use of distinguishable food stamps in stores or similar public locations,

we exclude income maintenance programs where in-cash or in-kind transfers are typically done

either using electronic tools, or related indistinguishable means, which tends to common in Latin

America. This is the case of program delivery of individual unemployment savings plans,

3

severance pay, and conditional cash transfers, which are common in many countries in the

region2.

We find that being in a social welfare program is negatively linked to trusting behavior

and thus, to social capital formation. Furthermore we also test whether this association is causal

by applying a two-stage instrumental variables approach to provide additional evidence that the

uncovered link appears to go causally, from social welfare program participation to a reduction

of interpersonal trust, and thus to social capital destruction. Overall, we believe that our findings

make a strong case on the link between welfare programs and interpersonal trust as we employ

representative samples at the city level, and to the extent that we are not only making emphasis

on methodological and experimental protocols, but also on sampling design and

representativeness, we increase the external validity of our results3.

The paper is organized as follows. The next section provides a brief review of the

literature. Section 3 describes the data collection process, including sampling experimental

design. Section 4 describes the empirical methodology applied. Section 5 provides empirical

evidence on the possible link between participation in social programs and interpersonal trust,

including an identification strategy that employs an instrumental variables approach. Finally,

section 6 summarizes and concludes.

2. Brief Review of the Literature

In spite of the fact that the issue of possible negative externalities related to social welfare

programs due to stigma and discrimination can be very relevant for policy design, it has been

largely overlooked by economists and policymakers alike. However, there is relevant related 2 For instance, in the case of the Dominican Republic funds for most social program transfers are delivered electronically to individual card holders who can check balances and make transactions anonymously or with little interaction with other members of the society (Busso and Galiani, 2014).3 Our experimental set up focuses on interpersonal trust but it does not allow us to study whether being in a welfare program is associated with trustworthiness, an issue that escapes our research scope, which we refer to in our conclusions.

4

research in other fields, including social psychology, sociology, and the medical sciences, where

researchers have studied the issue of stigma and its impact in terms of social interactions (e.g.,

Crocker, Major, and Steele, 1998; Watson and Corrigan, 2003). Furthermore, some researchers

have distinguished between public stigma, defined as ways in which the general public reacts to

stigma attached to a group, and self-stigma, defined as the reactions that individuals turn against

themselves because they are members of a stigmatized group (Augoustinos and Ahrens, 1994;

Judd and Park, 1993; and Krueger, 1996). Researchers have also identified several behaviors

associated with public stigma, all leading to discrimination, prejudice, and loss of trust and social

capital. According to this literature, stigmatized individuals are considered responsible for their

handicaps, as these are seen as arising from problems created by themselves. In this context, help

is withheld and avoidance follows as people choose not to interact with a stigmatized person that

is believed to be responsible for her lot in life (Watson and Corrigan, 2003). Furthermore, these

individuals are viewed as incompetent and thus requiring authority figures to care for them. This

also brings prejudice as well as discrimination (Hilton and von Hippel, 1996; Devine, 1995;

Crocker, Major and Steele, 1998). Thus, isolation and difficulties in social interactions follow

leading to a loss in trust and social capital (Stuber, et al., 2000; Watson and Corrigan, 2003).

Individuals from stigmatized groups tend to be aware of the beliefs about their groups so much,

that a sizeable fraction of them interiorize the stigma and apply it against themselves. As a result

of this sort of self-stigmatization, they end up suffering diminished self-esteem and self efficacy,

and isolation (Bowden, Schoenfield, and Adams, 1980; Kahn, Obstfeld, and Heiman, 1979). As

in the case of public stigma, people perceiving or feeling stigmatization are less likely to interact

with other individuals thus reinforcing the negative feelings of the stigmatized, to the point that

sentiments of marginalization, hatred, resentment, and lack of trust become predominant.

5

Paradoxically, entitlement programs that are designed to help excluded groups may end up

further marginalizing them as the social construct or social capital of the society becomes

structurally damaged (Gronfein and Owens, 2000).

According to this literature, the existence of networks of civil engagement and basic

norms of reciprocity are the most important prerequisites to accumulate social capital within a

society (Kumlin and Rothstein, 2005). Personal welfare state experiences tend to be generalized

into collective judgments of what population is experiencing as a global, thus influencing

political orientation and attitude formation and also affecting trust in politics, institutions, and

even trust among peers (Kumlim, 2002; Kumlin and Rothstein, 2005; Rothstein, 2000). The link

between personal experiences and trust is crucial as it determines the attitude for a voluntary

civil engagement and reciprocity, thus affecting the accumulation of social capital. In this sense,

when there is a general perception of trust and fairness in political and legal institutions, such pro

social behavior that enhances social capital is more likely to develop; while, a contrary context,

might influence citizens’ views on the reliability of other people as well, reducing interpersonal

trust to the detriment of the production of social capital. In other words, the formulation and

implementation of welfare programs implies a great scope for bureaucratic discretion and gives

rise to suspicions of arbitrary and discriminatory treatment in comparison with universal welfare

programs (Kumlin and Rothstein, 2005). These programs may also incite resentment on the

portion of the population who is not qualified for the benefits; being more exposed to political

resistance when non qualifiers are numerous and politically effective (Rothstein, 1998). On the

contrary, universal programs do promote equal treatment and principles of fairness, generating

experiences that build support for the welfare state and the political system. In fact, Kumlin and

Rothstein (2005) provide empirical evidence to sustain that personal experience that employ

6

means-tested welfare state tools seem to reduce interpersonal trust while experiences with

universal institutions tend to build it. In particular, they empirically test whether specific design

matters for the production of social capital in the case of Sweden. Their argument is that welfare

state institutions are not necessarily unfavorable to the production of social capital, being the

specific design of the welfare state policies a key issue on such production of social capital.

Thus, they find that contacts with universal welfare state institutions tend to increase social trust,

whereas experiences with means-tested welfare programs undermine it, suggesting that by

designing welfare-state institutions, governments can invest in social capital.

2. Data and Experimental Design4

As mentioned above, one of the advantages in our study is that we employ objective

measures of interpersonal trust based on experimental games. By using these data, instead of

subjective data, we avoid potential biases widely discussed in the literature (e.g., Bertrand and

Mullainathan, 2000)5. Our full sample consists of more than 3,000 individuals (half of them

assume the role of Player 1 and the other half the role of Player 2) and include samples collected

in Bogota, Buenos Aires, Caracas, Lima, Montevideo and San Jose. Not only are these data the

most comprehensive experimental data in Latin America, but they are also particularly unique

since the samples are specifically collected to be representative at the city level as the goal of the

sampling procedure was to obtain empirical distributions of individuals within combinations of

4 This section draws heavily from Candelo, et al. (2009) and Cardenas et al., (2012) as these papers are also part of a broader research initiative on trust and pro-social behaviors funded by the Research Department of the the Inter-American Development Bank, a multilateral institution based in Washington, DC. The paper by Candelo, et al., (2009) gives details related of the methodological approach of the overall field work and experimental protocols employed in both Cardenas, et al (2012) and this research. The research questions pursued in Cardenas (2012) and this piece are distinct and independent from each other. In particular, in the former paper the authors create a pro-social index using survey data and compare it with pro-social experimental data to show that both sources of information are complementary.5 Furthermore, our experimental approach allows us to measure interpersonal trust in an indirect way, which may help reduce reverse causality issues. Endogeneity issues are addressed in the next section.

7

characteristics resembling those of the populations in the cities (Candelo, et al., 2009). In

particular, we employed stratified random samples where strata were chosen on the basis of

education, average family income of the districts, gender, age, and the territorial units that make

up each city under study –in either quartiles or quintiles, depending on data availability.6

In order to guarantee homogeneity in the application of experimental protocols, the field

teams participated in a training workshop at the launching of this project7. This workshop

provided a uniform approach to implementation and related fieldwork details such as sampling

procedures, writing style, protocol, timing of actions (i.e., invitations, pre-survey, experiments,

post-surveys), elements to be included in experimental sessions and the construction of

questionnaires8. The individuals participating in the experiments were recruited randomly from

neighborhoods and invited a few days prior to the application of the experimental session in

order to gather information regarding their socio-economic background and social welfare

participation characteristics and to receive information about the expected gains from

participating in the experiments, which included a show-up fee and potential monetary gains

from participating in the games. The day before the experimental sessions, the participants were

reminded of it with a phone call or visit, and transportation was agreed or discussed. Information

on the socio-economic composition of the groups in each particular session was delivered as the

sessions progressed. The participants met in one room where they were able to see each other,

but they were not allowed to communicate during the session. During the recruitment process we

avoided having two people who knew each other within one session. As the sessions progressed,

participants received information about their peers, depending of the particular activity. Social

6 The age groups were: (i) 17-27; (ii) 28-38; (iii) 39-59 and (iv) 60-72.7 The launching workshop for this project was conducted in Bogota, Colombia during the fourth quarter of 2007.8 All experiments and related questionnaires were conducted in Spanish. A translation of the questionnaire and protocols is available from the authors upon request.

8

heterogeneity on individuals’ decisions in each particular session was made as salient and clear

as possible using the information collected on the socio-economic composition of the groups.

In order to carry out the field work we conducted a series of approximately 25

experimental sessions per city. The sessions were arranged so that at least three sessions per city

included only individuals from high-income strata and at least three other sessions included only

individuals from low-income strata; the rest combined individuals from all strata. Around 30

individuals were invited for each session, under the assumption that approximately one third

would not show up, thus allowing each experimental session to go forward with roughly 20 to 25

participants that lasted between two and three hours (Candelo, et al., 2009). Each experimental

session followed the exact same protocol, with the exact same sequence of activities, as a team of

researchers with experience in survey and field methods was selected to undertake the sample

design and conduct the experiments and surveys in each city. After participants completed the

surveys, which were designed to collect information on whether individuals participated in

education, health nutrition, or child care welfare programs, the payoffs from the experiments

were computed and the participants received their payments.

The sessions consisted of several activities each, all played in the same order. For this

paper we focus on the results of the first activity, a straightforward Trust Game, which was

applied using the strategy method (Berg, et al, 2003; Carpenter, et al., 2005). As it is well

known, session participants in this game are randomly assigned in pairs: half assume the role of

player 1 and the other half, that of player 2. The assignment of pairs was randomly made. Both

groups are simultaneously located in different rooms, and identities of the pairs are never

revealed, although each player receives information on key demographic characteristics of their

pairs: sex, age, schooling level and socio-economic stratum. Other socio-demographic variables,

9

such as if the person was or not in a welfare program, were not included as having done so may

have affected expectations about what the experiment would be, and also may have affected both

the behavior of the players, which could have compromised our findings. In this game, both

players receive an equal endowment, and player 1 is then asked to decide how much of this

endowment he or she wants to send to player 2, knowing that player 2 will then receive three

times that amount on top of the initial endowment everyone initially receives. In another room,

player 2 is asked to decide the amount to be returned to player 1 for each possible offer from

player 1, from a discrete set of fractions of amounts sent (0%, 25%, 50%, 75% and 100%).

Immediately before making their decisions, the individuals are also asked to predict the decisions

to be made by the other player. That is, the amount expected by player 2 from player 1, and

player 1’s expected returned amount from player 2. After both players make their decisions the

matching of their choices is made. This procedure allows us to separate the effects of

expectations from those of social background. Replications of this game around the world have

shown that people on average send half of the initial endowment to player 2, and that the returns

from player 2 to player 1 generate a net positive return for player 1 of about ten to twenty percent

of what was originally sent (Carpenter et al, 2005; Ashraf, Camerer and Loewenstein, 2005).

Provided that individuals’ attitudes towards risk may be a crucial determinant of a

player’s offering in this game, in this paper we also use information from a related experimental

activity based on the simple risk games first applied by Binswanger (1980) and later on by Barr

(2003). In this activity (which was the third activity of the experimental sessions) each player

makes individual decisions over three games that measure individual attitudes over risk,

ambiguity, and losses. In this paper we only use information that measures attitudes towards

10

risk9. Such measure comes from a game were participants are given a set of outcomes for six

50/50 lotteries that go from a sure low payoff to an all-or-nothing higher expected payoff. The

lotteries in between gradually increase both in expected value and in the spread between the low

and high payoff. Based on the outcomes of this game, we classify participants into three groups:

low risk averse, medium risk averse and high risk-averse. We use this resulting variable as a

control in our empirical specifications.10 For the purpose of this study, the sample only includes

those players who made a initial offer, which consists of 1,465 observations.

3. Econometric Approach

Following the standard interpretation of the trust game, we use the offer of the first player

to the second player, the measure of interpersonal trust, as our dependent variable. As mentioned

above, after both players receive an equal endowment player 1 is then asked to decide how much

of this endowment he or she wants to send to player 2. This amount is considered a good

measure of interpersonal trust, as player 1 knows that player 2 will receive three times the

amount given by him or her on top of the initial endowment everyone initially receives, and then

player 2 will return an amount to player 1 from a discrete set of fractions of amounts sent. In

addition, we include a vector of measures of welfare participation as our key variables of interest

along with basic controls, as described below. In particular, our reduced form follows the

specification:

(1)

where VolOfferi represents the voluntary offer of the particular individual i to another

individual in the trust game. We use the percentage of the initial endowment that player 1

9 Details on this specific game are provided in Candelo, et al. (2009) and Cardenas, et al., (2012). The other two experimental games applied were a Voluntary Contributions Game and a Risk Pooling Game. (Cardenas, et al., 2012).10 The exclusion of these variables among our controls delivers qualitatively similar results.

11

offered to player 2. Additionally, Xi is a vector of individual characteristics that includes age,

schooling, gender, and socio-economic level. Vector Zi captures information obtained from the

experimental sessions, in two dimensions: attitudes towards risk and (pre-game) expectations

about the behavior of the matched players. The vector M i contains socio-economic information

about the matched player; in particular, we control for gender and for differences between the

matched players in terms of schooling, age, and socio-economic level.

Our variable of interest, Wi , a measure of individuals’ participation in social welfare

programs, is defined in three alternative ways: (i) a simple dummy variable that captures whether

the individual is participant in a social welfare program; (ii) a variable defined as the percentage

of social programs of which the participants are beneficiaries, out of the list of social programs

that were listed in our survey; and (iii) an index that ranges from 0 to 4 depending on the groups

of social programs for which the individuals benefit, as the list of social programs is divided in

four general groups, education, health, nutrition, and child care For example, if the participant is

beneficiary of social programs related to only health and education, then his or her index will be

211. The exact definitions of these and all the other variables used in the paper are presented in

Table 112. Finally, i is a residual term.

Table 2 presents summary statistics, and Table 3 shows correlations between variables

used in the analysis. Related to the basic characteristics of the sample we find that the average

age of game and survey participants is 39, and that there is a reasonable gender balance as 54

percent of participants are women. Also, in our sample, 52 percent of the participants reside in a

low socio-economic level neighborhood and 30 percent in a medium-level neighborhood.

11 We also introduce this variable in percentages. Results do not change.12 Appendix 1 shows coverage of social welfare programs by city.

12

Finally, the average number of years of education reached by participants is almost 11 years13.

As mentioned above, Table 1 provides detailed definitions of variables.

4. Findings

In this paper we focus on the outcomes of the trust game and their link to participation in

social programs. In particular, Table 4 shows ordinary least squares results obtained when the

dependent variable is the voluntary offer of the first player in the trust game. All our regressions

include session dummies and include robust standard errors that are computed clustered at the

session level.14 We find that participation in social programs is negatively linked to the voluntary

amount offered by player 1, this variable interpreted as a measure of trust from the individuals to

their corresponding pairs. This result is statistically significant at conventional levels regardless

of the proxy employed to capture social welfare participation. Thus, our finding is consistent

with the idea that social welfare beneficiaries are associated with lower levels of interpersonal

trust, possibly due to stigma or discrimination, which is consistent with the literature of

disciplines such as sociology, social psychology and the medical sciences. This result, however

does not fully hold when studying each city separately. Whereas, overall our results are strongly

statistically significant, in some cities the sign is consistent but the significance is limited.15 On

the other hand, not all of our controls are statistically significant. We find that if the participant is

female and the second player is male, the participant will make a lower offer–compared to the

case where the participant is a male and the other player is female. In line with previous results

by Rabin (1993), the expectation of generosity of the matched individual (e.g., the second player)

appears to matter as well.

13 Further description of the sample can be found in Cárdenas et al. (2009).14 Results do not change when using ordered probit regressions instead. 15 Since our focus is on Latin America as a whole we do not present our city-level results. However they are available upon request.

13

It may be argued that endogeneity and in particular, reverse causality may be an issue

when testing the link between participation in social welfare programs as an explanatory variable

for interpersonal trust. In fact, as it stands it may be difficult to claim that participation in social

welfare programs causes a reduction in interpersonal trust and social capital. At the extreme, one

may argue that the opposite causal link is as likely to occur, that is, that more interpersonal trust

may drive higher participation in social programs either because interpersonal trust may also be

positively correlated to government and political trust or because more trusting individuals may

tend to have better social networks. While theoretically possible, as reverse causality is always

very difficult to be ruled out in empirical work, there are several reasons to be skeptical about

such issues between participation in social welfare programs and interpersonal trust. In fact, it is

unclear whether there is a link between interpersonal trust and political trust as several political

scientists have shown (Seligman, 1997; Benson and Rochon, 2004). In fact, some studies have

demonstrated a negative link between these two variables. For instance, Benson and Rochon

(2004) use World Value Surveys data to show that interpersonal trust is an important factor in

motivating political protest participation and raising the intensity of protest, thus negating a

positive link between interpersonal trust and political trust. These researchers further suggest that

high levels of trust make individuals likely to anticipate low expected costs of political trust and

participation.

Similarly, in the case of interpersonal trust linked to better social networks, it is difficult

to envision how improved social networks may increase participation in social welfare programs

at least in the context of most Latin American countries, for at least three reasons. First, it seems

unlikely that eligible individuals may not want to participate in a program that provides

substantial benefits at very-low-to-zero costs, at least under the basic assumption of rational

14

behavior. Thus, while strictly speaking it cannot be ruled out that an eligible, poor, needy

individual will not participate in, say, a government-sponsored social nutrition program, one

would have to argue that this may occur because of mistrust to such programs, tradition, personal

beliefs, and the like. Again, given that the cost of participating in such programs in both

monetary and non-monetary terms is essentially zero it does not seem likely that potential

beneficiaries would not participate in such programs, at least under basic standard neoclassical

considerations and if so, it would be reasonable to believe that such individuals choosing not to

participate would not be economically or statistically significant. Second, it is unclear why

better social networks may be able to provide potential beneficiaries with better transmission of

information of social welfare programs in the context of government oversaturation of such

social programs as, in fact, most Latin American countries own national networks (newspapers,

radio, and television), control affiliated media, and have dedicated Ministries that constantly

market the benefits of social welfare programs (Djankov, et al, 2003). Third, there is a very long

historical tradition of assistentialism in Latin American countries that date back to 1940s by

which populist dictators would provide a host of social programs and related favors to the lowest

income segments of the population in an exchange for implicit political support. The most

notorious historical example in Latin America is the case of Peron in Argentina, which has not

been, by any means unique in the region16. It is unlikely that there would be aversion by potential

beneficiaries to participating in welfare social programs sponsored by Latin American

governments and thus as, indeed, this has been historically expected. Thus, there is little reason

to see why social networks would needed influential in welfare program participation, further

undermining the idea of reverse causality between social welfare program participation and

16 In fact, Venezuela’s Chavism or Neo-Chavism is the most recent example of this long tradition (Hsieh, et al, 2011).

15

interpersonal trust, as stated in this paper. Finally, and in addition, the objective dependent

variable employed in this research seeks to reflect interpersonal trust in an “indirect” manner,

which we believe further helps minimize possible reverse causality issues.

Nevertheless, it may be argued that there may be some other channels that may drive

reverse causality from interpersonal trust to social welfare program participation to the ones

described above. Furthermore, it may be argued that endogeneity due to the presence of omitted

variables may still be an issue in our empirical results. Thus, in order to potentially correct for

this issue we proceed to employ an instrumental variables approach where the instrument chosen

is a categorical variable based on the ratio of children over income earners in the household. In

fact, chances are that households with higher household ratio of children over income earners

will have a higher probability of participating in social programs as the correlation coefficients

between participation in a social program and the ratio of children over earners using any our

three measures of participation is 0.478, 0.271, and 0.459 (See Table 3). On the other hand, there

appears to be no direct reason to expect that households with more children than income earners

will show different patterns of interpersonal trust as a result of a direct causal relationship

between them, since it is a measure of the households’ need.17

In fact, our main results hold when using the instrumental variables approach, as shown

in Table 5. The estimations presented are 2SLS, where the instrument included is the predicted

probability of the regressions of social programs participation on our proposed instrument and all

the control variables included in Table 4 estimations. For our first participation measure (i.e.

receives any social program), as it is a binary variable, this initial stage is a probit regression. For

17 The simple correlation between our instrument and the percentage offered to player 2 is -0.105. Also, we apply corresponding under-identification tests, namely, that the excluded instruments are relevant or correlated with the endogenous control. The null hypothesis that the equation is under-identified is rejected.

16

the other participation measures, given that they are left-censored, we run tobit regressions.

These initial stages of all instrumental variables regressions are in Appendix 2 and Appendix 3.18

We find that our three measures of participation in welfare social programs continue to yield

coefficients that are negative and statistically significant at conventional levels. The instrumented

estimators for the role of social programs on trust show higher magnitudes than the ordinary least

squares estimators, as expected. With respect to the other control variables, we find that some

gender combinations, the return offer expected from the second player, and medium risk

aversion are all still statistical significant. Initial probit and tobit estimations include session

dummies and city dummies.

In Tables 6 we go a step further and explore the linkages between social program

participation and interpersonal trust for three different social welfare domains separately,

education, health, and nutrition programs.19 In our ordinary least squares results we find an

analogous result of negative linkages between interpersonal trust and social welfare program

participation for every category, which is statistically significant at conventional levels.20

As before, we repeat the exercise of correcting for possible endogeneity with the use of

an instrumental variables approach. The results are shown in Table 7. In this case we confirm our

findings with respect to education, social health and nutritional programs. As expected, when

18 Additionally, it may be possible to argue that both trust and welfare participation may be linked to a third unobserved control, for instance, households’ shocks or life hardships (job loss, death of a bread winner, natural disasters, etc.). The occurrence of such negative shocks may be linked to a simultaneous change in interpersonal trust and individuals’ need of welfare, causing a spurious correlation among the two variables of interest. The design of the experiments and the data captured with the accompanying surveys do not allows us to explore such possibilities and thus leave these issues beyond the scope of this paper.

19 We are unable to consider the category child care programs because of lack of variation. In fact, less than two percent of our sample indicates being enrolled in such type of program.20 For the sake of completeness, in Appendix 4 we run our reduced form (1) but employ the offer of the second player. Unsurprisingly, we find that in this case, social participation in welfare programs has no bearing on reciprocity although the percentage expected to be received from matched player does matter.

17

using an instrumental variables approach, we find evidence of substantially higher coefficients in

our variables of interest, which suggest causal linkages between participation in specific social

welfare programs and interpersonal trust. Based on these instrumental variables findings, it may

be claimed that the participation of welfare programs appears to impact interpersonal trust

negatively, and thus contribute to damage social capital.

5. Summary and Conclusions

In this paper we offer experimental data on trust, which for the first time is studied in the

context of social welfare program participation. We apply a trust game to representative samples

of six Latin American cities with the aim of testing the extent to which the reception of social

programs is linked to a decrease in interpersonal trust. From the analysis of these experimental

data for representative samples of individuals we conclude that participation in social welfare

programs are linked with the destruction of social capital, as measured by interpersonal trust.

According to related literature, this may occur because of stigma or discrimination which

contributes to the increase of social distance between recipients and non-recipients. Furthermore,

our findings touch on the debate on whether low take up rates are due to transaction costs or

stigma and provide supporting evidence that they are due to the latter as stigma has been linked

with trust and social capital, but not with transaction costs. The results are robust to both the

inclusion of individual risk measures and to changes in specification when pooling our sample

and when testing most cities individually.

While, to the best of our knowledge, this is the first paper that addresses the issue of

welfare participation and social capital, the policy implications of this research may be even

more extensive. On the one hand, promoting higher take up rates would not necessarily be sound

policy as the trade off of damaging social capital would have to be taken into account. In the

18

context of Latin American countries, where social capital is not particularly high, there is a real

risk that social programs may be depleting such capital. On the other hand, under the assumption

that our dependent variable better reflects stigma rather than administrative costs, one shop stops

or centralized administrative offices for the delivery of social services would not be advisable.

It is important to highlight, however, that this paper studies only one aspect of trust,

interpersonal trust among agents. It is unclear whether trustworthiness issues may be affected,

measured by trust in the recipient of the trust game, which, at least in theory appear to be as

relevant as trust in the sender of the game. We expect that our future research will help address

these issues.

19

References

Ashraf, Nava, Colin Camerer, and George Loewenstein (2005), “Adam Smith, Behavioral Economist”, Journal of Economic Perspectives 19(3): 131–145.

Augoustinos, M. and C. Ahrens (1994) “Stereotypes and Prejudice: The Australian Experience”, British Journal of Social Psychology, 33: 125-141.

Barr, Abigail (2003) Risk Pooling, Commitment, and Information: An experimental test of two fundamental assumptions. Oxford University, Working Paper 187

Benson, Michele and Thomas Rochon (2004) “Interpersonal Trust and the Magnitude of Protests: A Micro and Macro Level Approach”, Comparative Political Studies, 37, 4: 435-457.

Berg, Joyce, John Dickhaut y Kevin McCabe (1995) Trust, Reciprocity and Social History. Games and Economic Behavior 10: 122-142;

Bertrand, Marianne and Sendhil Mullainathan (2001) "Do People Mean What They Say? Implications for Subjective Survey Data." Marianne Bertrand and Sendhil Mullainathan; American Economic Review, 91(2): 67-72

Binswanger, Hans (1980), “Attitudes towards Risk: Experimental Evidence from Rural India”, American Journal of Agricultural Economics 62: 395-407.

Blank, Rebecca and Patricia Ruggles (1996) “When Do Women Use AFDC and Food Stamps? The Dynamics of Eligibility versus Participation”, Journal of Human Resources, 31, 1: 57-89.

Bowden, C.L., L.S. Schoenfeld, and R.L. Adams (1980), Bowden, C.L., “Mental health attitudes and treatment expectations as treatment variables”, Journal of Clinical Psychology, 36: 653-657.

Burks, Stephen; Jeffrey Carpenter and Eric Verhoogen (2003). “Playing both roles in the trust game.” Journal of Economic Behavior & Organization, 51: 195-216.

Busso, Matias and Sebastian Galiani (2014) “The Causal Effect of Competition on Prices and Quality: Evidence from a Field Experiment”, Manuscript, University of Maryland, College Park.

Byrne, Peer (2000) “Stigma of Mental Illness and Ways to Diminishing it”, Advances in Psychiatric Treatment, 6: 65-72

Candelo, Natalia, Juan Camilo Cárdenas, Alberto Chong, Hugo Ñopo and Sandra Polania (2009). "Methodological Steps and Experimental Design to Measure Social Capital and Collective Action in Six Latin American Cities" Technical Paper, Research Department, Inter-American Development Bank. Washington, DC.

Cardenas, Juan Camilo, Alberto Chong, and Hugo Ñopo (2009), “To What Extent do Latin Americans Trust and Cooperate? Field Experiments on Social Exclusion in Six Latin American Countries”, Economia: Spring 2009, 9 (2): 45 – 94.

Cardenas, Juan Camilo, Alberto Chong, and Hugo Ñopo (2012) “Stated Behavior and Revealed Preference: Evidence from Six Latin American Countries”, Journal of Development Economics, 104: 16-33.

Carpenter, Jeffery, Glenn Harrison and John List (2005), Field Experiments in Economics, Vol 10, Research in Experimental Economics, JAI Press, Greenwich and London.

Chong, Alberto, Hugo Ñopo and Vanessa Rios (2009), “Do Welfare Programs Damage Interpersonal Trust? Experimental Evidence from Representative Samples for Four Latin American Cities” IDB WP #668. Washington, DC.

20

Crocker, J., B. Major and C. Steele (1998) “Social Stigma”. In D. Gilbert, S. T. Fioske and G. Lindzey (eds.) “The Handbook of Social Psychology”, vol 2, 4: 504-533), New York, McGraw Hill.

Currie, Janet (2004) “The Take up of Social Benefits” NBER Working Paper 10488Devine, P. (1995) “Prejudice and Out-Group Perception”, In: A. Tesser (ed.) Advanced Social

Psychology, New York: McGraw Hill.Djankov, Simeon, Caralee McLiesh, Tatiana Nenova, and Andrei Shleifer (2003) “Who Owns

the Media?” Journal of Law and Economics, 46, 2: 341-81Fiszbein, Ariel and Norbert Schady (2009) “Conditional Cash Transfers: Reducing Present and

Future Poverty”, The World Bank.Goffman, E. (1963) Stigma: Notes on the Management of Spoiled Identity. Engelwood Cliffs,

NJ: Prentice-Hall.Hilton, J. and W. von Hippel (1996) “Stereotypes”, Annual Review of Psychology, 47, 237-271.Hsieh, Chang-Tai, Edward Miguel, Daniel Ortega,and Francisco Rodriguez (2011), The Price of

Political Opposition: Evidence from Venezuela’s Maisanta, American Economic Journal: Applied Economics, 3: 196-214

Judd, C. and B. Park (1993) “Definition and Assessment of Accuracy in Social Stereotypes”, Psychological Review, 100: 109-128.

Kahn, M. W., Obstfeld, L., & Heiman, E. (1979). Staff conceptions of patients’ attitudes toward mental disorder and hospitalization as compared to patients’ and staff’s actual attitudes. Journal of Clinical Psychology, 35, 415-420.

Krueger, J. (1996) “Personal Beliefs and Cultural Stereotypes about Racial Characteristics, Journal of Personality and Social Psychology, 71: 536-548.

Kumlin, Staffan (2002) The Personal and the Political. How Personal Welfare State Experiences Affect Political Trust and Ideology., Department of Political Science, Göteborg University,

Kumlin, Staffan and Bo Rothstein (2005) “Making and Breaking Social Capital: The Impact of Welfare-State Institutions”. Comparative Political Studies, Vol. 38, No. 4, 339-365.

Levinson, Arik and Sjamsu Rahardja (2004) “Medicaid Stigma” Working Paper, Georgetown University

Levitt, Steven and John List (2007). What do Laboratory Experiments Measuring Social Preferences Reveal About the Real World?” Journal of Economic Perspectives, 21, 2: 153-174.

Moffitt, Robert (2005) “Policy Interventions, Low-Level Equilibria and Social Interactions” In Steven Durlauf and Peyton Young (eds.) Social Dynamics, Cambridge, MA: MIT Press.

Moffitt, Robert (1983) “An Economic Model of Welfare Stigma” American Economic Review, 73, 5: 1023-55

Neill, Helen; Ronald Cummings;Philip Ganderton; Glen Harrison and Thomas McGuckin (1994). “Hypothetical Surveys and Real Economic Commitments.” Land Economics, 70, 2:145-154.

Rabin, Matthew (1993), “Incorporating Fairness into Game Theory and Economics,” American Economic Review, 83, 1281-1302.

Remler, Dahlia and Sherry Glied (2003) “What Other Programs Can Teach Us: Increasing Participation in Health Insurance Programs”, American Journal of Public Health, 93, 1:67-74.

21

Rothstein, Bo (2001), “The Universal Welfare State as Social Dilemma”, Rationality and Society, Vol. 13, No. 2, 213-233.

Rothstein, Bo (2000), “Trust, Social Dilemmas and Collective Memories”, Journal of Theoretical Politics, Vol. 12, No. 4, 477-501.

Rothstein, Bo (1998) Just Institutions Matter: The Moral and Political Logic of the Universal State. New York. Cambridge University Press, 1998. 254 p.

Selden, Thomas, Jessica Banthin and Joel Cohen (1998) “Medicaid’s Problem Children: Eligible but Not Enrolled” Health Affairs, May-June 192-200

Seligman, Adam (1997) The Problem of Trust, Princeton University Press.Srivastava, Sameer and Tilcsik “Stigma, Sympathy, and the Double Edge of Strong Ties: Social

Capital Activation in Job Searches”Stuber, Jennifer, Kathleen Maloy, Sara Rosenbaum, and Karen Jones (2000) “Beyond Stigma:

What Barriers Actually Affect the Decisions of Low-Income Families to Enroll in Medicaid? Manuscript, George Washington University Medical Center, Center for Health Services Research and Policy.

Todd, Peter; Lars Penke; Barbara Fasolo and Alison Lenton (2007). “Different cognitive processes underlie human mate choices and mate preferences.” Proceedings of the National Academy of Sciences.

Watson, Amy and Patrick Corrigan (2003) “The Impact of Stigma on Service Access and Participation”, Manuscript, University of Chicago.

Wooldridge, J.M. (2002). Econometric Analysis of Cross Section and Panel Data, 752 pp. MIT Press, Cambridge, MA.

Wright, E. R., Gronfein, W. P., & Owens, T. J. (2000). Deinstitutionalization, social rejection, and the self-esteem of former mental patients. Journal of Health and Social Behavior, 41, 68-90.

22

Table 1 Variables Definitions

Variable DefinitionIndividual’s socio-demographic characteristicsAge (years) Age of the participant. Education (years) Number of years of education of the participant.Socio-economic level Categorical variable that indicates the socio-economic level of the participant:

low, medium or high. Each category was converted to a dummy variable. In all regressions, the first category (low socio-economic level) was the omitted dummy.

Matched players’ characteristics (related to individuals)Matched players that are woman

Dummy variable that takes the value of 1 when the matched player with the participant is a female and 0 otherwise.

Matched players that are men

Interaction variable that results from multiplying the participant’s gender by the matched player’s gender. It takes a value of 1, when both players are women; and 0, otherwise.

Average age difference between matched players

Difference in age of the matched player with the participant.

Average schooling difference between matched players

Difference in number of years of education of the matched player with the participant.

Socio-economic level compared to the matched player

Categorical variable that ranges between -2 to 2 as a result of subtracting the matched player’s socio-economic level from the participant’s one. Each category was converted into a dummy. In all regressions, the omitted dummy was the category “0”, when there was no difference between the matched players.

Experiment variablesInitial offer by Player 1 Percentage of money offered by player 1 to player 2 in the Trust Game. From

the amount received by player 1, he/she had five options: to give 0%, 25%, 50%, 75%, or 100% of his/her money to player 2.

Return offer by Player 2 The fraction of money that player 2 decided to send at the time of her/his decision. The numerator is the monetary amount sent at her/his move. The denominator is the initial endowment plus three times the amount sent by player 1.

Risk aversion Categorical variable that indicates the risk aversion level of the participant: low, medium or high. Each category was converted to a dummy variable. In all regressions, the first category (low risk aversion) was the omitted dummy.

Percentage expected to be returned by matched player (For player 1 in Trust Games regressions)

This variable can take values from 0% to 100%. It reflects player’s 1 expectation about the percentage to be returned by player 2, considering the different set of options he has (that set of options depends on the percentage of money gave by player 1).

Participation in social programsReceives any social program

Dummy variable that takes the value of one when the participant indicates that he/she or a member of his/her household is a beneficiary of a social program. The list of social programs is divided in four groups: education (E), health (H), nutrition (N), and child care (C). The list of specific programs included in each group varies depending on the city (programs in Buenos Aires and Caracas were not identified separately): Bogota: E: “Familias en Acción”, Labor training –SENA, “Jóvenes en

Acción”, “Hogares Infantiles” (ICBF or DABS), “Jardines Comunitarios”, and “Centro de Educación para Adultos”. H: “Régimen Subsidiado en Salud”, and “Jornada Nacional de Vacunación”. N: “Comedores comunitarios”, “Comedores Escolares” (ICBF), “Hogar de Bienestar” (FAMI), “Desayunos Infantiles” (ICBF), “Adulto Mayor” (PPSAM), and

23

Variable Definition“Red de Seguridad Alimentaria” (RESA). C: “Hogares Comunitarios”, “Club Juvenil o prejuvenil ICBF”

Lima: E: School uniform and shoes, Textbooks and school supplies, Labor training, and PRONOEI. H: “Seguro Integral de Salud”, “Campaña Nacional de Vacunación”, “Campaña de Planificación Familiar”, and “Control de Tuberculosis”. N: “Desayuno escolar”, “Vaso de leche”, “Comedor Popular”, “Canasta familiar” (PANFAR), “Alimento por trabajo”, “Comedor parroquial”, direct food aid, “Papilla u otro alimento para menores”. C: “Wawa Wasi”, and “Cuna”.

Montevideo: E: Textbooks and school supplies, School uniform and other clothing, Full time public school, “Asignaciones familiares”, and “Beca lineal”. H: “Programa de Educación Sexual y Planificación Familiar”, “Vacunación contra la gripe”, “Apoyo económico para control de embarazos”, and “Apoyo económico para control de niños”. N: “Comedores y merenderos”, “Reparto de canasta alimenticia”, “Reparto de leche en polvo”. C: “Verano solidario”, and CAIF.

San Jose: E: “Bono escolar”, Scholarships, “Transporte escolar”, and labor training. H: “Régimen no contributivo”, and “Vacunación gratuita”. N: “Comedor escolar”, “Comedor universitario”, “Comedor comunitario”, “CEN CINAI hogar comunitario”, “Leche”, “Paquetes de Alimentos”. C: “Guardería”.

Percentage of programs received

Percentage of social programs that the first players’ (in the trust games) households receive (out of all). If there are 10 social programs listed for that city, and the participant’s household received 2 of them, the variable value for this observation will be 20. It could take values from 0 to 100.

Reception of social programs (index)

Index that takes values from 0 to 4 depending on the groups of social programs that the first players’ (in the trust games) households receive. For example, if the participant is beneficiary of social programs related to health and education, then his index will be 2.

Reception of social program related to education, health, nutrition or child care

Dummy variable that takes the value of one when the participant indicate that his/her household is beneficiary of a social program related to education, health, nutrition or child care in his city; and zero, otherwise.

InstrumentsRatio of children over income earners in the household

Ordinal variable indicating the ratio of children over income earners in the household: Zero, 0 to 0.5, 0.5 to 1, 1 to 2, and More than 2. In all regressions, the category equal to zero was omitted.

24

Table 2Summary Statistics

Variable Mean Std. Dev.

Min Max

Household's participation in social programs        Receives any social program 0.281 0 1Percentage of programs received by the players' household (%) 3.199 6.805 0 61Reception of social programs (index) 0.445 0.806 0 3Receives a social program related to education 0.157 0 1Receives a social program related to health 0.174 0 1Receives a social program related to nutrition 0.088 0 1Receives a social program related to child care 0.031 0 1Individuals' socio-demographic characteristics        Age 39.172 15.176 17 80Years of education 10.942 3.591 0 22SE level

Low 0.52 0 1Medium 0.30 0 1High 0.18 0 1

CityBogotá 0.29 0 1Buenos Aires 0.38 0 1Caracas 0.06 0 1Lima 0.22 0 1Montevideo 0.04 0 1San José 0.01 0 1

Ratio of children over income earners in the household (Instrument)Zero 0.380 to 0.5 0.16 0 10.5 to 1 0.25 0 11 to 2 0.14 0 1More than 2 0.06   0 1

Differences between participants and matched playersGender difference

Both are women 0.31 0 1Both are men 0.24 0 1Participant is a woman and her match is a man 0.24 0 1Participant is a man and her match is a woman 0.22 0 1

Age difference 3.25 20.507 -54 53Schooling difference -1.12 4.351 -15 14SE level difference

Participant is at a lower SE level than his/her match 0.27 0 1Both are at the same SE level 0.55 0 1Participant is at a higher SE level than his/her match 0.19 0 1

Experimental variablesFirst player's initial offer 42.24 29.078 0 100Percentage expected to be returned by matched player 36.16 23.103 0 100Risk aversion

Low 0.17 0 1Medium 0.34 0 1High 0.50   0 1

Note: Summary statistics are based on weighted data using sample of main empirical specifications presented in Table 4 (1,465 observations). Standard deviations are only presented for continuous variables. Variables are defined in Table 1.

25

26

Table 3 Correlations matrix

First player's initial offer (Trust

measure)

Receives any

social program

Percentage of

programs received by the

players' household

Reception of social programs (index)

Receives a social program related to education

Receives any

social program related

to health

Receives a social program related

to nutrition

Receives a social program related to child

care

Age Years of

education

SE level

Percent. expected

to be returned

by matched player

Risk aversi

on

Differences between participants and matched players

Ratio of children

over income

earners in the

household

Age diff.

Schooling diff.

SE level diff.

First player's initial offer

1.000 -0.236 -0.165 -0.247 -0.355 -0.164 -0.118 -0.307

Age 0.035 -0.130 -0.062 -0.126 -0.119 -0.078 -0.089 -0.119 1.000Years of education 0.138 -0.306 -0.196 -0.298 -0.286 -0.213 -0.312 -0.235 -0.266 1.000SE level 0.127 -0.314 -0.259 -0.343 -0.238 -0.331 -0.459 -0.172 -0.192 0.636 1.000Percentage expected to be returned by matched player

0.260 0.077 0.076 0.068 -0.006 0.104 0.159 -0.183 0.084 -0.047 0.058 1.000

Risk aversion -0.033 0.078 0.038 0.065 0.048 -0.025 0.100 0.274 -0.030 0.066 -0.030 -0.020 1.000Gender differenceBoth are women -0.009 0.157 0.102 0.154 0.199 0.071 0.169 0.120 -0.033 -0.003 -0.055 -0.067 0.116 -0.159 0.117 -0.008 0.218Both are men 0.039 -0.201 -0.186 -0.196 -0.122 -0.201 -0.232 -0.162 0.014 -0.010 0.037 0.070 -0.174 0.124 -0.078 0.028 -0.136Participant is a woman and her match is a man

-0.076 0.018 0.001 -0.013 -0.179 0.062 -0.011 0.114 0.133 -0.014 0.004 0.018 0.047 0.170 -0.120 -0.033 -0.010

Participant is a man and her match is a woman

0.050 -0.008 0.023 0.017 0.048 0.044 0.013 -0.143 -0.120 0.028 0.021 -0.015 0.001 -0.127 0.062 0.015 -0.115

Age difference 0.053 -0.133 -0.064 -0.126 -0.157 -0.041 -0.114 -0.131 0.746 -0.212 -0.231 0.056 -0.018 1.000Schooling difference 0.020 0.025 -0.001 0.018 0.060 0.018 -0.042 -0.008 -0.244 0.623 0.333 -0.088 0.016 -0.332 1.000SE level difference 0.046 -0.023 -0.057 -0.035 0.046 -0.100 -0.137 0.098 -0.198 0.288 0.609 -0.029 -0.013 -0.214 0.510 1.000Ratio of children over income earners in the household (Instrument)

-0.105 0.478 0.271 0.459 0.518 0.298 0.360 0.416 -0.181 -0.192 -0.159 0.082 0.032 -0.146 -0.037 -0.027 1.000

Note: The matrix includes the Pearson, polychoric (tetrachoric), and polyserial (biserial) correlation coefficients, which were calculated according to their appropriateness for different continuous and categorical variables combinations.

27

Table 4Welfare Social Programs and Interpersonal Trust

Dependent variable: First player's initial offer(1) (2) (3)

Participation in social programs      Receives any social program=1 -9.5251***

(2.5423)Percentage of programs received -0.7173***

(0.1989)Reception of social programs (index1) -7.0492***

(1.4775)

Medium Risk Aversion 4.9799 4.5632 4.6809(3.6050) (3.6304) (3.5817)

High Risk Aversion -1.5221 -1.5188 -1.4968(3.2023) (3.2198) (3.2009)

Constant 36.5769*** 35.9658*** 38.1116***(9.6649) (9.5134) (9.4256)

Observations 1,465 1,465 1,465Individuals' socio-demographic characteristics  Yes Yes   YesMatched Players’ Characteristics Yes Yes YesCity Fixed Effects Yes Yes YesClusters 149 149 149Adjusted R-squared 0.2361 0.2417 0.2476* Significant at ten percent; ** significant at five percent; *** significant at one percent. Also controlled for matched players characteristics (gender, age, schooling, socio-economic status, and expectations of money return), individuals’ socio-demographic characteristics (age, education, and socio-economic status).

28

Table 5Welfare Social Programs and Interpersonal Trust, Instrumental Variables

Dependent variable: First player's initial offer(1) (2) (3)

Participation in social programs      Receives any social program=1 -15.7470*

(8.0837)Percentage of programs received -1.3456**

(0.5369)Reception of social programs (index) -9.5327**

(3.7503)Medium Risk Aversion 6.6696* 4.2082 4.5795

(3.4517) (3.4667) (3.3795)High Risk Aversion -1.3176 -1.2348 -1.3749

(3.1416) (3.0657) (3.0227)Observations 1,067 1,465 1,465Individuals’ socio-demographic characteristics Yes Yes YesMatched Players’ Characteristics Yes Yes YesCity Fixed Effects Yes Yes YesClusters 105 149 149Under identification test (Kleibergen-Paap LM statistic) 19.4670*** 23.1321*** 27.3927***Robust standard errors in parentheses, clustered at the session level. All regressions are run using 2SLS and include dummies per session. In (1), we calculated the predicted probabilities of a probit regression of the endogenous variable on the instruments and all control variables included in the main specifications. These predicted probabilities were used as instruments in the 2SLS regressions. A similar procedure was followed for (2) and (3), but using the predicted probabilities of tobit models. * Significant at ten percent; ** significant at five percent; *** significant at one percent.

29

Table 6Type of Social Programs and Interpersonal Trust

Dependent variable: First player’s initial offer(1) (2) (3)

Participation in social programs      Receives an education social program=1 -14.7525***

(3.1133)Receives a health social program=1 -10.6239***

(3.6944)Receives a nutrition social program=1 -6.5318

(4.0547)Medium Risk Aversion 5.1287 4.4100 4.8693

(3.6183) (3.5590) (3.6368)High Risk Aversion -1.4460 -2.1294 -1.6170

(3.2565) (3.2308) (3.2604)Constant 37.1910*** 35.4520*** 33.1797***

(9.6157) (9.4782) (9.6842)Observations 1,465 1,465 1,465Clusters 149 149 149Adjusted R-squared 0.2476 0.2355 0.2241* Significant at ten percent; ** significant at five percent; *** significant at one percent. Also controlled for matched players characteristics (gender, age, schooling, socio-economic status, and expectations of money return), individuals’ socio-demographic characteristics (age, education, and socio-economic status).

30

Table 7 Type of Social Programs and Interpersonal Trust, Instrumental Variables

Dependent variable: First player’s initial offer(1) (2) (3)

Participation in social programs      Receives an education social program=1 -16.9924**

(7.1625)Receives a health social program=1 -33.1684**

(14.5450)Receives a nutrition social program=1 -47.8183**

(22.4485)Medium Risk Aversion 8.2312** 2.0575 2.5738

(3.7810) (4.2274) (6.8829)High Risk Aversion 0.5570 -4.5467 0.3807

(3.5504) (3.8114) (6.5555)Observations 889 823 450Individuals’ socio-demographic characteristics Yes Yes YesMatched Players’ Characteristics Yes Yes YesCity Fixed Effects Yes Yes YesClusters 86 79 43Underidentification test (Kleibergen-Paap LM statistic) 27.4146*** 10.7141*** 6.4851***Robust standard errors in parentheses, clustered at the session level. Robust standard errors in parentheses, clustered at the session level. All regressions are run using 2SLS and include dummies per session. We calculated the predicted probabilities of a probit regression of the endogenous variable on the instruments and all control variables included in the main specifications. These predicted probabilities were used as instruments in the 2SLS regressions. * Significant at ten percent; ** significant at five percent; *** significant at one percent.

31

Appendix 1Individuals that Participate in Social Programs, By City

(percentage of all the first players)

Bogotá Buenos Aires Caracas Lima Montevideo San JoséAny social program 43.9% 5.5% 10.5% 53.9% 18.6% 17.4%Related to education 34.1% 0.7% 9.7% 19.3% 15.4% 9.6%Related to health 22.3% 0.3% 4.1% 46.9% 5.1% 10.6%Related to nutrition 8.3% 0.0% 3.8% 27.5% 3.6% 1.6%Related to child care 4.3% 4.5% 0.0% 0.3% 2.4% 0.4%

Note: Calculations are based on weighted data using sample of main empirical specifications presented in Table 4 (1,465 observations). 1/ Data is for 2006, excepting Uruguay, for which only 2005 was available. It includes expenditure in education, health and housing. Source: CEPAL (Web).

32

Appendix 2Instrumental Variables Initial Stage, Table 5

(1) (2) (3)Receives any

social programPercentage of

programs receivedReception of social programs (index)

Ratio of children over income earners in the household (Instrument, base=zero)0 to 0.5 0.5014** 6.9121*** 0.7927***

(0.2268) (2.2435) (0.2812)0.5 to 1 0.9807*** 11.9581*** 1.4720***

(0.2151) (2.2742) (0.2704)1 to 2 1.2478*** 13.1298*** 1.6473***

(0.2645) (2.4333) (0.2854)More than 2 1.3300*** 13.0662*** 1.7422***

  (0.2905) (2.6198) (0.3025)Individuals' socio-demographic characteristicsAge (years) 0.0008 -0.0097 -0.0004

(0.0071) (0.0751) (0.0087)Years of education -0.0478 -0.2784 -0.0497

(0.0408) (0.3394) (0.0382)SES (base=Low)

Middle -0.4614** -5.7201** -0.6812**(0.2330) (2.3020) (0.2656)

High -1.3411*** -13.8560*** -1.8013***  (0.4089) (4.3737) (0.5391)Matched player' charcteristics (related to individuals)Gender (base= Participant is a man and the matched is a woman)  

Both players are women=1 0.2334 2.3477 0.2232(0.1544) (1.7564) (0.1841)

Both players are men=1 -0.0869 -2.0977 -0.1619(0.2194) (2.2023) (0.2708)

Participant is a woman and the matched is a man=1

0.1050 1.1629 0.0433(0.1889) (2.0374) (0.2203)

Age difference -0.0061 -0.0245 -0.0044(0.0049) (0.0537) (0.0057)

Schooling difference 0.0177 0.0841 0.0174(0.0224) (0.2029) (0.0248)

SES difference (base=Both are equal)Participant is at a lower SES -0.1279 -1.1534 -0.2284

(0.2164) (2.0417) (0.2538)Participant is at a higher SES 0.4578* 4.3783 0.5111

  (0.2690) (2.6728) (0.3226)Experimental variablesPercentage expected to be returned by matched player

0.0019 0.0055 0.0001(0.0028) (0.0262) (0.0030)

Risk aversion levels (base=Low)Medium -0.0043 -1.3994 -0.1187

(0.2369) (2.3488) (0.2957)High 0.0440 0.4124 0.0045

(0.2152) (2.2076) (0.2709)Observations 1,067 1,466 1,466Clusters 105 149 149Robust standard errors in parentheses, clustered at the session level. For (1), we used a probit model; and for (2) and (3), tobit models (left-censored, lower limit equals zero). The predicted probabilities from these regressions were used as instruments in the 2SLS regressions presented in Table 5. * Significant at ten percent; ** significant at five percent; *** significant at one percent.

33

Appendix 3Instrumental Variables Initial Stage, Table 7

(1) (2) (3)Receives an education

social programReceives a health

social programReceives a nutrition

social programRatio of children over income earners in the household (Instrument, base=zero)

0 to 0.5 1.5018*** 0.5504** -0.2324(0.3521) (0.2580) (0.4227)

0.5 to 1 2.1667*** 0.7842*** 0.4151(0.3147) (0.2506) (0.4342)

1 to 2 2.3311*** 0.7866*** 0.5464(0.3148) (0.2554) (0.4434)

More than 2 2.5989*** 0.3815 0.8798*  (0.3295) (0.3489) (0.4640)Individuals' socio-demographic characteristicsAge (years) 0.0065 -0.0058 0.0003

(0.0082) (0.0104) (0.0126)Years of education -0.0841** 0.0161 -0.0007

(0.0426) (0.0431) (0.0404)SES (base=Low)

Middle -0.4357 -0.6877*** -0.6612*(0.2655) (0.2450) (0.3430)

High -1.8359*** -1.2584*** -0.5596  (0.4883) (0.4884) (0.6353)Matched player' characteristics (related to individuals)Gender (base= Participant is a man and the matched is a woman)  

Both players are women=1 0.0448 0.1077 0.4618(0.2268) (0.1733) (0.3076)

Both players are men=1 -0.0795 -0.2097 -0.2772(0.2882) (0.2484) (0.4138)

Participant is a woman and the matched is a man=1 -0.5809** 0.2159 0.2103

(0.2459) (0.2197) (0.2953)Age difference -0.0064 0.0057 -0.0073

(0.0052) (0.0058) (0.0075)Schooling difference 0.0143 0.0069 -0.0176

(0.0314) (0.0255) (0.0344)SES difference (base=Both are equal)

Participant is at a lower SES -0.1904 -0.2189 -0.2057(0.2953) (0.2675) (0.3523)

Participant is at a higher SES 0.4659 0.1774 -0.1707  (0.2995) (0.2786) (0.3968)Experimental variablesPercentage expected to be returned by matched player

-0.0039 -0.0006 0.0014(0.0046) (0.0032) (0.0048)

Risk aversion levels (base=Low)Medium 0.0747 -0.3686 -0.1921

(0.3282) (0.2647) (0.3445)High 0.0220 -0.2153 0.2449

(0.3018) (0.2477) (0.3734)Observations 889 823 450Clusters 86 79 43Robust standard errors in parentheses, clustered at the session level. For all regressions, we used probit models. The predicted probabilities from these regressions were used as instruments in the 2SLS regressions presented in Table 7. * Significant at ten percent; ** significant at five percent; *** significant at one percent.

34

Appendix 4The Case for Reciprocity

Dependent variable: Second player's return offer(1) (2) (3)

Participation in social programs      Receives any social program=1 -0.7558

(2.2290)Percentage of programs received 0.0092

(0.1494)Reception of social programs (index) -0.0981

(1.2087)Individuals' socio-demographic characteristics    Age (years) 0.0977 0.0983 0.0982

(0.0868) (0.0869) (0.0870)Years of education -0.1323 -0.1166 -0.1233

(0.4010) (0.3981) (0.3953)SES (base=Low)

Middle -2.4342 -2.3741 -2.3951(2.4056) (2.4109) (2.4072)

High -4.0897 -3.9374 -3.9735(3.7729) (3.7145) (3.7375)

Matched player' characteristics (related to individuals)    Gender (base= Participant is a man and the matched is a woman)

Both players are women=1 -4.2464** -4.2710** -4.2646**(2.1089) (2.1165) (2.1138)

Both players are men=1 -4.0156* -4.0432* -4.0464*(2.2026) (2.2206) (2.2113)

Participant is a woman and the matched is a man=1

-4.2895** -4.3104** -4.3107**(1.9920) (1.9986) (1.9972)

Age difference -0.0021 -0.0006 -0.0011(0.0617) (0.0619) (0.0618)

Schooling difference -0.1712 -0.1686 -0.1673(0.3028) (0.3000) (0.3001)

SES difference (base=Both are equal)Participant is at a lower SES -2.6715 -2.6530 -2.6526

(3.0256) (3.0188) (3.0242)Participant is at a higher SES 5.3047** 5.2898** 5.2866**

(2.2507) (2.2340) (2.2343)Experimental variables      Percentage expected to be returned by matched player

0.2122*** 0.2131*** 0.2129***(0.0325) (0.0328) (0.0326)

Constant 19.1998*** 18.7073*** 18.8701***(6.3130) (6.2625) (6.2205)

Observations 1,522 1,522 1,522Clusters 149 149 149Adjusted R-squared 0.2806 0.2804 0.2804Robust standard errors in parentheses, clustered at the session level. All regressions are run using OLS and include dummies per session. * Significant at ten percent; ** significant at five percent; *** significant at one percent.

35