Friend or Foe: Conflict vs. Cooperation in High Stakes Prisoners Dilemmas

1
Friend or Foe Conflict vs. Cooperationin High
Stakes Prisoners Dilemmas

• Peter Coughlan
• William Gates
• Elda Pema

2
Prisoners Dilemma in Defense Economics

• Classic example Arms races conflict escalation
  (Schelling 1980, Brams 1985, Sandler 1992, etc.)

Nation 2
Disarm / De-Escalate
Arm / Escalate
Disarm / De-Escalate
3 3
1 4
Nation 1
Arm / Escalate
4 1
2 2
3
Prisoners Dilemma in Defense Economics

• Modern example International cooperation and the
  global war on terror (Hartley Sandler 1995)

Nation 2
Punish State Sponsors of Terrorism
Dont Punish State Sponsors of Terrorism
Punish State Sponsors of Terrorism
3 3
1 4
Nation 1
Dont Punish State Sponsors of Terrorism
4 1
2 2
4
Prisoners Dilemma in Defense Economics

• Pareto-improving moves are possible but they
  hinge on cooperation
• Q what factors influence cooperation?
• In one-shot prisoners dilemma experiments
  cooperation rates are about 33
• Experimental evidence limited to situations where
  stakes are relatively low or imaginary
• How would higher stakes affect behavior in
  prisoners dilemma situations?

5
Friend or Foe

• June 2002, Game Show Network
• 2 seasons first 40 shows produced before going
  on air 65 more shows produced after the first 40
  aired
• Potential for learning in second season
• Each show consists of 6 strangers randomly split
  into 2 groups (G1, G2)
• Each person from G1 selects someone from G2 as
  partner
• If multiple players select the same person, the
  choosee picks a partner from the choosers

A B C
D E F
strong
G1
G2
weak
medium
6
Friend or foe

• The three pairs play trivia rounds in isolation
  contributing to a trust fund of x
• In a final stage decide to share as friends or
  fight over as foes
• Each player can press a hidden button to choose
  foe and the winnings are distributed similar to
  the prisoners dilemma game
• However, the stakes are quite high potential
  earnings vary from 200 to 22,000. Actual
  winnings vary from 200 to 15,000 with an
  average of 3,300 per game.

Friend
Foe
Friend
Foe
7
Empirical evidence from Friend or Foe

• Oberholzer-Gee et al. (2004)
• About 55 of players choose friend even at high
  stakes
• Stakes do not affect the play strategy
• Find evidence of players guessing friendliness
  of opponent based on observable characteristics
  in second season, players who are expected to
  cooperate fare better (monetarily) than players
  more likely to choose foe
• List (2006) (uses data from 39 shows)
• Stakes insignificant
• Women, whites, older players, and
  non-Californians more likely to cooperate
• Partner attributes do not influence strategy
  (they influence earnings)
• Statistical discrimination at the partner
  selection stage
• women, whites, and non-Californians more likely
  to be chosen

8
This study

• Take a closer look at how teams are formed and
  how signals are interpreted
• In reality teams form endogenously, and
  experimental data can rarely replicate this
  aspect
• Focus on the selection of opponents
• Do people choose opponents based on their
  expected contribution, or on their predicted
  probability of cooperating?
• What characteristics identify cooperative
  players?
• and communication that leads to cooperation
• At the end of the trivia rounds, players make a
  closing argument that explains their upcoming
  choice, plead with the opponent to cooperate, or
  explain previous behavior
• What arguments are effective in inducing
  cooperation?
• Do players deceive in closing statements?

9
Findings

• Relative contribution to total earnings does not
  affect play strategy
• i.e. preferences for fair division do not seem
  to arise from the perceived fair return to own
  contribution
• Individual characteristics influence strategy
• People tend to pick as opponents individuals who
  are more likely to cooperate
• Whites are more likely to get picked
  Californians, people in managerial, sales, and
  administrative occupations are less likely to be
  picked. (correct stereotyping)
• Conditional on cooperation, people in
  professional occupations are more likely to be
  chosen as partners (expected contribution).

10
Alliances

• At the team level the similarities between
  players affect cooperation
• Teams where both partners are female, non-CA,
  more mature are more likely to cooperate
• Pairs that match on more than one demographic
  characteristic are more likely to cooperate
• Strength of alliance influences strategy teams
  with both players expressing some willingness to
  be matched (strong picks) are more likely to
  cooperate

11
Cooperation Communication

• Arguments positively correlated with the
  probability of cooperation
• I picked you
• I trust you
• Im honest
• Arguments negatively correlated with the
  probability of cooperation
• I need the money
• Half is good enough
• Arguments opponents make that induce cooperation
• (Opponent says) Im honest
• People choose foe when opponent shows his/her
  hand
• (Opponent says) Lets be friends

12
Discussion

• These findings indicate that stakes and relative
  contribution may not be as important as
  previously thought
• From a theoretical standpoint, cooperation rates
  are still too high
• It appears that individual characteristics and
  communication are critical in explaining these
  cooperation rates

13
Conclusions Implications

• International cooperation resembles a multi-stage
  prisoners dilemma game with non-overlapping
  generations
• foreign policy often designed and implemented by
  different administrations
• Can use Friend or Foe evidence on communication
  to understand cooperation in the international
  arena (e.g. first and second Gulf wars)
• Implications regarding alliance formation and the
  role of confidence building
• NATO which countries should be joining next?
• What kind of communication assures cooperation?