The effect of out-group competition on individual behavior and out-group perception in the Intergroup Prisoner’s Dilemma (IPD) game

The effect of out-group competition on individual behavior
and out-group perception in the
Intergroup Prisoner’s Dilemma (IPD) game1

Harel Goren

The Hebrew University of Jerusalem
Center for Rationality and Interactive Decision Theory
and the Department of Psychology

Biographical note: The author received his Ph.D. in social psychology from The
Hebrew University of Jerusalem and has recently finished a one year post-doctoral
visit to The Department of Management and Policy at The University of Arizona.


Abstract
Hebrew University of Jerusalem students participated in two experiments of
repeated play of the Intergroup Prisoners’ Dilemma (IPD) game, which involves
conflict of interests between two groups and, simultaneously, within each group. The
experiments manipulated the level of competition exhibited by the out-group
members (i.e., their level of contribution to their group’s effort in the conflict).
Consistent with the hypothesis that participants use strategies of reciprocal
cooperation between groups, higher levels of out-group competition caused
participants to increase their contribution and lower levels caused them to decrease it.
In addition, participants had accurate recall of the contribution levels of out-group
members, and they attributed motivations to out-group members in a manner that
reflected their level of contribution. The nature of reciprocation with the out-group is
discussed in light of both behavioral and cognitive data.

Key words: Intergroup conflict, Team games, Prisoner’s dilemma, reciprocal
strategies, Intergroup perception.

2


Introduction
What motives govern individual behavior in intergroup conflicts? The answer
to this question depends to a large extent on how the conflict is conceptualized. Social
scientists have often modeled intergroup conflict as a two-person game (Allison,
1971; Axelrod, 1984; Brams, 1975; Snidal, 1986), necessarily assuming that the
interest of the individual is identical to that of his group. Thus, if it is rational for the
group to compete it must also be rational for the individual group member to do so.
Other researchers recognized that what is best for the group is not necessarily best for
the individual group member. Most notably, Campbell (1972) observed that
contribution to the collective group effort is not rational from the perspective of the
individual since “Group-level territoriality has always required that the soldier
abandon for extensive periods of time the protecting of his own wife, children and
home” (p. 24).
The conflict between individual interest and group interest referred to by
Campbell (1972) is a problem of public goods provision (Rapoport and Bornstein,
1987; Bornstein, 1992). It stems from two facts. First, the payoffs associated with the
outcomes of inter-group conflicts (e.g., territory, political influence, higher wages) are
equally available to all the members of a group, regardless of their contribution to the
group’s effort. Second, although the size of these public goods increase the more
group members contribute, the individual’s contribution to the group’s effort is
typically too costly (in terms of money, time, effort or risk taking) to be justified on a
rational basis. Therefore, self-interested rational group members are expected to free
ride on the contribution of others. Of course, if everyone else free ride as well, the
group would lose the competition and the public goods.

3

To capture the intra-group and inter-group levels of conflict, Bornstein (1992)
devised the Intergroup Prisoner’s Dilemma (IPD) Game. The game as operationalized
in the present study involved a competition between two teams with three players in
each team. Each player received an endowment of 2 points and had to decide whether
or not to contribute his endowment towards the group’s effort. After decisions were
made, a bonus was paid to each player according to following scheme: if all players in
Team A contributed, while no players in Team B contributed, each player in A
received a bonus of 6 points and each players in B received 0 points. If there were 2
more contributors in Team A than in Team B, each player in A received 5 points and
each player in B received one point. If there was one more contributor in Team A than
in Team B, each player in A received 4 points, whereas each player in B received 2.
Finally, in case of an equal number of contributors in both teams, each player in both
Teams received a 3-point bonus. In addition to the bonus, a player who decided not to
contribute kept the 2-point endowment. The payoffs to a member of Team A as a
function of his decision to contribute (C) or not to contribute (NC), the number of ingroup contributors (mA) and the number of out-group contributors (mB), appears in
Figure 1.
The payoff parameters of the IPD game were such that: First, withholding
contribution was the dominant individual strategy; that is, regardless of what the ingroup and out-group members did, the individual earned an extra point by not
contributing. Second, the dominant strategy for each team was to have all of its
members contribute, regardless of what the out-group did. In the present experiment,
each team player earned 1 more point if all group members (including him)
contributed than if they all did not. Third, all members of both teams were better off if
they all withheld contribution than if they all contributed. When no one contributed (a

4


0:0 tie) each player earned 5 points whereas if all contributed (a 3:3 tie), each player

earned only 3 points. No-contribution was, in fact, the collectively (i.e., Pareto)
efficient outcome of the game, the one which maximize the earnings of all six
participants.

Figure 1. Payoff to a member in team A as a function of the decision to contribute (C)
or not to contribute (NC), the number of in-group contributors (mA) and the
number of out-group contributors (mB).

7
6

Payoff

5
4

C

3

NC

2
1
0
0
0

1
0

2
0

3
0

0
1

1
1

2
1

3
1

0
2

1
2

2
2

3
2

0
3

1
3

2
3

3
3

mA , mB

The first and second properties of the IPD game define the intra-group payoff
structure as a three-person PD game or a social dilemma (Dawes, 1980). Although the
in-group’s payoffs decrease the more out-group players contribute, the structure of the
intra-group dilemma remains constant regardless of the number of out-group
contributors.2 As can be seen in Figure 1, in all four intra-group PD games
(corresponding to 0, 1, 2 and 3 out-group contributors in the IPD game) the cost of
contribution for the individual and the benefit (i.e., externality) it produces for the
team are the same.

5


Therefore, if one assumes that individual behavior is motivated solely by self-interest
-- the assumption of narrow rationality -- one should expect no contribution in the
one-shot IPD game, irrespective of the out-group’s behavior. Similarly, if one assumes

that individuals are motivated only by a concern for the collective in-group interest,
one should expect full contribution, regardless of what the out-group does. Of course,
in reality, participants are likely to be concerned with both self-interest and common
group interest to various degrees. Nonetheless, any fixed combination of self-interest
and group interest should lead to a constant contribution rate, irrespective of the
number of out-group contributors.
What if participants are predisposed to maximize the relative difference in
payoffs between the in-group and the out-group? The assumption that people are
motivated to achieve positive self esteem by making the in-group positively distinct
from the out-group, is central to social identity theory (Hogg & Abrams, 1988; Tajfel
and Turner, 1986). This competitive inter-group motivation was demonstrated in
numerous laboratory experiments using the minimal group paradigm (for reviews see
Brewer, 1979; Diehl, 1990; Messick & Mackie, 1989; Tajfel, 1982). However, in the
IPD game individual behavior, even if governed by a motivation to maximize the
groups’ payoff difference, should not be affected by the behavior of out-group
members. This is so, because in the IPD, individual contribution increases the ingroup’s payoff by 3 points and reduces the out-group’s payoffs by 3 points, regardless
of the out-group behavior. Therefore, no matter what the out-group does, individuals
who wish to maximize the payoff difference between their team and the other team
should always contribute.
Participants’ decisions are expected to be affected by out-group behavior in the
one-shot IPD only to extent that they are motivated to “win” or at least not “lose” the

6


game. That is, if it is important to them that their group earns more than the out-group
(by any margin) or that it earns not less than the out-group. The notion that
participants are motivated to get at least as much as out-group members do (i.e. –
motivated not to “lose” the game) is supported by studies done in the minimal group
paradigm, which showed that equity considerations affect participants’ decisions

(Diehl, 1989; Ng, 1981). These studies suggest that part of the reason for biased
allocations in the minimal group paradigm is that participants expect biased
allocations by out-group members. In trying to achieve an equitable allocation
participants discriminate against out-group members themselves. In the IPD game,
which has a symmetric payoff structure, equity considerations dictate that participants
would contribute at the level they expect out-group players to contribute.3
Repeated interaction in the IPD game: The repeated IPD game is different
from the one-shot game in two important ways: First, in an on-going interaction
behavior can be dependent on the earlier choices of other players, whereas in a oneshot game this is not possible. This opens the possibility for using strategies of
reciprocal cooperation in light of which contribution is seen as a rational strategic
move. Second, an iterated environment provides players with an opportunity to learn
the structure of the strategic situation and adapt their behavior accordingly -- an
opportunity that they do not have in a one-shot game.
Several experiments that examined the dynamics of contribution in the iterated
IPD game (Bornstein, Erev & Goren, 1994; Bornstein, Winter & Goren, 1996; Goren
& Bornstein, 1999) show that, when communication among players is prohibited
(both within and between teams), contribution decreases steadily as the game
progresses. These works maintain that the gradual decrease in contribution levels is
most readily accounted for by individual rationality (i.e., selfishness). To explain this

7


finding all one needs to assume is that players adapt their choice behavior as they
become more experienced, so that choices that have led to good outcomes in the past
are more likely to be repeated in the future (Harley, 1981; Maynard-Smith, 1984;
Selten, 1991). Since withholding contribution is the unconditionally best individual
strategy in the IPD game, this simple principle of reinforcement learning, known as
the law of effect (Thorndike, 1898), would inevitably move players in the direction of
no contribution. This interpretation receives substantial support from computer

simulations which, using Roth and Erev’s (1995) quantification of the law of effect,
closely reproduce the experimental results.
However, in addition to being the selfish (narrowly rational) individual
strategy, withholding contribution is also the cooperative strategy vis-a-vis the other
group, since it always increases the total out-group payoff by 3 points (1 point for
each individual out-group player). Therefore, low contribution levels could, in
principle, reflect an evolution of cooperation between the two teams.
Research on the two-person prisoner’s dilemma game has shown that mutual
cooperation evolves over time (Radlow, 1965; Rapoport, Am. & Mowshowitz, 1966;
Rapoport, An. & Cammah, 1965). Further studies have shown that strategies of
reciprocal cooperation, like TIT-FOR-TAT (Axelrod, 1984), are influential in bringing
mutual cooperation (Oskamp, 1971; Komorita, Hilty, & Parks, 1991; Wilson, 1971).
Reciprocity is defined as a norm that “prescribes that we should help those who have
help us in the past and retaliate against those who have injured us” (Komorita, Parks
& Hullbert, 1992, p. 608). Gouldner (1960) viewed the reciprocity norm as universal
and contributing to the stability of social structure.
It is possible that such tendencies for reciprocation are generalized to
intergroup contexts (Patchen, 1987). Rabbie (1998), for instance, explains the typical

8


finding in the Group-Individual Discontinuity paradigm (see review in Insko &
Schopler, 1998) as a reciprocity effect. According to Rabbie, it is possible that the
increase in competitive choices of groups (in comparison to individuals) results from
the fact that groups tend to reciprocate exploitative choices more than individuals.
Using between-team reciprocation in the IPD can help bring about the Pareto
efficient outcome of no contribution. By reacting to out-group contribution in kind
players can deter such behavior of out-group members. This is so, because their
contribution would reduce the out-group earnings in the round that follows. Reacting

to out-group contribution in kind is also consistent with the notion that people employ
equity considerations in inter-group interactions. Even though equity and reciprocity
are not identical, the use of reciprocal strategies can reduce the difference between the
final outcomes of the two teams.
In an attempt to differentiate between the process of learning and that of
between-team reciprocation in intergroup interaction, Goren and Bornstein (1999)
conducted an experiment in which the IPD game was played repeatedly under
different matching protocols. Their ‘Matching protocol’ manipulations related to
changes in the composition and matching of groups from one round to the other
during the repeated IPD experiment. In the fixed-matching condition, team
composition and matching between teams was constant throughout the entire game.
This condition mimics a naturally occurring intergroup interaction, in which people
belong to the same distinct group and two groups repeatedly interact with each other.
In the mixed-matching condition, the participants were randomly assigned to teams,
which were randomly paired for an IPD game at the beginning of each round of the
experiment. Both matching protocols provided participants with the opportunity to
learn the structure of the one-shot IPD game. But, whereas the fixed protocol enables

9


them to use between-team reciprocal strategies, the mixed protocol hinders any form
of effective reciprocation. The results showed no effect of matching condition on
contribution, nor an interaction of condition with time, thus failing to support
between-group reciprocation.
The experiment by Goren and Bornstein (1999) constituted a weak test for the
between-group-reciprocation hypothesis. In their “natural” experimental setting it is
quite difficult to disentangle the effect of between-team reciprocation from that of
learning. If all participants are learning the same thing, namely the dominant strategy
of withholding contribution, it is difficult to detect those players who (potentially)

make their decision contingent on the behavior of out-group players. Reciprocation in
this case (to the extent that it occurred in the fixed-matching condition) can possibly
alter the speed of the dynamic process but not reverse its direction.
Given the problem in interpreting the results of Goren and Bornstein (1999),
the present experiment used a more direct approach to examine between-group
reciprocation. Instead of using “real” out-group players, this study used simulated
ones. Using virtual players, whose behavior is predetermined, enables assessment of
the extent to which individuals react to the out-group. The virtual out-group players in
this study displayed different levels of contribution at different periods of the game.
Based on the principal of between-group reciprocation it was hypothesized that the
contribution levels of the participants’ would be affected by those of the virtual outgroup players. Specifically, participants were expected to contribute more following
high levels of out-group contribution, and less following low levels of out-group
contribution.
It is important to note that, although learning and reciprocation lead to the
same outcome, the two processes assume fundamentally different strategic aims. The

10


learning explanation simply maintains that, since it is the individual’s short-term
interest not to contribute, people will eventually learn to free ride. The fact that this
results in an outcome that is best for all players of both groups, is just a by-product of
this simple process of individual adaptation. Hypothesizing that players reciprocate
with out-group members, on the other hand, presupposes that people consciously
condition their behavior on that of the out-group, in a calculated attempt to bring
about a more beneficial outcome.
The hypothesis that participants employ between-team reciprocal strategies
requires that they would be attentive to the behavior of the out-group. This conjecture
was tested by examining how well the participants remember the contribution
behavior of out-group members during the game.

The motivations attributed to out-group members were also assessed in an
attempt to find out whether these motivational attributions reflect the pattern of outgroup behavior. Recent research indicates that people tend to reciprocate the
motivation they perceive others to have. Van Dijk and Wilke (1999) found that when
participants attributed the behavior of another player to self interest they contributed
more when their own contribution was necessary for the attainment of a public good
and less when it was not necessary for public good attainment. This is consistent with
the participants’ own self-interest since the public good was worth more then the
participants’ endowments (i.e., participants reciprocated selfishness by acting selfishly
themselves). The same effect was not obtained when participants interpreted the other
player’s actions as resulting from a motivation for fairness.
If the perception of motivations is involved in between-group reciprocation
one should expect participants of the current study to attribute different motivations to
out-group players who show different levels of contribution. Specifically, participants

11


should attribute competitive motivations to out-group members who contribute a lot,
and benign motivations to out-group members who refrain from contribution.
Furthermore, studying the motivations that participants attribute to out-group
members can help clarify the reasons that participants have for reciprocation with it.
(This point will be further developed in experiment 2.)
Below I report two experiments that tested these hypotheses. The first
experiment was designed to look mainly at the behavioral aspects of between-team
reciprocation, while the second focused more on the perception of out-group
motivations in the repeated IPD game.
Experiment 1
This experiment contrasted two conditions, involving different contribution
patterns by the out-group. In the High-to-Low (HTL) treatment, the virtual out-group
players started with a medium level of contribution, then contribution increased to a

very high steady level, decreased to a steady low level, and finally ended with a
medium level again. In the Low-to-High (LTH) treatment the pattern of contribution
by the virtual out-group members was the exact mirror image of the pattern in the
HTL treatment. Out-group players in the LTH condition contributed at a medium
level, then decreased contribution to a low level, increased it to a high level and then
contributed at a medium level again toward the end. The proportions of contribution
by the virtual out-group players in conditions HTL and LTH appear in Figure 2
(computed over 12 blocks of 5 rounds each).
To have a sufficiently powerful manipulation, the levels of contribution by the
virtual out-group members (in the ‘high’ periods) were higher than those usually
found in the repeated IPD game. In addition, the reason for contrasting these two
conditions is that the usual finding in the repeated (fixed) IPD game is a gradual

12


decrease in contribution levels. This pattern loosely fits the HTL condition and
therefore it is important to confirm that this pattern can reverse itself in the LTH
condition, where the out-group members first display a low level of contribution and
only later increase it. If such a reverse pattern is found this will serve as a clear
confirmation for the use of between-team reciprocation. It should be noted that since
the overall levels of contribution in the two virtual out-groups were identical (50%
overall contribution by each out-group member in both conditions) there was no point
in assessing the attribution of different motivations to the two different virtual outgroups in this experiment.4
Procedure
The participants were 60 undergraduate students at the Hebrew University of
Jerusalem (30 in each condition), with no previous experience with the task.
Participants participated in experimental sessions of 12 people each. When they
arrived at the laboratory the participants were seated in separate cubicles facing a
personal computer, and were given verbal and written instructions concerning the

rules and payoffs of the game. The instructions were phrased in terms of the
individual's payoffs as a function of his or her own decision (to contribute or not) and
the decisions made by the other players. The payoffs were summarized in a table,
which was available to the participants throughout the experiment. Participants were
not instructed to maximize their earnings, and no reference to cooperation or defection
was made. Participants were given a quiz to test their understanding of the game, and
the instructions were repeated until the experimenter was convinced that all the
participants understood the payoff rules. Participants were also told that to ensure the
confidentiality of their decisions they would receive their payment in sealed envelopes

13


and leave the laboratory one at a time with no opportunity to meet the other
participants.
Participants played 60 rounds of the IPD game with the payoff parameters
described earlier. The 12 participants were divided into four three-person teams, and
were told that the same two teams would compete against each other throughout the
entire game. However, each team actually played against a virtual out-group under
one of the two the conditions described above (HTL or LTH).
Following each round, each participant received detailed feedback concerning
the decision made by every other in-group and out-group member in that round. In
addition, the participants received information about the total number of in-group
contributors, the total number of out-group contributors, their own earnings (in points)
in the last round, and their cumulative earnings.
The number of rounds to be played was not made known.5 Following the last
round, the points were added up by the computer and cashed in at the rate of IS 1 for 8
points (1 Israeli Shekel was equal to $0.29 at the time the experiment took place, and
the average participant earned IS 36.6, or about $10.5). Participants then filled out a
questionnaire that included questions about the level of contribution by out-group

members in the first and last 20 rounds. They were then debriefed on the rationale and
purpose of the study, and were paid and dismissed individually.
Results
Contribution rates: The 60 rounds were divided into 12 blocks of five rounds
each, and the mean proportion of contributions per block was calculated. These means
appear in Figure 2. The issue of the possible dependency among players of the same
team was addressed by averaging the contribution proportions of all three team

14


members and using this team mean as the unit for analysis. Thus a 2 (experimental
condition) by 12 (blocks) mixed factorial design was used in the analysis.
The results of the ANOVA show a significant interaction effect of condition by
block (F(11,198)=5.15, p<0.05). The pattern of this interaction show that the participants
generally followed the levels of contribution displayed by the different virtual outgroups. The participants tended to contribute more when the level of contribution by
out-group members was high, and less when it was low. The graphs of actual
contribution in the two conditions cross and alternate at about the same points in time
as those of the two virtual out-groups. Participants in the HTL condition contributed
more in the first part of the game than in the second part, while participants in the
LTH condition showed the exact opposite pattern.
The main effects for condition and block were non-significant (F(1,18)=0.21,
p=0.65; F(11,198)=0.46, p=0.93, respectively). Note that if individuals use between-team
reciprocal strategies, this is exactly the pattern of results one would expect. This is
because the total contribution rate by the virtual out-group players was identical in the
two conditions, and the contribution rate averaged across the two conditions is about
50% in each block..
Figure 2: Experiment I - Virtual out-group members’ behavior and observed
behavior (proportion of contribution) by block in the two experimental
conditions.

Prop. of Contribution

1
0.8

VIRT'-HTL

0.6

VIRT'-LTH
Res HTL

0.4

Res LTH
0.2
0
1

2

3

4

5

6

7

8

Block

15

9

10

11

12


Dependency of contribution on out-group behavior: If participants were
making their contribution decisions contingent on out-group behavior, then their
decisions should be correlated with the number of out-group contributors in previous
rounds. Table 1 shows the percent of contribution in round t (2 to 60) as a function of
the number of out-group contributors in round t-1 (1 to 59) in the two experimental
conditions. The frequencies shown in the table are computed over the decisions of all
participants in each condition. The table shows that in both conditions the proportion
of contribution increased as the number of out-group contributors in the previous
round increased from 0 to 3. In the HTL condition, contribution is about 29% when
the number of out-group contributors in the previous round is 0 or 1. Contribution
increases to a little over 40% when the number of out-group contributors is 2 or 3. In
the LTH condition, contribution is about 28% when the number of out-group
contributors in the previous round is 0 and it increases almost linearly up to 52%

when the number of out-group contributors in the previous round is 3.
Table 1: Experiment I - Percent of contribution (at round t) following 0, 1, 2 or 3
out-group contributors (at round t-1) in the two experimental conditions.*
# of out-group
contributors, t-1
condition
High to Low
Low to High

0

1

2

3

29.2
(17)
28.5
(16)

28.5
(11)
36.0
(15)

41.6
(15)
42.1

(11)

44.4
(16)
52.2
(17)

* Numbers in parentheses present the frequencies of each level of out-group
contribution in rounds 1 to 59 (t-1).

16


To have an adequate and unconfounded assessment of the correlation between
the individual’s contribution and the number of out-group contributors in the previous
round a separate logistic regression model was estimated for each participant. The
participant’s decisions were predicted by the number of out-group contributors and by
the number of other in-group contributors in the previous round. The regression
coefficient for the first predictor reflect changes in the individual’s contribution due to
out-group behavior, which can not be attributed to any concomitant tendency for
reciprocating in-group members’ behavior.
In the HTL condition the average logistic regression coefficient for the number
of out-group contributors in the previous round was 0.49 and in the LTH condition it
was 1.07. Both of these averages are significantly different from zero (t(29)=3.88,
p<0.001; t(28)=2.23, p<0.05 - respectively) providing yet another proof for the use of
between-team reciprocal strategies. (The positive logistic regression coefficients attest
to a positive correlation between the predicted event - contribution, and the predictor
variable - number of out-group contributors.)
The corresponding average logistic regression coefficients for the number of
in-group contributors in the previous round were smaller: 0.23 in the HTL condition

(t(29)=1.59, p=0.12) and 0.46 in the LTH condition (t(28)=2.29, p<0.05).
Memory of out-group contribution levels: Participants were asked to recall the
number of contributions made by each participant in the out-group in the first 20
rounds and last 20 rounds of the game. The average level of contribution recalled
(across the three out-group members) was computed and analyzed in a 2 (condition)
by 2 (first 20/last 20 rounds) mixed design ANOVA. Figure 3 shows the averages of
recalled contribution levels of the out-group in the first and last 20 rounds of the game
in the two conditions. (The average numbers of contributions actually made by virtual

17


out-group members in the first and last 20 rounds were 15 and 5 (respectively) in the
HTL condition, and 5 and 15 (respectively) in the LTH condition.)

Figure 3: Experiment I - Average recalled number of out-group contributions in the
first and last 20 rounds of the game in the two experimental conditions.

16

Recalled Contribution

14
12
10
8
6
4
2
0

HTL - Beg

LTH - Beg

HTL - End

LTH - End

The analysis shows that, as expected, participants were quite attentive to the level of
contribution by the virtual out-group players. This fact is manifested in a significant
interaction effect between experimental condition (HTL or LTH) and the game period
(first or last 20 rounds) that participants were asked to recall (F(1,58)=52.7; p<0.01).
The main effects were non-significant. Since there was no difference between the two
experimental conditions with regard to the overall level of contribution by the outgroup, there was no reason to expect main effects of condition or of game period.
Nevertheless, as shown in figure 3 participants clearly over-estimated the outgroup’s contribution when it was at a low level, and under-estimated the out-group’s
contribution, when it was at a high level. This “regression to the mean” is statistically
significant. When out-group players contributed at a high level (15 contributions per

18


player, on average, in 20 rounds), participants recalled a significantly lower level of
contribution. This was true both for the HTL condition were participants recalled on
average 12.72 in the first 20 rounds (t(29)=4.37, p<0.01), and the LTH condition where
they recalled 12.64 contributions in the last 20 rounds (t(29)=3.66, p<0.01). Similarly,
when out-group contribution was low (5 contributions per player, on average, in 20
rounds), participants remembered it as significantly higher. (M=7.8, t(29)=3.75, p<0.01,
in the HTL condition, and M=8.89, t(29)=6.31, p<0.01 in the LTH condition).
Discussion
The results of experiment 1 clearly demonstrate that individual behavior in the

repeated IPD game is affected by the behavior of out-group members. Generally
speaking, participants increased their own contribution when the competition by the
out-group was intense and decreased it when the out-group competition declined. As a
result, there was a significant correlation between participants’ decision to contribute
and the number of out-group contributors in the previous round. Participants also had
a fairly good recollection of the dynamics of out-group behavior. These results are
consistent with the hypothesis that individuals are using strategies of between-team
reciprocation in a continuous interaction with the same out-group.
The current results are inconsistent with explaining the gradual decline in
contribution in previous IPD experiments as resulting only from individual learning.
Especially, individual learning models would not predict an increase in contribution in
later rounds of the game as was demonstrated in the LTH condition.
As an illustration, a simulation was run in which the simulated players used
Roth and Erev’s (1995) learning algorithm and were confronted by each of the two
virtual out-group contribution patterns. Fifty “groups” of simulated players were run
in each of the two conditions. The results appear in figure 4. As can be easily seen in

19


the figure, there was no interaction between time and the two virtual out-groups
contribution patterns (F(11,1078)=1.03, p=0.42, using the same ANOVA as the one used
on the actual participants’ data). The simulation results show only a decline in
contribution over time without any responsiveness to the virtual out-groups’ behavior
(F(11,1078)=8.60, p=0.0001, for the block effect). This pattern was contradicted by the
results of the actual participants in the experiment who responded to the level of outgroup contribution.
Figure 4: Experiment I - Simulated data of “players” using Roth and Erev’s (1995)
learning model and confronted by the two virtual out-groups.

Prop. of Contribution

1
0.8
0.6

HTL-SIM
LTH-SIM

0.4
0.2
0
1

2

3

4

5

6

7

8

9

10

11

12

Block

It is important to note that the contribution level of the (actual) participants
was on average considerably lower than that of the (virtual) out-group members. This
lower contribution level clearly rules out the possibility that participants use strict
between-team reciprocal strategies that respond to every contribution by the out-group
with a contribution of their own.
The most plausible explanation for the above results is that some mixture of
between-team reciprocation and individual learning was taking place. In other words,

20


while participants changed their contribution behavior in reaction to that of the outgroup, they also gradually learned that, regardless of what the out-group does, they
are better off holding on to their endowments. This possibility will be discussed
further, in light of the results of experiment 2.
Experiment 2
The focus of experiment 2 is on whether the contribution levels by out-group
members affect the way they are being perceived by in-group members. As mentioned
in the introduction, it is likely that the perception of other participants’ motivations is
involved in the reciprocation process. Since experiment 1 supported the hypothesis of
between-team reciprocation, the aim of experiment 2 was to find whether participants
would attribute different motivations to out-group members according to their
different levels of contribution.
Bornstein and Ben-Yossef (1994) argued that a high contribution level in the

IPD game may reflect either competitive (maximizing-relative-difference) motivation,
or an increase in “patriotism” (maximizing-ingroup-gain), or both. Comparing the
IPD game with a parallel single-group PD game Bornstein and Ben-Yossef found
contribution twice as high in the IPD game. Participants also rated in-group and outgroup members on several motivation scales. Consistent with the high contribution in
the IPD condition participants there rated both in-group and out-group members as
more competitive, more “patriotic” and less motivated by self-gain than in the singlegroup PD.
In the current experiment participants were asked to rate out-group members
on four motivation scales: self-gain interest (maximizing individual gains),
“patriotism” (maximizing in-group gains), competitiveness (maximizing relative gain
difference) and intergroup cooperation (maximizing the joint gain of both teams). It

21


seemed reasonable to measure attributions on the ‘max-joint’ motivation scale since
between-team reciprocation can bring about an outcome that is good for both teams.
The expectation was that higher levels of contribution by out-group members would
lead participants to perceive them as more competitive and “patriotic”, and that lower
contribution levels would lead to perceiving them as more cooperative vis-a-vis the
in-group.6
Measuring the motivations that participants attribute to out-group members
can help clarify the reasons behind the use of between-team reciprocation. For
instance, the predictions just mentioned assume that participants are equally focused
on the two aspects of reciprocal strategies – reciprocating cooperative choices and
retaliating against competitive choices. However, participants may be more focused
on one aspect of reciprocation than the other. In that case, the manipulation of outgroup contribution level may affect the ratings of some motivations but not of other.
This is especially likely if participants engage in reciprocation because they are
motivated to get as much as the out-group (motivated not to “lose” the game). If this
is the case, participants may be more concerned with the out-group’s competitive
motivation than with its cooperative motivation, and the manipulation of out-group

contribution will affect ratings on the competitive (max-rel) scale more than on the
cooperative (max-joint) scale.
Experiment 2 addressed the issue of out-group perception by manipulating the
variability in contribution behavior of individual players within the out-group, in
addition to manipulating their average contribution level. The literature on inter-group
perception has documented a tendency to view out-group members as more similar to
each other than in-group members. This perceptual bias termed “the out-group
homogeneity effect” (Linville & Fisher, 1998; Quattrone, 1986) should be accentuated

22


when there is conflict of interests between the groups (Judd & Park, 1988). It is
therefore interesting to test whether, in the IPD game, participants pay attention to the
behavior of individual out-group members, and, consequently, attribute different
motivations to different out-group members depending on their individual conduct.7
Therefore, experiment 2 entailed a two-factor design. One factor involved the
average contribution level of the virtual out-group that was either high (70%
contribution throughout the 40-period game) or low (30% contribution). The high
contribution (Hi-Cont) and low contribution (Low-Cont) conditions were crossed with
two levels of out-group variability, a low variability level (Low-Var) and a high
variability level (Hi-Var). In the Low-Var conditions the three out-group players
contributed at almost the same rate. In the Hi-Var conditions the virtual out-group
players contributed at different rates.
It was hypothesized that contribution levels would increase over time in the
‘Hi-Cont’ conditions (where out-group contribution was high) and that contribution
would decrease with time in the ‘Low-Cont’ conditions (where out-group contribution
was low). In other words, it was hypothesized that there will be an interaction effect
of time with the ‘Average contribution’ factor. In addition, in line with between-team
reciprocation, a positive lagged correlation between participants’ decisions and those

of the out-group was expected. (If between-team reciprocation is strong enough, then
one can also predict a main effect of the ‘Average contribution’ factor, with more
contribution by participants in the Hi-Cont conditions then in the Low-Cont
conditions. However, this effect can be greatly influenced by the initial levels of
contribution in the different conditions. The speed with which participants react to the
level of contribution of the virtual out-group can also influence the likelihood of
attaining a main effect for the ‘Average contribution’ factor.)

23


Following experiment 1, one can expect participants to have good recollection
of out-group contribution behavior. To the extent that participants differentiate
between out-group members who contribute a lot and those that contribute at a low
rate, one can also expect the participants to attribute different motivations to outgroup members according to their contribution levels (as discussed earlier). This
effect should manifest itself in main-effect differences between the Hi-Cont and LowCont conditions. To the extent that participants pay attention to the behavior of
individual out-group members, this effect should also manifest itself in differences
between motivation ratings for indivudual out-group members.
Experimental Design and Procedure
The participants were 96 undergraduate students at the Hebrew University of
Jerusalem (24 in each condition), with no previous experience with the task.
Participants played 40 rounds of the IPD game. Each three-person team played against
a virtual out-group of one of the four kinds described above (Hi-Cont/Hi-Var, HiCont/Low-Var, Low-Cont/Hi-Var or Low-Cont/Low-Var). The exact numbers of
contributions by each out-group player during the entire 40 rounds were as follows:
Hi-Cont/Low-Var condition: 29, 28 and 27 contributions; Hi-Cont/Hi-Var condition:
36, 30 and 18 contributions; Low-Cont/Low-Var condition: 13, 12 and 11
contributions; Low-Cont/Hi-Var condition: 22, 10 and 4 contributions. In all
conditions the initial levels of contribution of the virtual out-groups were medium.
This means that the differences between the level of contribution of the high
contributing virtual out-groups and of the low contributing virtual out-groups

increased as the game progressed.8
Following the last round, the points were added up by the computer and
cashed in at the rate of IS 1 for 7 points (the average participant earned IS 25.8, or

24


about $7.5). Participants filed out a post-experimental questionnaire in which they
were inquired about the number of contributions made by the out-group players
during the 40-round game, and their perceptions of the out-group players’
motivations. Participants were told that they would be rewarded for being accurate in
recalling contribution levels.9
In all other details the experimental procedure was identical to that used in
experiment 1.
Results
Contribution rates: The 40 rounds of play were divided into 8 blocks of five
rounds each, and the mean proportion of contributions per block was calculated.
These means appear in Figure 5. Again, the unit of analysis was the average
proportion of contribution by the three members of the same team.
These average proportions were analyzed by a 2 (‘Average contribution’: ‘HiCont’ vs. ‘Low-Cont’) by 2 (‘Variability’: ‘Hi-Var’ vs. ‘Low-Var’) by 8 (block) mixed
design ANOVA. This analysis revealed a significant main effect of ‘Variability’
(F(1,28)=4.10, p=0.053) and a significant interaction between the ‘Average contribution’
and ‘Variability’ factors (F(1,28)=4.51, p<0.05). The main effect for ‘Average
contribution’ was not significant.
Both the main effect for variability and the interaction effect were not
predicted by the hypotheses. Comparing the mean levels of contribution in the 4
experimental conditions reveal that as expected these means were relatively high in
the Hi-Cont conditions (Hi-Cont/Low-Var - 36.7%; Hi-Cont/Hi-Var - 36.3%) and low
in the Low-Cont/Low-Var condition (22.3%). However, for some reason, participants
in the Low-Cont/Hi-Var condition contributed at a high rate of 40.1% (similar to those

in the Hi-Cont conditions). This resulted in the significant effects above.

25