Professional Documents
Culture Documents
Collective Action
Collective Action
Laboratory Experiments
Author(s): David Goetze
Source: The Journal of Conflict Resolution, Vol. 38, No. 1 (Mar., 1994), pp. 56-86
Published by: Sage Publications, Inc.
Stable URL: http://www.jstor.org/stable/174401
Accessed: 28/10/2010 04:57
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=sage.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
Sage Publications, Inc. is collaborating with JSTOR to digitize, preserve and extend access to The Journal of
Conflict Resolution.
http://www.jstor.org
Comparing Prisoner's Dilemma,
Commons Dilemma, and Public
Goods Provision Designs in
Laboratory Experiments
DAVID GOETZE
Utah State University
In the social sciences, the term collective action has acquired a meaning
distinct from its more traditional reference to a group of people engaging in
actions directed toward a common end. The contemporary social science
sense includes the notion that incentives confronting individuals in a group
may not be consistent with achievement of cooperative or collective ends
that would benefit everyone in the group. This understanding implies that
cooperative actions by individuals generate benefits for everyone in the
group, but the self-interested individual benefits even more from choosing
not to cooperate (Elster 1989, 25; Axelrod 1984, 8; Dawes 1980, 169). In
Rapoport1974;Rapoport,Guyer,andGordon1976;OrbellandWilson 1978;
Hardin 1982), commons dilemmas (Dawes, McTavish,and Shaklee 1977;
Hardinand Baden 1977; Edney 1980; Ostrom 1990; Ostrom,Walker,and
Gardner1992), andproblemsin the provisionof public goods (Bohm 1972;
Kamien, Schwartz, and Roberts 1973; Snidal 1979; Smith 1979, 1980;
Marwell and Ames 1979, 1980, 1981; Isaac andWalker1988).
Unfortunately,we lack a systematic method for clarifying how these
representationsdifferfromand are similarto one anotherin ways thatmight
permitus to (1) dismiss those differencesas unlikelyto affect behavioror (2)
control for those differences so as to begin to cumulate findings across
representations.2Some apparentdifferencesmay involve nothingmore than
superficialvariationin the presentationof game incentives, or theirdesigns
may comprise qualitativelydifferent incentive characteristicsthat render
comparisons and cumulationdifficult, if not impossible. This article dis-
tinguishesvariationsin presentationandinvestigateswhetherthe threegame
types shareincentive commonalitiesthatwould permitmeaningfulcompar-
ison andcumulation.If experimentscanbe understoodas beingarrayedalong
common incentiveparameters,then a means shouldexist for controllingfor
incentives-the most importantfeaturesof any experimentaldesign.
The articleconcludesthatcommonincentiveparametersdo exist among
prisoner'sdilemma(PD), commonsdilemma(CD), and public goods provi-
sion (PG) designs, and thatthese featuresare readilyidentifiablefor games
thathave linearreturnsto cost contributionsand "summation"technologies
for aggregatingcontributionsinto supply levels.3 In contrast,nonlinearity
and technologies other than summationcan sometimes generate a mix of
qualitativelydifferentincentivefeaturesthatmake cumulationdifficultwith
available methods. For those designs that are amenableto cumulation,the
article formulates methods for renderingfindings comparable.First, the
articleprovidesa more precise definitionof collective action dilemmas.
which two or more actors are able to make choices between two or more
actionsandthe expectedoutcomeis Paretoinferiorto some otherobtainable
outcome (Taylor 1987, 19; Elster 1985, 1989, 25; Sandler 1992, 23). An
expected outcome is the result obtained when actors make rational,
uncoordinated choices that maximize self-gain. Expected outcomes for
nonrepeatedCADs, as defined here, are all Nash equilibria in which each
actor chooses his or her best strategy, given that other actors have done
likewise (Kreps 1990, 402-10). Having arrivedat an equilibrium,no actor
has an incentive to changestrategygiven the choices of otheractors(Sandler
1992, 15-17).
Pareto inferior outcomes are simply those for which anotheroutcome
exists thatmakesat least one personbetteroff withoutmakinganyoneworse
off. The latter outcome is Pareto superiorto the former.The discrepancy
between featuresof expected outcomes and featuresof outcomes that are
Paretosuperiorto themis whatmakesCADs interestingfor social scientists.
It defines a set of crucialsocial situationswhere incentiveslead individuals
away from cooperationthatwould have createdvalue for everyone (Dawes
1980, 169; Sandler1992, 45).
TABLE 1
Illustrative Prisoner's Dilemma Game
Player2's choices
Player l's choices
Cooperate Defect
Cooperate a b
7.50,7.50 0, 11.25
Defect c d
11.25, 0 3.75, 3.75
NONEXCLUDABLE GOODS
TABLE2
ChickenandAssuranceGames
withoutproblems,butprospectsarenotablybrighterthanfor strictcollective
action dilemmas. Our attention focuses on CADs because they involve
relatively pure forms of problematiccooperationand serve as good, unen-
cumberedexperimentalmodels for testingtheoriesaboutbehavior.
b = r(c)
where
b = benefitfromcontribution of a costunit,
c = sizeof theunitin whichcostcontributions canbe made,
r = a returnsfunctionrelatingbenefitsto costs.
For the examplein Table3, when c is equalto $1.00, as in the second cell
from the left in the top row, the benefit createdis $.75, so b/c and,hence r =
.75. In the last cell in the top row, four $1.00 cost units are contributedand
the benefitcreatedpercost unitis $3.00/4, so againb/c = .75. In this example,
the ratioof benefits to costs is always .75 no matterhow many cost units are
or have been contributed.1 Hence benefits are a linearfunctionof costs.
10. The ratio remainsthe same but may be less apparentin othercells where contributions
must be summedfromthe representativeindividualand others.In the second cell fromthe left
in the second row, $2.00 are contributed-$1.00 from the representativeindividualand $1.00
from others.The benefit createdis $1.50, as expected with an r value of .75. The entry in the
cell is only $.50, however, because the $1.00 cost contributionmade by the representative
individualhad to be subtractedfrom the total to yield the individual'snet payoff.
Goetze/ COMPARINGDESIGNS 67
TABLE3
Payoffs to a RepresentativeIndividualin a
Three-PersonPublic Goods Game (returnto contribution= .75)
Representative
individual Otherscontribute
contributes
$0 $1.00 $2.00 $3.00 $4.00
$20 -0.5: :0 .
0750 1.50 2.25 3.00
$1.00 05 05 1.25 2.00 2.75
$2.00 -0.50 0.25 1.00 1.75 2.50
may vary accordingto who is doing the contributingas when the amountof
defense good producedis mostly dependenton the effortof soldiersdefend-
ing a critical portion of a defense line (a "weakest-link"technology in
Sandler's [1992] terminology).13
bi = E rij(ck)
i=lj= l,k=
where
bi = the benefit to the ith individual,
ri = a returnsfunctionfor each beneficiarythatrelatesbenefitsto contributionof a cost unit
for each of thej contributors,
Ck = a cost unit for which k cost unitshave alreadybeen contributedby individualj.
Goetze/ COMPARINGDESIGNS 69
TABLE4
GeneralizedPublic Goods Game with
SummationTechnologyandLinearReturnsFunction
Individual Otherscontribute
contributes
0 c (n -j)c nc
0 0 b(n -j) bn
c b(n -j) + b - c bn + b - c
(m - i)c (m - i) (b - c) (m - i)(b - c) + b -
(m i) (b - c) + b(n -) m(b- c) + b(n -j)
mc m(b- c) m(b - c) + b b(n -j) + m(b - c) bn + m(b - c)
designs involving actorsthat have equal incomes (the most common treat-
ment of income in laboratoryexperiments)and hence abilitiesto contribute
equal numberof cost units to provisionof the public good.
As noted, the formatof PG matriceslists payoffs for a single individual
whose payoff possibilities, at least for this design type, are identical to all
otherinvolved actors.Therows distinguishthe contributionchoices (number
of cost units that can be contributedto provision of a public good) for this
representativeindividual.The columnsdistinguishthe cumulativecontribu-
tion choices that can be made by all other actors. The cells of the matrix
indicate the payoff consequences for the representativeindividual of any
combinationof choices made by the individualand otheractors.
The upper left-handcell is the amountreceived by the individualwhen
the individualmakes zero contributionto the publicgood andwhen all other
actors contributezero as well. Nothing is venturedby any of the actors, so
nothingis gained.In thecell immediatelybelow is thepayoffto the individual
when the individualcontributesone cost unit to provisionof the public good
while all otheractorscontinueto contributenothing.The individualreceives
the benefit (b) from the public good generatedby her or his contributionof
one cost unit and loses the amountof the contribution(c). Proceedingdown
this column, the individual'spayoff is simply multipliedby the numberof
cost units she or he contributesup to the limits of her or his income (m cost
units).
In the cell immediatelyto the rightof the upperleft-handcell is the payoff
received by the individualwhen the individualcontributeszero cost units to
provisionof a public good andthe cumulativecontributionof all otheractors
is one cost unit. The individual receives the benefit (b) generatedby the
contributionof thatcost unit but does not sufferthe loss of thatunit (c). The
cost of that unit is borne by some other actor whose payoffs are not listed.
Proceedingacrossfrom the upperleft-handcell, the individual'spayoffs are
multipliedby the numberof cost units all other actors contributeup to the
cumulative limits of their incomes (n cost units). The individual would
receive the benefits of those contributionswhile sufferingnone of the costs.
The entries in all other cells of the matrixare determinedby summingthe
payoffs thatwould be generatedby the representativeindividual'schoice of
cost units to contributeand the payoffs that would be generated by the
cumulativecontributingchoices of all otheractors.
The upperleft-handcell containstheNash equilibrium.Moving down any
of the columns shows that an individual'schange in value or payoff for an
additionalunitcontributionis alwaysb - c. The definitionof a CAD specifies
thatthe cost mustbe greaterthanthe benefit (so as to ensurethatcooperating
or contributingis problematicfrom an incentiveperspective),so this termis
72 JOURNALOF CONFLICTRESOLUTION
always negative (at least for PG games thatalso happento be CADs). Hence
any individualwould always be worse off making a unilateralmove away
from the zero contributionchoice (the upperleft-handcell).
The positive incentive to forego contributingan additionalcost unit to
provision of the nonexcludablegood (defect incentive) is the lost value
associatedwith an additionalcontributionof a cost unit-the absolutevalue
of b - c. The individual'stotal change in value for contributingall the units
she or he is capable of contributingis m(b - c), the lower left-hand entry
correspondingto the payoff she or he receives if no other actors contribute
to the public good. The change in value thatthe individualexperiences for
contributingall possible units is the same, no matterhow many units other
actors contribute,as confirmedby subtractingentries in the top row from
those in the bottomrow. The resultis always m(b - c).
More useful still as an indicatorof the incentiveto defect is the inverseof
the value of the returnsfunction, that is, 1 - b/c. The value of the returns
function(b/c) specifies what portionof the cost contributionis retainedas a
benefit on actual contributionto provision of the nonexcludable good,
whereas the inverse specifies what portion is lost and thus indicates the
relative degree to which contributionsare discouraged.For the example in
Table3, the value of the returnsfunctionwas $.75/$1.00 = .75 so the inverse
is 1.00 - .75 = .25. An individualloses one-fourthof every cost unit she or
he contributes.The inverse of the value of the returnsfunction is constant
irrespectiveof the numberof cost units contributed.It also is standardized
for the size of the cost unit andconsequentlyhas much appealas a summary
measurethat can be employed to comparedefect incentives across games
with differentstructures(althougheasy comparisonsstill requirethat all of
the games have summationtechnologies and linearreturns).
The lower right-handcell containsthe payoff associatedwith the Pareto
best outcome. It is the only outcome (the set of payoffs for all players) that
necessarilycreatesmore value thanthe Nash equilibriumfor all playersand
more value than any otheroutcome thatis superiorto the Nash equilibrium.
The upperright-handcell yields a higher payoff (bn) than the lower right-
handcell for the individualrepresentedin thematrix,butthe payoffsfor other
players would be worse because they would fail to benefit from the repre-
sented individual'slevel of cost contributions(zero). Someone will always
be made worse off wheneverthere is a decline in the numberof cost units
contributed.At the same time, a definitionalpresumptionof a CAD is that
some mannerof contributions(cooperation)by individualswill make every-
one betteroff thanin the absenceof contributions.As alreadynoted,to ensure
thateveryoneis betteroff fromadditionalcontributions,those additionsmust
be symmetricacross contributors(or someone might forfeit a cost contribu-
Goetze/ COMPARINGDESIGNS 73
PRISONER'S DILEMMAS
and specify "payoff wealth" ratherthan just payoffs resulting from cost
contributions.Hardin(1982) andTaylor(1987, 139) note thatPD games are
usually representedin a 2 x 2 matrixthat shows outcomes from two inter-
dependentchoices of two players,althoughN-persongames may be accom-
modated in the same matrix structureusually by designating one of the
playersas "allotheractors"who arepresumedto makeuniformchoices. Four
differentoutcomes are specified for an individualactor, which result from
the permutationsfor the single cooperateand not cooperatechoices of the
two players.If one understandscooperateto be synonymouswith contribut-
ing to a nonexcludablegood, thenthe PD game can be seen as allowing only
a single level of contributionto a nonexcludablegood in contrastto the
multiplelevels of contributionsthattypicallyconfrontplayersin PG games.
The preferencefor presentingPD games with only a single choice of
contributionprobablystems fromthe way in whichPD games aredefined-
by the nature of the inequalities among the four outcomes available to a
player.The neatnessof the inequalitieswoulddiminishwith additionallevels
of choice. In particular,PD games are typically defined as a matrix of
outcomesfor whichT > R > P > S whereTrefersto the (temptation)outcome
for individuali when i defects andthe otherplayercooperates;R refersto i's
(reward) outcome when both players cooperate; P refers to i's (punish-
ment) outcome when both players defect; and S refers to i's (sucker)
outcome when i cooperatesandthe otherplayerdefects (Axelrod 1984, 206;
Taylor 1987, 14).
MatrixA in Table5 depictsthese generaloutcomesandinequalities-the
arrowsindicate the directionof the relevantinequalities.The first entry in
each cell is the outcome for player 1; the second entryfor player 2. Matrix
B (drawnfrom Table 1) containsspecific outcomesthat are configuredas a
prisoner'sdilemma.The configurationof relationsin both matricesensures
thatdefect/defectis the Nash equilibriumandthatthe defect/defectoutcome
is Pareto inferior to cooperate/cooperate.Moreover, the set of relations
defining a PD can be interpretedas evident because the matrices involve
choices to contributeor not contributecost unitsto provisionof a nonexclud-
able good. Every contributionis presumed to involve a net loss for the
contributingindividual.Both T andP can be understoodas not contributing
choices, so net losses for contributingmeans that T must be greaterthan R
andP mustbe greaterthanS. Everycontributionalso resultsin a nonexclud-
able augmentationof value for the otherplayer(s)and as long as the returns
value is sufficiently high (see footnote 18), cooperate/cooperateis greater
thandefect/defectandR must thereforebe greaterthanP.
The outcomes specified in a PD game sometimes include a component
not presentin PG games. PG games usuallypresentto a playerthe payoffs,
76 JOURNALOF CONFLICTRESOLUTION
TABLE5
Prisoner's Dilemma Game
MatrixA MatrixB
Cooperate Defect Cooperate Defect
that is, net changes in her or his wealth resulting from the different levels of
contributions that she or he and other players in the game can make. In
contrast, PD games typically specify possible payoff wealth, determined by
adding the initial income provided the actors to the payoffs created by
contributing costs to provision of the good. This composition of entries in a
PD matrix makes sense when reflecting on the similarities between defect
actions and not contributing actions. The term "defect" suggests a failure to
take some positive action in a PD game just as "not contributing" denotes
inaction in a PG game. Defect and not contributing are functionally equiva-
lent. Moreover, alternative actions in these games (cooperate and contribute)
have similar consequences for the players-they generate a return to the
acting individual and an external benefit to others.
The defect/defect cell of a PD game and the cell of a PG game in which
no one contributes to a nonexcludable good are not always recognized as
conceptually equivalent, in part because the latter registers zero value for the
payoff, whereas in PD games, defect/defect entries other than zero are quite
common. If defect/defect is truly equivalent to the absence of contributions
to a nonexcludable good, why should any value, positive or negative, be
registered in that cell? A compelling explanation of the difference is that the
cells of a PD matrix aggregate with payoffs the initial income of the players.
One can interpret the entries in the defect/defect cell of PD game not as
payoffs from provision of a good because no contributions to a good are
imagined in this cell, but solely as the amount of income with which the
players start the game.17 Other cells of a PD matrix can be understood as
17. Initialincome refersto income availableto an individualfor contributingin a particular
collective action situation.
Goetze/ COMPARINGDESIGNS 77
TABLE6
GeneralizedPrisoner'sDilemma Game with
SummationTechnologyandLinearReturnsFunction
Player2
Player 1
Cooperate Defect
(contribute) (not contribute)
Cooperate
(contribute) 2b- c + e, 2b- c + e b - c + e, b+ e
Defect
(not contribute) b + e, b- c + e e, e
COMMONS DILEMMAS
SUBTRACTABILITY
19. For a discussionof factorsthat affect the feasibility of dividinga commons, see Goetze
(1987).
20. Olson (1965, chaps. 1 and 2) typically assumedsubtractabilityin his analysis of size
effects, as did Hardin(1982) in his sequel to Olson's seminal study.
80 JOURNALOF CONFLICTRESOLUTION
Comparingresultsfromgames employingothertechnologiesandnonlin-
ear returnsis, however,very problematic.This circumstanceis troublesome
because most real-worldcollective action dilemmas involve asymmetries
acrossactorsandnonlinearreturns.Importantempiricalfield studies26 of many
of thesesituationshavealreadybeenundertaken, butbecausetheymust accept
existing asymmetriesand nonlinearities,the resultsare not easily compared
with each other, much less with the results of laboratoryexperiments.
A large numberof laboratoryexperimentshave been conducted using
summationtechnologies and linear returnsso that meaningfulcomparison
and accumulationcan be accomplished with these studies alone. Their
symmetryandlinearityraisequestions,however,abouthow adequatelythey
representreal-worldsituationsand whetherresultsfrom these experiments
informus aboutreal-worldbehavior.27 My own surmiseis thatexperiments
with summationtechnologies and linearreturnsconfrontpeople with fairly
simple situationsthat are relatively unfetteredby strategiccomplications.
Simplicity helps people understandthe collective action characterof the
game and enables researchersto explore how people behave when they are
awarethatthey confronta collective action situation.
Moreover,use of these simple games is analogousto the physicist's use
of a vacuum for studyingthe laws of motion. In the real world (at least on
the earth'ssurface),vacuumsarenot naturallyencountered.Particlesbehave
in an environmentcomplicated by the effects of other particles and by
randomlyintrudingforces that are incidentalto fundamentalprocesses that
thephysicistwantsto identify.Thephysicistcannotisolatethese fundamental
processes in the real world, but she or he still knows that they are operating
and are fundamentalto understandingbehaviorsthat are observablein the
real world.The inducedvacuum,althoughcontrived,allows these processes
to be isolated and studied.Similarly,a game with a summationtechnology
and linear returnsis not naturallyencounteredin the real world, but the
contrived game allows us to isolate many of the fundamentalfeatures of
collective action dilemmasand to observe how people behave in response.
These recordedbehaviorsshouldgive us importantclues aboutwhy people
behave the way they do in more complicatedand sometimesindecipherable
real-worldenvironments.
REFERENCES