University of Massachusetts AmherstScholarWorks@UMass AmherstResource Economics Department FacultyPublication Series Resource Economics

2002

Economic inequality and burden-sharing in theprovision of local environmental qualityJC Cardenas

J Stranlund

C Willis

Follow this and additional works at: https://scholarworks.umass.edu/resec_faculty_pubs

This Article is brought to you for free and open access by the Resource Economics at ScholarWorks@UMass Amherst. It has been accepted forinclusion in Resource Economics Department Faculty Publication Series by an authorized administrator of ScholarWorks@UMass Amherst. For moreinformation, please contact scholarworks@library.umass.edu.

Recommended CitationCardenas, JC; Stranlund, J; and Willis, C, "Economic inequality and burden-sharing in the provision of local environmental quality"(2002). Ecological Economics. 191.Retrieved from https://scholarworks.umass.edu/resec_faculty_pubs/191

ANALYSIS

Economic inequality and burden-sharing in the provision oflocal environmental quality

Juan Camilo Cardenas a, John Stranlund b,*, Cleve Willis c

a Facultad de Estudios Ambientales y Rurales, Uni�ersidad Ja�eriana, Trans 4 c 42-00 (Piso 8), Bogota, Colombiab Department of Resource Economics, 214 Stockbridge Hall, Uni�ersity of Massachusetts-Amherst, 80 Campus Center Way,

Amherst, MA 01003-9246, USAc College of Food and Natural Resources, Stockbridge Hall, Uni�ersity of Massachusetts-Amherst, Amherst, MA 01003, USA

Received 16 July 2001; received in revised form 7 December 2001; accepted 13 December 2001

Abstract

A large, but inconclusive, literature addresses how economic heterogeneity affects the use of local resources andlocal environmental quality. One line of thought, which derives from Nash equilibrium provision of public goods,suggests that in contexts in which individual actions degrade local environmental quality, wealthier people in acommunity will tend to do more to protect environmental quality. In this paper we report on experiments performedin rural Colombia that were designed to explore the role that economic inequality plays in the ‘provision’ of localenvironmental quality. Subjects were asked to decide how much time to devote to collecting firewood from a localforest, which degrades local water quality, and how much to unrelated pursuits. Economic heterogeneity wasintroduced by varying the private returns to these alternative pursuits. Consistent with the Nash equilibriumprediction, we found that the players with more valuable alternative options put less pressure on local water quality.However, the subjects with less valuable alternative options showed significantly more restraint relative to their pureNash strategies. Furthermore, they were willing to bear significantly greater opportunity costs to move their groupsto outcomes that yielded higher average payoffs and better water quality than the Nash equilibrium outcome. © 2002Elsevier Science B.V. All rights reserved.

Keywords: Local environmental quality; Burden-sharing; Economic inequality; Experiments

1. Introduction

Many rural communities in the developingworld depend to a large degree on access to localnatural resources and on local environmentalquality. A large body of literature has emergedthat addresses how economic heterogeneity—in-

* Corresponding author. Tel.: +1-413-545-6328; fax: +1-413-545-5853.

E-mail address: stranlund@resecon.umass.edu (J. Stran-lund).

PII: S0921-80 09 (01 )00285 -3

equality of wealth, income, or economic opportu-nity within a community—affects the use of localresources and environmental quality. Taken as awhole, this literature is inconclusive (Varugheseand Ostrom, 2001; Bardhan and Dayton-Johnson,2002). One line of thought, however, originateswith Olson’s (Olson, 1965) well-known hypothesisthat wealthier people in a group will tend to takeon a larger share of providing a public good thantheir poorer counterparts. Refining Olson’s hy-pothesis, Bergstrom et al. (1986) suggested thatthis effect is enhanced by greater income inequal-ity.1 These results suggest the hypothesis that incontexts in which individual actions degrade localenvironmental quality, the wealthier members ofthe community will do more to protect localenvironmental resources than their poorer neigh-bors because they will limit their damaging ac-tions to a greater extent. Greater inequalitysuggests a further shift of the burden of environ-mental protection to the wealthier members of acommunity.2

The theoretical predictions of how income in-equality affects the provision of public goods aregenerated from models of Nash non-cooperativebehavior; that is, using the standard assumptionof purely self-interested strategic behavior. How-ever, we know from a wealth of experimentalliterature that subjects in experiments involvingpublic goods do not typically play purely self-in-terested strategies.3 Rather, they make choices

that seem to balance pure self-interests againstgroup interests. Consequently, groups of individu-als in public good environments tend to achievemore efficient outcomes than Nash equilibriumoutcomes.

In this paper we report on experiments wedesigned to explore the role that economic in-equality plays in the ‘provision’ of local environ-mental quality in the developing world. Ratherthan test Nash equilibrium predictions about howeconomic inequality affects local environmentalquality, we explore burden sharing in achievingbetter-than-Nash equilibrium outcomes. Specifi-cally, we ask the question of whether the richermembers of a community do more than theirpoorer neighbors to help the community escapethe adverse welfare and environmental effects ofpurely self-interested strategic behavior, orwhether it is the other way around.

Our experiments were performed in three ruralareas of Colombia. The areas were chosen be-cause villagers in each region have significantinterests in local natural resources and environ-mental quality. In fact, the experiments were de-signed to approximate an environmental qualityproblem that rural villagers in developing coun-tries are likely to face. Subjects were asked todecide how much time to devote to collectingfirewood from a local forest, and how much todevote to an alternative market pursuit such aswage labor or farming one’s own land. Time spentcollecting firewood has adverse environmentalconsequences because it leads to soil erosion andworsened local water quality due to increasedsedimentation. The public goods aspect of theproblem is that individual restraint in firewoodcollection improves water quality, which is abenefit shared by all in the community.4

1 They also show that an income redistribution from poorerindividuals to wealthier individuals will lead to increased pro-vision of a public good if the redistribution leads more of thepoorer members of a community to choose to freeride com-pletely on the contributions of their richer neighbors. Experi-mental tests of this hypothesis by Chan et al. (1996) offer somesupport. On the other hand, Bergstrom, Blume and Varianpredicted that a small redistribution of income—small enoughso that the set of contributors to a public good is notchanged—would not affect aggregate contributions. Chan etal. (1999) test and reject this hypothesis.

2 Sandler (1992) notes that the link between income inequal-ity and the provision of a public good proposed by Olson andBergstrom, Blume and Varian is sensitive to assumptionsabout preferences for the public good, the technology thatcharacterizes provision of the good, and the exact form ofstrategic interaction.

3 For a review of this literature, see Ledyard (1995).

4 While we chose to place our subjects in a public goodenvironment, many local resource dilemmas in the developingworld are better characterized as common pool resource prob-lems. Since public good models are structurally different fromcommon pool models, one should be cautious about using ourresults about burden-sharing in the provision of local environ-mental quality to draw conclusions about local use of acommon pool resource. Further research that extends ourapproach to examine burden-sharing in the exploitation ofcommon pool resources will likely yield valuable insights.

While subjects faced symmetric private returnsfrom time spent collecting firewood and symmet-ric shared costs from water quality degradation,economic heterogeneity was introduced by vary-ing the private returns to the alternative marketpursuit.5 Two treatments of the experiment wereadministered—one in which the subjects faceddifferent returns in the market, and the other inwhich subjects faced symmetric returns to thisalternative. The model that we used to generatethe payoffs for the experiments was constructedso that average returns to the market alternative(per unit of effort) are the same in both thesymmetric and asymmetric treatments. More im-portantly, the Nash equilibria of the two treat-ments yield the same aggregate amount of timespent collecting firewood, and hence, local waterquality. However, individual Nash equilibriumchoices differ in the two treatments. Consistentwith the prediction of Olson and Bergstrom,Blume and Varian, individual equilibrium choicesin the symmetric treatment are also symmetric,but in the equilibrium of the asymmetric game,the high-wage players— those with more valuablealternative market options—are predicted toshow more restraint in exploiting the local forest,thereby doing more to protect local water qualitythan their less advantaged neighbors.6

Each group of subjects played a number ofrounds of the game without being able to commu-nicate with others in their group, and then contin-ued to play additional rounds in which they wereallowed to engage in non-binding and non-threat-

ening communication between rounds. Allowingsubjects to communicate with each other wasmotivated by the fact that many local environ-mental and natural resource dilemmas are ad-dressed through cooperative efforts by localresidents (Ostrom, 2000). Moreover, an over-whelming amount of experimental evidence sug-gests that face-to-face communication is effectivein enhancing cooperation in experiments of thistype (Hackett et al., 1994; Ostrom et al., 1994;Ledyard, 1995).

The results of our experiments yield intriguinginsights about burden-sharing in the provision oflocal environmental quality. As predicted by theequilibrium of our asymmetric game, the high-wage players spent less time harvesting firewoodthan their poorer counterparts, thereby puttingless pressure on local water quality. As expected,however, the subjects did not choose pure Nashstrategies. Across the board the groups achievedoutcomes with higher average payoffs and betterwater quality than the equilibrium predictions,thereby setting the stage for exploring burden-sharing in achieving these more efficient out-comes. Our results from this exploration areunequivocal. By calculating the differences be-tween individuals’ actual choices and their Nashbest-responses, we found that although the high-wage players put less pressure on local environ-mental quality, their choices ultimately were quiteclose to their Nash best-responses. On the otherhand, the low-wage players showed significantlymore restraint relative to their Nash best-re-sponses. Thus, the restraint necessary for thesegroups to achieve better-than-Nash equilibriumoutcomes came largely from the low-wage sub-jects. Furthermore, by calculating individuals’foregone payoffs from not playing pure Nashstrategies, we found that the low-wage playerswere, on average, willing to bear significantlygreater opportunity costs than their richer coun-terparts to help their groups achieve better-than-Nash equilibrium outcomes.

Face-to-face communication only enhanced thewillingness of the low-wage subjects to bear thecosts of moving their groups to more efficientoutcomes. In fact, this was the main effect ofcommunication, since through these rounds of the

5 In real settings, these returns may vary for a number ofreasons, including unequal land ownership, access to credit, ordifferences in education levels. Of course, there are otherdimensions of economic inequality. The typical public goodsmodel would focus on differences in exogenous income(Bergstrom et al., 1986; Chan et al., 1996, 1999). Others havefocused on asset inequality (Bardhan and Dayton-Johnson,2002). For a set of common pool experiments, Hackett et al.(1994) vary the exogenous endowment of an input acrossindividuals. Whether our results will hold along other dimen-sions of economic inequality is a question for future research.

6 One should be clear that we are not proposing this resultas a general statement about Nash equilibrium provision ofenvironmental quality in the developing world, only that it isconsistent with one line of thought in the literature.

experiment the high-wage subjects stuck close totheir Nash best-responses. Not surprisingly, thesegroups fared much better when they were allowedto communicate between rounds than when theywere not.

Not only do our results suggest that economicasymmetry plays an important role in the mannerin which groups share the burden of achievingbetter-than-Nash equilibrium outcomes, they alsosuggest that heterogeneity may have an importantrole in determining aggregate outcomes that gounrecognized in models of purely self-interestedprovision of a public good. Even though aggre-gate levels of exploitation of the local forests werepredicted to be the same for both treatments, thegroups with unequal market wages put signifi-cantly less pressure on local environmental qualitythan the groups with symmetric market returns.While the subjects in our symmetric treatmentalso did not choose pure Nash strategies, andachieved better outcomes in doing so, on averagethey did not assume as large a burden of movingtheir groups to more efficient outcomes as thelow-wage subjects in the asymmetric treatmentdid for their groups. Consequently, the heteroge-neous groups consistently enjoyed higher levels ofenvironmental quality than their homogeneouscounterparts.

2. Experimental design

As noted in the introduction, we designed ourexperiments to confront our subjects with a localenvironmental quality problem that would closelymimic their actual experiences. Toward that end,our field experiments were undertaken in threerural areas in Colombia where villagers have sig-nificant interests in local natural resources andenvironmental quality. Payoffs for the games weregenerated from a model of individual efforts tocollect firewood from local forests. Higher levelsof effort devoted to firewood extraction leads togreater erosion and sedimentation of local watersupplies; thus, restraint in harvesting firewoodfrom local forests generates the pure public goodof better water quality.

2.1. The payoffs

A simple model of a fixed number of individu-als who exploit a local forest for firewood gener-ated the payoffs for our experiments. In eachround of the games, each individual was given anendowment of time e, of which all or part couldbe allocated to collecting firewood. Let xi denotethe amount of time individual i spends collectingfirewood. This effort yields a private benefit,which we assume takes the quadratic formg(xi)=�xi−�(xi)2/2, where � and � are strictlypositive and are chosen in part to guaranteeg(xi)�0, for xi� [1, e ]. The strict concavity ofg(xi) indicates diminishing marginal private re-turns to time spent collecting firewood.

Alternatively, the individual could allocate hisor her labor to an alternative market pursuit suchas wage labor or farming one’s land. Let wi

denote the individual’s return to time devoted tothe market alternative. Henceforth, we refer tothis return as the market wage. In one experimen-tal treatment this wage will vary among individu-als in a group, while in the other treatment allindividuals in a group will face the same marketwage. Note that i ’s decision to provide (e−xi)units of labor to the market alternative yields apayoff of wi× (e−xi).

Subjects were told explicitly that their decisionto spend time extracting firewood would affectwater quality in the area adversely. We assumedthat water quality q is a quadratic function of theaggregate amount of time individuals in the com-munity spend collecting firewood; specifically,q(� xj)=q0− (� xj)2/2, where q0 is interpreted tobe water quality in the absence of firewood ex-traction. The value of q0 was chosen, in part, toguarantee q(� xj)�0 for all feasible � xj.

Define ui(xi, � xj)=q(� xj)+g(xi)+wi× (e−xi). Parameters were chosen, in part, to guaranteethat ui(xi, � xj)�0 for all possible xi and � xj. Tofacilitate scaling individual payoffs, we take anindividual’s payoff function to be a positive,monotonic transformation F of u. In particular,F(ui)=k× (ui)�, where k and � are positive con-stants. An individual’s payoff function is then,

Ui(xi, � xj)=k× [ f(� xj)+g(xi)+w× (e−xi)]�

=k��

q0−(� xj)2

2

�+�

�xi−�(xi)2

2�

+wi× (e−xi)��

. (1)

Each group consisted of n=8 subjects, andeach subject was allocated e=8 units of time ineach round. Pre-testing of the experimental de-signs at the University of Massachusetts-Amherstand the Humboldt Institute for Biodiversity inVilla de Leyva, Colombia, led us to denominateunits of time as months per year. Scale concernsled us to choose the following remaining parame-ter values: k=4/16 810, q0=1372.8, �=97.2;�=3.2, and �=2. Therefore:

By varying the market wage rate, we generatedtwo treatments of the experiment. For the base-line case we assumed that wi=30 for each of theeight individuals in a symmetric-wage group. Forthe other treatment we assigned a lower marketwage, wi=20, to six members of a asymmetric-wage group, and a higher wage, wi=60, for theother two members. Note that average marketwages are the same for the symmetric-wage andasymmetric-wage groups.

Subjects were given a table of payoffs generatedby Eq. (2). Table 1 is a slightly shortened versionof the payoff table given to each of the subjects inthe symmetric market wage treatment (the bottomseven rows are deleted here to conserve space. Theshaded cells indicate an individual’s pure Nashstrategy. The subjects received complete tableswithout the shading). Tables 2 and 3 are short-ened versions of the payoff tables given to thehigh-wage and low-wage subjects, respectively, inthe asymmetric-wage treatment.

2.2. Nash strategies and the balance betweenself-interested and group-oriented beha�ior

The treatments were designed to generate thesame aggregate amount of time devoted to fire-

wood harvesting in the two Nash equilibria, al-though individual equilibrium choices would bedifferent. The Nash equilibrium of the symmetricmarket wage game is for each of the eight playersto spend 6 months collecting firewood. In thisequilibrium, each subject receives 155 points. Inthe equilibrium of the asymmetric market wagegame, the high-wage subjects are predicted to

spend no time collecting firewood, choosing in-stead to devote all of their time to the alternativemarket pursuit. On the other hand, the low-wagesubjects are predicted to devote all of their en-dowment of 8 months to collecting firewood.7

Consistent with the predictions of Olson (1965)and Bergstrom et al. (1986), in the asymmetric-

Ui(xi, � xj)=� 4

16810��

1372.8−(� xj)2

2+97.2xi−

3.2(xi)2

2+wi× (8−xi)

�2

. (2)

7 One may verify that these choices constitute Nash equi-libria from the payoff tables. Recall that in a Nash equilibriumeach player’s choice has to be a best-response to all the otherplayers’ choices. Consider first the symmetric market wagetreatment in which each player is predicted to spend 6 monthscollecting firewood. From Table 1, if each of seven of the eightplayers chooses 6 months for a total of 42, then the best-re-sponse of the eighth player is to choose 6 months. Nowconsider the equilibrium of the asymmetric-wage treatment inwhich the high wage players choose to spend no time collect-ing firewood, while each of the low-wage players devote all oftheir 8 months to collecting firewood. Using Table 2— thepayoffs for a high-wage player in this treatment—note that inthe proposed equilibrium the choices of all low-wage playersand the other high-wage player sum to 48 months. A high-wage player’s best-response is, therefore, to spend no timecollecting firewood. For a low-wage player, the sum of theother players’ choices in the proposed equilibrium is 40months. From Table 3, a low-wage player’s best-response is tospend 8 months collecting firewood.

Table 1Individual payoffs and Nash responses for symmetric-wage subjects

Table 2Individual payoffs and Nash responses for high-wage subjects in asymmetric-wage treatment

Table 3Individual payoffs and Nash responses for low-wage subjects in asymmetric-wage treatment

wage treatment the high-wage subjects are ex-pected to take on a disproportionate burden ofprotecting local water quality. In this asymmetricequilibrium, low-wage subjects receive 191 points,while the high-wage subjects receive only 117points. In both the symmetric and asymmetric-wage treatments, aggregate equilibrium time spentcollecting firewood is 48 months.8

Although Nash equilibrium choices are a stan-dard benchmark for games of this type, we aremore interested in analyzing off-equilibriumchoices and comparing these to individuals’ pureNash strategies; that is, their payoff-maximizingchoices, given the choices of others in their group.In fact, we take the difference between an indi-

vidual’s Nash best-response to the choices of theother subjects in his or her group and his or heractual choices to be an indicator of how thatperson balances self-interests against the interestsof the entire group.

To illustrate our approach, consider the payoffsfor an individual in our symmetric-wage treat-ment provided in Table 1. Recall that each exper-iment involved eight participants. Suppose thateach of seven players chooses to spend 2 monthscollecting firewood from local forests. Since thesesubjects are choosing to spend 14 months in totalcollecting firewood, Table 1 indicates that theeighth player’s Nash best-response is to spend 8months collecting firewood. This choice is madepurely out of self-interest without regard for thewelfare of the others in the group. Note thatplayer eight’s payoff in this outcome is 776 points,while each of the other seven receive 535 points

(each player chooses 2 months, while the sum ofthe others’ choices is 20 months).

Now imagine that the eighth player chooses 3months instead of 8, while the other seven sub-jects continue to choose 2 months. This is asignificantly more group-oriented choice— it iscostly because the eighth player’s payoff is now652 points instead of 776; however, each of theother players’ payoffs increase from 535 points to606 (each continues to choose 2 months, but thesum of the others’ choices is now 15 months).With a choice of 3 months instead of 8, the eighthplayer bears a personal opportunity cost of 124points to increase the payoffs to the others in thegroup by 497 points! Much of our analysis of theexperimental data in the next section is basedupon the differences between the subjects’ actualchoices and their Nash best-responses: choicesthat are close to Nash responses indicate relativelyself-interested behavior, while those that are fur-ther away indicate stronger group-regardingbehavior.

2.3. The subjects and experiments

As noted before, we intended our experimentsto confront subjects with an environmental prob-lem that closely mimicked their actual experi-ences. Our field experiments were undertaken inthree rural areas in Colombia. In the village ofEncino, located in the eastern Andean region,residents extract firewood and log timber on asmall scale in local tropical cloud forests. Waterfor consumption and irrigation comes nearly un-treated from local rivers. Of the three areas thatwe visited, the relationship between forest coverand water quality is most critical in Encino, andthe residents of this village are acutely aware ofthe problem. Water quality degradation caused byforest cover losses is less severe in the villages ofCircasia and Filandia in the Quindio coffee regionin the mid-Andes, but nevertheless is a significantproblem. In Quindio, subjects for our experimentswere drawn specifically from a group of familieswhose livelihood is related to the extraction andprocessing of natural fibers from local forests. Asin Encino, water is drawn from local rivers, andresidents are aware that extracting forest products

8 The other common benchmark for games of this type is thewelfare maximizing outcomes. Although this benchmark is notvery useful for our purposes, it is interesting to note how poorlythe subjects fare in the Nash equilibria, relative to what ispossible. It is easy to show that in the efficient outcome of thesymmetric-wage game, each player would choose to spend only1 month harvesting firewood and would receive a payoff of 645points. This is more than four times greater than the Nashequilibrium payoffs. Efficiency in the asymmetric-wage treat-ment would require the high-wage players to forego spendingtime collecting firewood, and the low-wage players to spend only1 month collecting firewood. In this outcome the high-wageplayers would receive 801 points, about 6.8 time greater thantheir Nash equilibrium payoffs, and the low-wage players wouldreceive 602 points, or just over three times their Nash payoffs.

can lead to lower water quality. In Nuqui, locatedon the Pacific coast, villagers harvest coastal man-groves for firewood and other wood products, buttheir water comes from further inland; hence, theydo not experience a direct link between theirexploitation of local sources of wood and waterquality. However, they face a similar dilemmabecause their exploitation of the mangroves forwood adversely affects coastal fish populationsupon which also they depend.

In total, 120 subjects participated in the experi-ments. The subjects were distributed into 15 groups,10 of which played the symmetric-wage treatment,and five the asymmetric-wage treatment. In the caseof the asymmetric-wage treatment, the assignmentof high-wage and low-wage payoff tables was donerandomly. In each of the three settings, the partic-ipants generally knew each other well, having livedin the same village for most of their lives (we avoidedhaving close relatives in the same group). Schooling,age and income levels varied significantly withineach group. Most participants had fewer than 6years of schooling, roughly half were between 30and 50 years old, and all were 16 or older.

Each session of the experiment involved eightsubjects and two monitors. The subjects sat atindividual desks that were distributed in a circlewith enough separation between the desks so theycould not look at another’s work. Except in periodswhen communication was allowed, the desks facedaway from the center of the circle. In each round,each subject would choose how many units of time,between zero and eight, to spend collecting fire-wood. Subjects were given a payoff table and largeposters of these tables were placed on a wall of the‘field lab’. In the symmetric-wage treatment, indi-viduals knew that the other participants consultedthe same table. In the asymmetric-wage treatment,individuals knew how many high-wage and low-wage players there were, but not which individuals,other than themselves, were low-wage and high-wage players. Although individuals could not knowin advance the decisions of others, they knew thepayoffs upon which their decisions were based.Once a subject made a decision for a particularround, this decision was written on a slip of paperin private. When all subjects had made theirdecisions, a monitor collected each slip of paper and

gave them to another monitor who recorded thedecisions and calculated the total for the group. Thistotal was announced to the subjects, who thendetermined their own payoffs from their payofftables. Subjects kept a record of their own payoffsas a check on the monitor’s record.

Each session began with welcoming remarkswithin which the subjects were told that the sessionwould last approximately 2 hours. A monitor thenread the instructions to the participants (the instruc-tions are available from the authors). Results frompre-tests of the experiment led us to decide not togive the subjects written instructions because of thewide variation in levels of literacy. The instructionsexplained the basic setting of the game, how pointswere earned, how these points were converted tocash at the end of the session, and the proceduresof the game. The instructions included three differ-ent examples to familiarize the subjects with thepayoffs and the procedures. Two practice roundswere conducted. The monitor asked for questionsat several points, and when there were no furtherquestions the game began with round 1. In additionto posters of the payoff tables, large posters of theforms the subjects used during the game and theexamples from the instructions were placed on onewall of the ‘field lab’.

Each of the 15 groups played 8–11 initial roundsof the game without being allowed to communicatewith the others in their group or with the monitors.The subjects were not told how many rounds wouldbe played in the initial stage, or that the rules forcommunication would be changed at some point.After the initial rounds the monitors stopped thegame and announced a new set of rules for theforthcoming rounds. A monitor read from a newposter, which was subsequently placed on the wall,announcing that from now on the individuals in agroup would be allowed 5 minutes of discussionbetween rounds. Their discussions could be aboutanything, but they could not threaten each other oragree to transfers of cash or points after the session.Between rounds the subjects turned their desks toface each other. Once 5 minutes elapsed, the subjectswere required to turn their desks back around tomake their decisions in private. The groups playedin this way for another 9–12 rounds, again withoutknowing when the final round would be.

Fig. 1. Average months harvesting firewood.

At the end of each session, total points for eachindividual were calculated, and they were paid thatnumber in pesos for their participation. Eachindividual also received a gift (a common house-hold item such as a lamp, table set, machete, etc.)for participation, which was not related to theirplay during the experiments. For the villages inwhich the experiments were conducted, a dailyminimum wage centered around 7000 pesos (aboutUS $5.40 at the time). The average payoff (includ-ing payoffs for practice rounds) was, as planned,about 1.5 days of work at the local minimum wage.

3. Results

We begin the analysis of the experimental databy considering average choices of time spent har-vesting firewood. Fig. 1 and the first block ofrows in Table 4 summarize these decisions. Asindicated earlier, we ended the first and secondstages at different points so that the subjectscould not anticipate the terminal rounds. Allgroups played 8–11 rounds in the first stage,within which they were not able to communicatewith each other. We, therefore, consider only thefirst eight rounds of first-stage decisions for eachgroup. All groups played nine rounds in the sec-ond stage in which they were allowed to commu-nicate with each other between rounds, and somea few more; therefore, we consider only the firstnine rounds of second-stage decisions for eachgroup.

Let us focus first on the high-wage and low-wage players’ choices in the asymmetric payofftreatment. After the initial round in which aver-

age choices for both types of players were between4 and 5 months harvesting firewood, there is alarge separation of average time spent harvestingfirewood. As predicted, the high-wage playersspent less time harvesting firewood than the low-wage players. In the first three rounds of theno-communication stage the average choice of thehigh-wage players was 3.667, while for the low-wage players the average was 4.767. This differ-ence is statistically significant (P-value=0.040from the Wilcoxon–Mann–Whitney rank sumtest— this test is used for comparing meansthroughout the analysis). Both types of subjectsreduced their amounts of time collecting firewoodas the experiment proceeded through the no-com-munication stage. The high-wage subjects reducedtheir average time collecting firewood from 3.667in the first three rounds of the no-communicationstage to 2.462 in the last three rounds (P-value=0.084), while the low-wage subjects reduced theiraverage time in the forest from 4.767 in the firstthree rounds to 4.128 in the last three rounds(P-value=0.035). Notice that high-wage playerscontinued to exploit the local forest to a lesserdegree than low-wage players in the last threerounds of the first stage. The difference betweentheir average choices, 2.462 for the high-wagesubjects as compared with 4.767 for the low-wagesubjects, remained statistically significant (P-value=0.004).

As expected, the players put significantly lesspressure on local environmental quality when theywere allowed to communicate between rounds inthe second stage. However, the main effect ofcommunication was to induce the low-wage play-ers to reduce their harvesting of firewood. From

Table 4Summary data

Communication roundsNo-communication rounds

Rounds 6–8Rounds 1–3 Rounds 1–3 Rounds 7–9

Months in the forestAsymmetric payoffs

2.462 2.100High wage 2.4673.6674.128 3.2784.767 3.044Low wage

Combined 3.7124.492 2.983 2.9004.388 3.783 3.6164.388Symmetric payoffs

EarningsAsymmetric payoffs

$522.08 $618.80$406.63 $634.37High wage$447.86 $503.44Low wage $498.39$347.71$466.41 $532.28$362.44 $532.38Combined$371.41Symmetric Payoffs $444.47$368.17 $460.14

De�iations from best-responsesAsymmetric payoffs

−2.300High wage −0.192 1.467 1.4003.872 4.7223.233 4.956Low wage3.224Symmetric payoffs 3.9753.146 4.223

Opportunity costs of non-Nash choicesAsymmetric payoffs

$6.85 $16.30 $18.10High wage $18.57$74.42 $125.76$44.68 $138.99Low wage$33.49 $58.00 $65.50Symmetric payoffs $35.61

the last three rounds of the no-communicationstage into the first three rounds of the communi-cation stage, the low-wage players reduced theiraverage time spent harvesting firewood from4.128 to 3.278 (P-value=0.035) and a bit furtherto 3.044 for the last three rounds of the communi-cation stage. On the other hand, the high-wageplayers reduced their time spent harvesting fire-wood only slightly from 2.462 in the last threerounds of the no-communication stage to 2.100 inthe first three rounds of the communication stage.This difference is both small and statistically in-significant (P-value=0.492). In the last threerounds of the communication stage, the high-wage players’ average choices edged back up tothe same level as their average choices in the lastthree rounds in the no-communication stage.

As the asymmetric-wage subjects reduced theirtime spent harvesting firewood, their averageearnings improved. The second block of rows ofTable 4 indicates that in the no-communication

stage the average earnings of the high-wage sub-jects rose from about $407 in the first threerounds to about $522 in the last three rounds(P-value=0.005), while the earnings of the low-wage subjects rose from about $348 to about $448(P-value=0.000). This continued into the com-munication stage. Average earnings for the high-wage players rose from $522 in the last threerounds of the no-communication stage to nearly$619 in the first three rounds of the communica-tion stage (P-value=0.010), and to over $634 inthe last three rounds of this stage. The same thingoccurred for the low-wage players, but at lowerlevels. Note that the subjects’ earnings were muchhigher than the Nash equilibrium earnings of $117per round for each high-wage player and $191 perround for each low-wage player.

Now let us compare the performance of theasymmetric-wage groups to that of the symmetric-wage groups. Even though the payoffs were con-structed so that the Nash equilibria of the

Fig. 2. Average deviations from Nash best-responses.

asymmetric and symmetric-wage treatments yieldsthe same level of aggregate time harvesting fire-wood, and hence local water quality, on averagethe asymmetric-wage groups put less pressure onthe local forest. The first block of rows of Table 4and Fig. 1 indicate that after starting out at aboutthe same levels, by the last three rounds of theno-communication stage the combined averagechoices of the asymmetric-wage groups fell belowthose of the symmetric-wage groups. In the lastthree rounds of the no-communication stage theasymmetric-wage subjects were averaging 3.712units of time harvesting firewood, while the sym-metric-wage subjects were averaging 4.388 monthsharvesting firewood (P-value=0.018).

Like the subjects in the asymmetric-wage treat-ment, the symmetric-wage subjects reduced theirexploitation of the forest when they were allowedto communicate with each other. Their averagechoices fell from 4.388 for the last three rounds ofthe no-communication stage to 3.616 for the lastthree rounds of the communication stage (P-value=0.001). However, they continued to ex-ploit the forest to a greater degree, and hence,experience lower water quality, than their asym-metric-wage counterparts. By the last rounds ofthe communication stage, the average choices ofthe asymmetric-wage subjects were 2.900 as com-pared with 3.616 for the symmetric-wage subjects(P-value=0.0173). Clearly, economic heterogene-ity played a role here in producing better environ-mental outcomes, even though the two treatmentswere constructed to produce the same Nash equi-librium levels of water quality.

Although the snapshot provided by analyzingaverage choices is illuminating, it does not give us

a complete picture of how economic inequalityaffects burden-sharing in achieving the outcomesthat are consistently better than Nash equilibriumoutcomes. For this we analyze the average devia-tions of the participants’ decisions from their indi-vidual Nash best-responses in each round. Anindividual’s Nash best-response in a particularround is his or her payoff-maximizing choicegiven the actual aggregate choices of the rest ofthe group in that round. We then calculated thedifference between each individual’s best-responseand their actual choice. These deviations fromNash best-responses are summarized in Fig. 2 andin the third block of rows in Table 4. Note thatvalues close to zero indicate that players madechoices that were, on average, close to pure Nashstrategies, while positive values indicate averagechoices of time spent harvesting firewood thatwere less than individual best-responses. Thesepositive deviations indicate more group-orientedchoices than purely self-interested Nash strategies,and hence, a greater willingness to accept part ofthe burden of achieving more efficient outcomesthan the Nash equilibrium outcomes.9

Focusing first on the high-wage and low-wageplayers’ choices in the asymmetric payoff treat-ment, note the large separation between the aver-age deviations from Nash best-responses of thetwo types of players throughout the sessions(these differences are always statistically signifi-

9 One could argue that individual deviations from pureNash strategies could be explained, at least in part, by subjectshaving a difficult time locating their best-responses. If this wasthe sole reason for these deviations, we would expect thataverage deviations would consistently be statistically indistin-guishable from zero, which is not what we observe.

cant). In the first five rounds or so of the first stage,the high-wage players’ average deviations fromtheir best-responses are negative, indicating thatthey actually spent more time harvesting firewoodthan suggested by their Nash strategies. Thesechoices are very inefficient— they are not individu-ally optimal, and they are more environmentallydamaging than their Nash strategies. As the high-wage players realized the inefficiency of theirchoices, they moved to their purely self-in-terested Nash best-responses in the last threerounds of the first stage. The low-wage players, onthe other hand, made choices throughout the firststage that were consistently well below their Nashbest-responses. By the last three rounds of theno-communication stage, the low-wage playerswere making choices, on average, that were nearly4 units of time lower than their self-interested Nashstrategies would indicate.

Both types of players made significant movestoward achieving better-than-Nash equilibriumoutcomes when they were allowed to communicatewith each other. The high-wage players’ deviationsfrom their Nash best-responses increased fromabout zero (−0.192) in the last three rounds of theno-communication stage to 1.467 in the first threerounds of the communication stage (P-value=0.057). Their average deviations from best-re-sponses decreased slightly to 1.400 in the last threerounds of the communication stage, but the differ-ence between the first three rounds and the lastthree rounds of this stage is not statistically signifi-cant. Similarly, the low-wage players’ deviationsfrom their best-responses rose from 3.872 in thelast three rounds of the no-communication stage to4.772 in the first three rounds when communica-tion was allowed (P-value=0.035), and remainedat about that level through the rest of that stage.

Our comparison of the subjects’ actual choicesto their purely self-interested Nash strategies yieldsa very clear message. Even though the richerplayers put less pressure on the environment thanthe poorer players, the poorer players madechoices that were significantly more group-ori-ented, thus taking on the primary responsibility forachieving outcomes that were consistently moreefficient than the Nash equilibrium outcome.

The symmetric-wage subjects also made choices

that were, on average, more group-oriented thantheir pure Nash strategies. Their choices averagedabout 3.224 units of time lower than their Nashbest-responses in the last three rounds of theno-communication stage, and fell to 4.223 unitslower by rounds 7–9 of the communication stage(P-value=0.000). Though the symmetric-wageplayers made choices that were significantly moregroup-oriented than their pure Nash strategies,they did not do so to the same extent as thelow-wage subjects in the asymmetric-wage treat-ment, particularly in the communication rounds.Average deviations from best-responses for thesymmetric-wage subjects in the first three rounds ofthe communication stage were about 3.98, whilefor the low-wage subjects in asymmetric-wagetreatment they were about 4.722 (P-value=0.021).This separation continued for rounds 7–9 in thecommunication stage—4.223 for the symmetric-wage subjects and 4.956 for the low-wage subjectsin the asymmetric-wage treatment (P-value=0.033). While these differences are not very large,the fact that the low-wage subjects in the asymmet-ric-wage groups were willing to take on moreresponsibility for moving their groups to moreefficient outcomes than their symmetric-wagecounterparts is the primary reason the asymmetric-wage groups were able to attain higher levels ofenvironmental quality.

Another way of looking at the distribution of theburden of achieving better-than-Nash equilibriumoutcomes is to examine what it costs individuals toshow restraint in harvesting firewood relative totheir Nash best-responses. Toward this end, foreach individual in each round we calculated thedifference between the individual’s actual payoffand what they would have earned had they playedtheir Nash best-responses instead. This calculationindicates an individual’s personal opportunitycosts of not making a purely self-interested choice,and hence, the costs they are willing to bear to helptheir group achieve more efficient outcomes. Theseopportunity costs are summarized in Fig. 3 and thelast block of rows in Table 4.

This exercise simply reinforces our finding thatthe asymmetric-wage groups were able to reachand sustain better-than-Nash equilibrium out-comes largely because the low-wage subjects were

Fig. 3. Average opportunity costs of non-Nash choices.

willing to take on the burden of doing so. In thelast three rounds of the no-communication stage,the low-wage subjects were bearing opportunitycosts of not making purely self-interested choicesthat averaged $74.42 per subject per round, whilethe opportunity costs borne by the high-wageplayers were merely $6.85 (P-value=0.0001). Inthe communication stage the difference is moredramatic—$125.76 for the low-wage players inthe first three rounds of this stage as comparedwith $16.30 for the high-wage subjects (P-value=0.0001), and about $140 versus $18 for rounds7–9 of this stage (P-value=0.0001).

Recall that from the perspective of the devia-tions from Nash best-responses, the willingness ofthe low-wage players in the asymmetric-wagegroups to make more group-oriented choices thantheir symmetric-wage counterparts does not ap-pear to be that great. However, from the perspec-tive of the opportunity costs borne to make moregroup-oriented choices, the low-wage players inthe asymmetric-groups showed significantlygreater willingness to bear the costs of movingtheir groups to better-than-Nash equilibrium out-comes than the symmetric-wage subjects. Begin-ning in the last three rounds of theno-communication stage and on to the end of thesessions, the opportunity costs of making non-Nash choices borne by the low-wage players inthe asymmetric-groups were consistently morethan twice the opportunity costs borne by thesymmetric-wage subjects (these differences in thecosts of non-Nash choices are always statisticallysignificant: the P-values are always around0.0001). Again, not only were the low-wage play-ers willing to accept the greater burden of achiev-

ing better outcomes than the high-wage subjectsin their own groups, but they also did so to agreater extent than the subjects in the symmetric-wage treatment.

4. Concluding remarks

This study strongly suggests that any examina-tion of the effects of economic inequality on theprovision of shared environmental quality in ruralcommunities of the developing world should rec-ognize two modes of burden-sharing. The first,which derives from hypotheses of Olson (1965)and Bergstrom et al. (1986), focuses on the ques-tion of whether it is the more- or less-advantagedmembers of a community who are likely to putless pressure on the local environment. Our mod-ification of this approach yielded the Nash equi-librium prediction that the richer members of acommunity— those with more valuable alterna-tive options in our construct—would put lesspressure on the local environment than their less-advantaged neighbors. Our experimental datalend strong support to this hypothesis.

However, a single-minded focus on this modeof burden-sharing misses an important point—subjects in public goods environments rarelychoose pure Nash strategies, and typically achievemore efficient outcomes in doing so. Thus, themore important contribution of this study is torecognize and analyze another mode of burden-sharing; that is, the question of which individualsare likely to bear the burden of moving theirgroup to these more efficient outcomes. Our re-sults concerning this mode of burden-sharing are

unequivocal. Even though the members of ourgroups with the more valuable alternative pursuitsput less pressure on the local environment, thosewith less valuable options took on significantlymore of the responsibility of moving their groupsto outcomes with higher payoffs and better envi-ronmental quality.

These results may have important policy implica-tions for whether and how governments shouldintervene in these contexts. A restriction on the useof a resource in order to improve local environmen-tal quality may not produce the intended outcomeif the policy is designed without regard for the waya community attacks the problem on its own. Forexample, a policy that focuses on restraining thebehavior of the less-advantaged in a communitybecause they tend to put more pressure on localenvironmental quality might be counterproductiveif doing so weakens their willingness to accept mostof the burden of moving their community to moreefficient outcomes. From the perspective of ourfindings, at the very least, such a policy appears tobe unfair. Even a policy that is construed to bemore fair, like a uniform quota, may turn out tobe ineffective if it causes the less advantaged tomake more self-interested choices.

Our results may also have implications for theeffects on local environmental quality of povertyalleviation programs. Many have argued that con-fronting rural poverty by enhancing the privatemarket opportunities of poorer individuals (withmore or better land, better access to credit, or moreeducation) could lessen the pressure they place onlocal natural resources and environmental quality(Durning, 1989; Leonard, 1985, 1989). However,our results suggest that those without good privatealternatives seek more cooperative solutions for theuse of shared natural resources and environmentalquality. Thus, government provision of better pri-vate opportunities to the rural poor may not havethe desired impact on local environmental qualityif these programs reduce the motivation of theseindividuals to make more efficient choices concern-ing their local environments. Further research isnecessary to explore this hypothesis more rigor-ously.

Since our analysis focused on out-of-equilibriumoutcomes, we had no a priori expectation of how

our heterogeneous groups of subjects would sharethe burden of achieving better-than-Nash equi-librium outcomes. But our results simply invite thequestions: What motivated the less-advantagedmembers of the heterogeneous groups to take ona disproportionate share of the burden of movingtheir groups to more efficient outcomes? From theother perspective, why did not the richer membersof these groups respond as the poorer members didand make more group-oriented choices? Perhapsthe answers lie in the establishment of behavioralnorms, or notions of equity in sharing the burden.It is possible that the high-wage members of thegroups, by making choices that put less pressure onthe environment, signaled a standard of restraintthat the low-wage subjects felt compelled to emu-late. It is also possible that the high-wage members,because they made less damaging choices, felt thatthey were doing their ‘fair share’ to protect theenvironment, and hence, were not motivated to domore than what was indicated by their purelyself-interested Nash strategies. As plausible as theseconjectures are, they probably do more to suggestfurther research than to explain how economicinequality affects burden-sharing in these contexts.

And further analyses are needed to check therobustness of our results in other contexts. As wenoted in the introduction, further research is neces-sary to check our results across different naturalresource and environmental dilemmas, as well asdifferent dimensions of economic inequality. Evenif contrary patterns emerge, we stand to gainvaluable lessons about how communities in thedeveloping world distribute the burden of con-fronting local environmental and natural resourcedilemmas, and about how public polices can helpthem do so more effectively.

Acknowledgements

Financial support for the field work was pro-vided by the MacArthur Foundation, the Institutode Investigacion de Recursos Biologicos Alexandervon Humboldt, the WWF Colombian program,and Fundacion Natura-Colombia. In the US, theresearch was supported by the Cooperative StateResearch, Extension, Education Service, US De-

partment of Agriculture, Massachusetts Agricul-tural Experiment Station, under Project No. 799,Manuscript No. 3298, and by a Resources for theFuture dissertation grant. We are indebted forhelpful advice and suggestions to Sam Bowles,James Walker, and Elinor Ostrom, and absolvethem from responsibility for any shortcomingsthat remain. In Colombia we must thank the fieldpractitioners and fellows from Humboldt, WWFand Fundacion Natura who helped pre-test andconduct the experiments. Very special thanks aredue Luis Guillermo Baptiste and Sara Hernandezat Humboldt, Carmen Candelo at WWF andJuan Gaviria, Nancy Vargas and Danilo Salar atNatura.

References

Bergstrom, T., Blume, L., Varian, H., 1986. On the privateprovision of public goods. J. Public Ecol. 29, 25–49.

Bardhan, P., Dayton-Johnson, J., 2002. Unequal irrigators:Heterogeneity and commons management in large-scalemultivariate research. In: National Research Council (Edi-tor), The drama of the commons. National Academy Press,Washington, DC.

Chan, K., Mestelman, S., Moir, R., Muller, R.A., 1996. Thevoluntary provision of public goods under varying incomedistributions. Can. J. Ecol. 29 (1), 54–69.

Chan, K., Mestelman, S., Moir, R., Muller, R.A., 1999.Heterogeneity and the voluntary provision of public goods.Exp. Ecol. 2, 5–30.

Durning, A.B., 1989. Poverty and the environment: reversingthe downward spiral. Worldwatch Paper No. 92. Washing-ton D.C.: The Worldwatch Institute.

Hackett, S., Schlager, E., Walker, J., 1994. The role of com-munication in resolving commons dilemmas: experimentalevidence with heterogeneous appropriators. J. Environ.Ecol. Manage. 27 (2), 99–126.

Leonard, H.J., 1985. Divesting Natures Capital—The PoliticalEconomy of Environmental Abuse in the Third World.Holmes and Meier, New York.

Leonard, H.J., 1989. Environment and the Poor: DevelopmentStrategies for a Common Agenda. Overseas DevelopmentCouncil, Washington DC.

Ledyard, J.O., 1995. Public goods: a survey of experimentalresearch. In: Kagel, J., Roth, A. (Eds.), Handbook ofExperimental Economics. Princeton University Press,Princeton.

Olson, M., 1965. The Logic of Collective Action. HarvardUniversity Press, Cambridge, MA.

Ostrom, E., 2000. Collective action and the evolution of socialnorms. J. Ecol. Perspect. 14 (3), 137–158.

Ostrom, E., Gardner, R., Walker, J., 1994. Rules, Games andCommon Pool Resources. University of Michigan Press,Ann Arbor.

Sandler, T., 1992. Collective Action: Theory and Applications.University of Michigan Press, Ann-Arbor.

Varughese, G., Ostrom, E., 2001. The contested role of hetero-geneity in collective action: some evidence from communityforestry in Nepal. World Dev. 29 (5), 747–765.