Economics Letters 123 (2014) 144–148

Contents lists available at ScienceDirect

Economics Letters

journal homepage: www.elsevier.com/locate/ecolet

Solidarity among the poor

Angela C.M. de Oliveira a,∗, Catherine C. Eckel b, Rachel T.A. Croson c

a University of Massachusetts Amherst, Department of Resource Economics, 203 Stockbridge Hall, 80 Campus Center Way, Amherst, MA 01003, USAb Texas A&M University, Department of Economics, 4228 TAMU, College Station, TX 77843, USAc University of Texas, Arlington, Dean, College of Business, 701 S. West Street, Room 334, Box 19377, Arlington, TX 76019, USA

h i g h l i g h t s

• We design and implement a visual version of the solidarity game for use in low-literacy populations.• We find significant evidence of conditional gifts (informal risk sharing) in a low income population.• Less than 7% of participants do not make any conditional gifts. These individuals are more risk tolerant than other participants.• We find substantially more ‘fixed gift’ behavior than previous studies, over 40% of the participants.

a r t i c l e i n f o

Article history:Received 20 August 2013Received in revised form15 January 2014Accepted 24 January 2014Available online 7 February 2014

JEL classification:C93D81

Keywords:SolidarityField experimentInformal risk sharingSocial preferencesPoverty

a b s t r a c t

We conduct a field experiment with low-income subjects in Dallas, Texas. We examine voluntary,informal risk sharing using a visual representation of the solidarity game developed for low-literacypopulations. We find substantially more ‘fixed gift to loser’ behavior and less ‘egotistical’ behavior than inprevious studies. Individuals who display ‘egotistical’ behavior are more risk tolerant. The amount of theconditional gifts is positively related to age, income, and connection to the community. However, trustand empathy, which are commonly discussed as drivers for solidarity, are not significantly related to theamount given.

© 2014 Elsevier B.V. All rights reserved.

1. Introduction

Solidarity, a driving force behind risk sharing, is a type of in-direct reciprocity; taking care of others who have ended up in abad financial situation, purely by chance. Informal risk sharing ar-rangements are most often observed among individuals living ator below the poverty line, and provide an important financial al-ternative for those with few market-based options for borrowing

Abbreviations: SO98, Selten and Ockenfels, 1998.∗ Corresponding author. Tel.: +1 413 545 5716.

E-mail addresses: acm.de.oliveira@gmail.com, adeolive@resecon.umass.edu(A.C.M. de Oliveira), ceckel@econmail.tamu.edu (C.C. Eckel), croson@uta.edu(R.T.A. Croson).

http://dx.doi.org/10.1016/j.econlet.2014.01.0250165-1765/© 2014 Elsevier B.V. All rights reserved.

or insurance. However, very little is understood about the behav-ioral propensity to risk-pool, nor about the potential behavioral re-sponses that might result. On the one hand, risk pooling provides asafety net for individuals most susceptible to shocks. On the otherhand, it reduces incentives to self-insure against losses. Previousresearch on solidarity, or risk pooling more generally, has focusedon establishing the phenomenon and understanding underlyingmotivations for self-selecting into risk-pooling groups (Barr andGenicot, 2008; Büchner et al., 2007; Charness and Genicot, 2009;Selten and Ockenfels, 1998), cultural differences (Brosig-Kochet al., 2011; de Beer and Berg, 2012a,b; Ockenfels and Weimann,1999), the role of social networks in risk-pooling decisions (Attana-sio et al., 2012; Fafchamps and Lund, 2003); luck, deservingnessand wealth differences (Chaudhuri et al., 2005; Trhal and Rader-macher, 2009). However, these studies have primarily focused on

A.C.M. de Oliveira et al. / Economics Letters 123 (2014) 144–148 145

Fig. 1. Solidarity game instruction page.

either student samples (e.g., Charness and Genicot, 2009) or less-developed countries (e.g., Attanasio et al., 2012).1 We contribute tothe literature by developing and implementing a variation of theSolidarity game (Selten and Ockenfels 1998, hereafter SO98) witha simple, visual representation, and examining behavior in a low-income urban neighborhood in the US.

In the solidarity game, participants are placed in random andanonymous groups of three. Each has an independent 2/3 chanceof receiving $75 and 1/3 chance of receiving nothing. Beforeoutcomes are known, subjects decide how much of their earningsto send, conditional on winning, to those who lose.

We find evidence of substantial informal risk sharing whenthe opportunity is available. While we identify the same types ofgiving behavior observed in previous studies, we find a differentdistribution of types, with substantially more ‘fixed gift to loser’behavior: over 40% of individuals make decisions that guaranteea set minimum payoff to their group members, even though thegroups are anonymous. We further find that ‘egotistical’ individu-als, thosewho do notmake conditional gifts, aremore risk tolerant.Conditional gifts are positively related to income and connection tothe community.

2. Experimental design and field implementation

We adapt the SO98 design for a low-income population byintroducing a visual representation and by increasing the stakes,so that subjects can win $75 with 2/3 probability and $0 with 1/3probability. Fig. 1 shows the graphic representation. Each of thethree players has a bagwith twowinning chips,markedW , and onelosing chip, marked L. To determine payment, person pulls a chipout of the bag, and that chip determines whether they win or lose.

Before pulling a chip, each subject has to make two decisions.The decision form is shown in Fig. 2. The form shows two situa-tions: When the subject and one other person win (top panel) andwhen the subject was the only winner (bottom panel). They wereinstructed to write down the amount they wanted to put in theirwallet and the amount they wanted to send to the loser(s) in eachsituation.2

Experiments were conducted as part of a larger field study ex-amining neighborhood quality and neighborhood change.3 Sub-jects were chosen randomly from 496 individuals who completed

1 Notable exceptions are de Beer and Berg (2012a,b) who use an urbanenvironment (Amsterdam), but one that is more financially affluent. For brevity weprovide illustrative examples, not an exhaustive review of the literature.2 Note that while we did not force individuals to send the same amount to each

of two losers, 183 out of 199 who completed the game chose to send an identicalconditional gift to each. Full instructions are available from the authors. Note thatSO98 conduct a double-blind study whereas ours is not. All subjects complete theirbooklets using a code number, but the number is not randomly assigned.3 More details are available at

http://www.utdallas.edu/~murdoch/NeighborhoodChange/index_nc.html.

Fig. 2. Solidarity game decision form.

the detailed household survey where one participant per house-hold was recruited from a random selection of tax parcels in theneighborhood (Leonard et al., 2011). A total of 201 subjects par-ticipated in the experimental sessions in October 2009, November2009, and February 2010 and ranged in size from two to nineteensubjects, with a mean of 10.4 Subjects could participate in onlyone session. All sessions were run at a centrally-located field sta-tion maintained for this study, and transportation was providedwhen necessary. The same lead experimenter ran all sessions, withtrained assistants drawn from both the community and from Cen-ter for Behavioral and Experimental Economic Science at the Uni-versity of Texas at Dallas. Subjects arrived, gave informed consent,and were paid a $20 show-up fee.

Subjects participated in a series of experiments to elicit prefer-ences for individual risk, correlated risk, skewness, and time pref-erences as well as the dictator, trust, and solidarity games. Soli-darity game results are this study’s focus. We additionally use theaverage gift from a comparative dictator game (DG) in the analysis.Subjects made four DG decisions with other anonymous strangersfrom their community. They received some limited informationabout recipients, possibly (but not necessarily) including gender,marital status, number of children, employment status and disabil-ity status.

Experimental tasks were followed by a survey. Further, somesubjects completed additional surveys as part of the larger study,which were conducted on different dates/times. The experimentalgames were always run in the same order, with no feedbackbetween tasks.5 One game was chosen at random for payment forall subjects in a session. Average earnings were $50.16 (min = $0,max = $170), plus the $20 show-up fee.

3. Aggregate gifts and strategies

We begin with a discussion of aggregate results for the baselinesolidarity game, Appendix A details all of the conditional gifts.

4 Session size is never statistically significant in our analysis (either indepen-dently or in interaction with the key variables) and so it is omitted.5 This design choice means that we cannot explicitly test for order effects, nor

can we rule out the influence of order on the contribution levels chosen. Paying oneactivity, with no feedback between activities, should help minimize these effects:Subject only receives feedback for the task for which they will be paid.

146 A.C.M. de Oliveira et al. / Economics Letters 123 (2014) 144–148

Table 1Summary of total conditional gifts and strategies, by treatment.

Data, stakes Mean gift % of endowment 1-loser premium Conditional gift classification, % of respondents1 loser 2 losers Egotistical Fixed sacrifice Fixed gift Intermediate

South Dallas $75 24.91 37.28 2.67 6.53 21.11 42.21 15.35SO98a—West DM 10 24.60 31.20 3.15 21 36 16 11OW99b—East DM 10 16.20 20.20 3.21 47 N/A N/A N/Aa Selten and Ockenfels (1998).b Ockenfels and Weimann (1999).

Table 2Level of voluntary risk-sharing, tobit.

Observables Preferences1 loser 2 losers 1 loser 2 losers

Female −0.771 −6.239*−0.835 −6.359*

(1.97) (2.84) (1.79) (2.66)Age, years 0.141 0.230* 0.116 0.215*

(0.07) (0.11) (0.07) (0.11)Income 0.813 1.733* 0.650 1.619*

(0.56) (0.80) (0.52) (0.77)Mean DG – – 0.312*** 0.407***

(0.07) (0.10)Empathy – – 0.233 0.306

(0.18) (0.27)Risk tolerance – – −0.609 −1.568

(0.62) (0.91)Identify – – 1.987 2.364

(1.73) (2.57)Know Name – – 3.889** 4.144*

(1.25) (1.85)GSS trust – – 0.219 −0.391

(0.35) (0.52)Constant 12.076*** 19.259***

−1.910 5.448(3.66) (5.27) (5.90) (8.71)

LNL −658.87 −711.68 −639.57 −696.65χ2 , 6.46 16.38 45.05 46.44(Prob > χ2) (0.00) (0.00) (0.00) (0.00)

Tobit with 16 observations censored at zero for one loser and 18 censored at zerofor two losers. N = 178 due to missing observations on some of the surveymeasures, standard errors in parentheses. The dependent variable is the amountof the conditional gift.

* p ≤ 0.05.** p ≤ 0.01.*** p ≤ 0.001.

Table 1 shows the mean total conditional gift for one and twolosers as a percentage of the endowment, as well as the ‘1-loserpremium’, defined below. The average per-person gift is $18.68 forone loser (whowould receive gifts from twowinners) and the totalconditional gift for two losers is $27.96, or $13.98 each.

Examining decision rules more closely, only 13 subjects (6.5%)contribute zero in the case of both one and two losers, termed‘egotistical’ by SO98, which is substantially smaller than found inSO98 (21%) or Ockenfels and Weimann (1999, 47%–48%). By andlarge, subjectsmake positive conditional gifts. Twopatterns appearto be focal: In the first, termed ‘exact fixed total sacrifice’ (fixedsacrifice hereafter) by SO98, the decision maker offers the sametotal amount to losers, whether there are one or two. One can thinkof this decision rule as being similar to a tithe: a set amount, basedon the endowment and the decision makers’ preferences, is givenregardless of the need. We find that 21.1% of individuals makepositive contributions consistentwith this decision rule, comparedwith their 36%.6

6 This figure does not include their category, fixed total sacrifice ‘up to rounding’,where subjects essentially make a fixed sacrifice, except that they round to thenearest integer. We do not have any amounts that are not integers. Including thosesubjects brings their percentage of subjects exhibiting fixed sacrifice behavior to52%.

The second pattern corresponds to the case where the decisionmaker wants each of the losers to get the same amount, termed‘fixed gift to loser’ (fixed gift hereafter). If there is one loser, thedecision maker contributes $X, and if there are two losers, thedecision maker contributes $2X. We find that 42.2% of decisionsare consistent with this pattern compared with 16% in SO98. Notethat 34 subjects exhibit a desire to achieve an egalitarian outcome,sending $25 to one loser and $50 (or $25 each) to two losers. If thedecisionmaker believes that the secondwinnerwill contribute $25to the loser if they win, then this pattern of conditional gifts wouldresult in payoffs of $50 each if there are two winners and $25 eachif there is one winner.

A useful aggregate measure of giving behavior is the ‘1-loserpremium’, proposed by SO98. It summarizes how much more aloser receives if there is only one loser in the group compared towhen there are two losers.7 We find that one loser can expect 2.67times more than when there are two losers, compared with 3.15in SO98. This suggests that, in the aggregate, the decision rule inour population is closer to the fixed gift than fixed sacrifice: a fixedsacrifice decision rule would produce a factor of 4, and a fixed gifta factor of 2. The prevalence of fixed gift as opposed to fixed sac-rifice behavior is particularly interesting in this population: SO98show that fixed sacrifice behavior is inconsistent with altruisticutility maximization and argue that it reflects a self-serving normwhereas fixed gift is an other-regarding behavior.8

4. Individual choices

A number of factors may influence the decision to voluntarilymake a conditional gift. In introducing the game, SO98 suggestpossible underlying mechanisms: ‘‘Solidarity means a willingnessto help people in need who are similar to oneself but victims ofoutside influences such as unforeseen illness, natural catastrophes,etc’’. (p. 518). This definition proposes that to identifying withthe potential losers, and either altruistic tendencies or beingempathetic towards the needy are likely to influence gifts. Further,though gifts are not reciprocated in this environment, if the gameis measuring some aspect of the subjects’ informal risk-sharingnetwork, then indirect reciprocity and trust as well as individualrisk tolerance are likely to impact the choicesmade as well. Table 2shows the correlates of conditional gifts.

The dependent variable is the amount of the conditional gift, indollars. The first pair of columns includes some of the observablecharacteristics often considered in charitable giving studies: Gen-der, Age, and Income. We find mixed evidence for gender: womengive less when there are two losers, but there is no difference foronly one loser. This is somewhat in line with previous studies:Büchner et al. (2007) find no significant differences by gender inthe solidarity gamewhereas Charness and Genicot (2009) find thatwomen transfer less in their risk-sharing game and Brosig-Kochet al. (2011) find a marginally negative impact on contributions.However, SO98 find women less likely to be egotistical.

7 The 1-loser premium is calculated as follows: (mean gift to one loser ×

2)/(mean gift to two losers/2).8 We would like to thank Axel Ockenfels for highlighting this point.

A.C.M. de Oliveira et al. / Economics Letters 123 (2014) 144–148 147

We further find older individuals transferring larger amounts.Individualswith higher incomes alsomake higher conditional gifts.Our incomemeasure is a lumpy indicator from0 to 8. Zero indicatesan annual household income of less than $10 K, 8 indicates an an-nual household income of between $80 and $90 K. Higher numbersstep up into the next $10K range.9 Our estimates indicate that indi-viduals with higher incomesmake larger conditional gifts, but onlysignificantly for the case of two losers.

The second pair of columns includes preference andmotivationvariables that have been discussed in the literature. First, we seethat the mean dictator gift is positively and significantly related togifts in the solidarity game. Our empathymeasure is from theDavis(1980) Interpersonal Reactivity index. Like Büchner et al. (2007),we do not see a significant relationship between the empathy scaleand the conditional gifts.

Risk Tolerance is the Eckel–Grossman risk measure (Eckel andGrossman, 2008).10 We find no significant effect of risk toleranceon the amount an individual is willing to contribute. However, wedo find (not shown) that risk significantly correlates with strategychoice, and in the manner expected (similar to Charness andGenicot): Individuals who are more risk tolerant are more likelyto make choices consistent with the egotistical decision rule, andless likely to make choices consistent with any of the other rules.11

Based on the definition of solidarity put forth in SO98, the abilityto identify with the needy might play a role in the gift decision.We therefore include two measures of identity and relatedness.‘Identify’ is a dummy variable equal to one if the subject indicatesthat they strongly agree that they see themselves as a member oftheir neighborhood. ‘Know Name’ is the number of individuals inthe experimental session that the subject reports they know byname (mean 0.38, max 4, within the estimation sample). Whileself-reported identificationwith the community is not significantlyrelated to gift behavior, knowing the name(s) of others in thesession is—even though all decisions are anonymous, groups arerandomly matched, and the likelihood of being matched withsomeone you know is small. Finally, trust, measured using thetypical General Social Survey trust question, is not related toconditional gift behavior.12

Knowing the name of individuals in one’s session could causeparticipants to feel closer to those in the session or it could causeindividuals to feel more closely related to their community. If thelatter is true, we might expect to see a correlation between ‘knowname’ and ‘identify’. However, this is not supported by the data(corr = 0.01, p = 0.98). We could also think of splitting the databy thosewho do and do not knowat least one person in the session,shown in Table 3.

An interesting pattern develops: Among those who know noone in the room, individuals who identify strongly with the com-munity make significantly higher conditional gifts. This is not thecase for individuals who know the name of at least one individ-ual, where the effect is negative and insignificant. For individualswho know at least one person in the session, the number of peo-ple whose name is known is related to the gift to one loser but notthe gift to two losers. Taken together, this result is consistent witha type of substitution between identities: General neighborhood

9 For the estimation sample, the householdmedian is in the $10–$20 K range andthe household mean is 1.20 (std. dev. 1.72), or in the $20–$30 K range.10 Risk takes a value of one if the subject is not willing to take on any risk, twothrough four indicate decreasing levels of risk aversion, five indicates expected-value maximization, and six indicates risk seeking behavior.11 Note that none of the other demographic or preference variables aresignificantly related to strategy choice except that women are less likely to displayintermediate behavior.12 We additionally tested trust, as measured by a trust game. The amount sentin the trust game is not significantly related to behavior and reciprocity is onlyrobustly related to behavior when the second player is sent nothing (which isessentially a dictator game). Results are available from the authors upon request.

Table 3Level of voluntary risk-sharing by ‘Know Name’, tobit.

Know 0 (N = 128) Know > 0 (N = 52)1 loser 2 losers 1 loser 2 losers

Identify 5.363* 6.372*−4.586 −5.983

(2.15) (3.23) (2.93) (4.27)Know Name – – 5.624** 4.254

(2.03) (2.96)LNL −445.35 −492.43 −187.10 −206.36

Tobit, standard errors in parentheses. The dependent variable is the amount of theconditional gift. The analysis includes all of the variables in Table 2, with estimatessuppressed.

* p ≤ 0.05.** p ≤ 0.01.

identity is a significant factor in the decision process until thosewho are closer in terms of social distance are involved. Then, theidentity associated with that group may dominate in this decisionenvironment.

5. Closing discussion

We develop and implement a visual representation of the SO98solidarity game for use in low-income/low-literacy populations.On the whole, the evidence indicates substantial levels of volun-tary, informal risk sharing in this population. The magnitudes ofthe conditional gifts are particularly striking given the low in-comes and apparent need in this population: 32.3% of subjects in-dicated that the money they earned would be used to pay bills and39.3% indicated the funds would go towards necessities (food, gas,medicine).

Contrary to previous studies, the most common decision ruleobserved in our population is fixed gift rather than fixed sacrifice:this is particularly interesting since SO98 suggests that fixed giftis a more altruistic behavior. However, we do not find universalsupport for the role of previously hypothesized underlying moti-vations impacting the choice of conditional gifts. While individu-als who are more risk tolerant are more likely to make egotisticaldecisions, it does not impact the amount of the conditional gift.We find mixed evidence that women make lower conditional giftsand that older subjects make higher conditional gifts. Connectionto the community is associated with higher gifts. Understandingthe behavioral foundations of this behavior, as well as the role ofcommunity, provides an interesting avenue for future research.

Acknowledgments

The authors completed the work for this manuscript while atthe University of Texas at Dallas in addition to their present affil-iations. A number of people assisted at various stages of this re-search.Wewould especially like to thankNatalia Candelo Londoño,Elizabeth Pickett, Lance Mattingly, Tammy Leonard, and otherresearchers associated with the Neighborhood Change ResearchInitiative at the UT Dallas. Axel Ockenfels, Lata Gangadharan,Alexander Smith, and Christine Binzel provided helpful commentson previous versions of this manuscript. Funding provided by theNational Science Foundation, HSD award # 0827350.

Appendix A

See Table A.1.

Appendix B. Supplementary data

Supplementary material related to this article can be foundonline at http://dx.doi.org/10.1016/j.econlet.2014.01.025.

148 A.C.M. de Oliveira et al. / Economics Letters 123 (2014) 144–148

Table A.1Conditional gift tables.

Notes: The x1 column indicates the conditional gift for one loser, and x2 indicatesthe conditional gift for two losers. The numbers in the cells are subject countsfor that (x1, 2x2) pair. Light gray shading indicates fixed sacrifice, dark grayshading indicates fixed gift, and diagonal shading indicates intermediate behavior.Contingency table tests confirm that the gifts are not independent, p < 0.00.

References

Attanasio, O., Barr, A., Cardenas, J.C., Genicot, G., Meghir, C., 2012. Risk pooling, riskpreferences, and social networks. Am. Econ. J. Appl. Econ. 4, 134–167.

Barr, A., Genicot, G., 2008. Risk sharing, commitment, and information: anexperimental analysis. J. Eur. Econ. Assoc. 6, 1151–1185.

Brosig-Koch, J., Helbach, C., Ockenfels, A., Weimann, J., 2011. Still different afterall these years: solidarity in East and West Germany. J. Public Econ. 95,1373–1376.

Büchner, S., Coricelli, G., Greiner, B., 2007. Self-centered and other-regardingbehavior in the solidarity game. J. Econ. Behav. Organ. 62, 293–303.

Charness, G.B., Genicot, G., 2009. Informal risk sharing in an infinite-horizonexperiment. Econom. J. 119, 796–825.

Chaudhuri, A., Gangadharan, L., Maitra, P., 2005. An experimental analysis of groupsize and risk sharing. Working paper.

Davis, M.P., 1980. A multidimensional approach to individual differences inempathy. JSAS Catalog Sel. Doc. Psychol. 85.

de Beer, P., Berg, M., 2012a. Conditions and motives for voluntary sharing: resultsof a solidarity game experiment. AIAS Working Paper No. 124.

de Beer, P., Berg, M., 2012b. Solidarity in a multicultural neighbourhood: results ofa field experiment. AIAS Working Paper No. 125.

Eckel, C., Grossman, P.J., 2008. Forecasting risk attitudes: an experimental studyusing actual and forecast gamble choices. J. Econ. Behav. Organ. 68, 1–17.

Fafchamps, M., Lund, S., 2003. Risk-sharing networks in rural Philippines. J. Dev.Econ. 71, 261–287.

Leonard, T.C.M., O’Brian Caughy, M., Mays, J.K., Murdoch, J.C., 2011. Systematicneighborhood observations at high spatial resolution: methodology andassessment of potential benefits. PLoS One 6, e2022.

Ockenfels, A., Weimann, J., 1999. Types and patterns: an experimental East-West-German comparison of cooperation and solidarity. J. Public Econ. 71, 275–287.

Selten, R., Ockenfels, A., 1998. An experimental solidarity game. J. Econ. Behav.Organ. 34, 517–539.

Trhal, N., Radermacher, R., 2009. Bad luck versus self-inflicted neediness—anexperimental investigation of gift giving in a solidarity game. J. Econ. Psychol.30, 517–526.