Bargaining and Coalition Formation Dr James Tremewan - - PowerPoint PPT Presentation

Bargaining and Coalition Formation

Dr James Tremewan (james.tremewan@univie.ac.at) Culture, Biology, and Bargaining

Culture and Bargaining

Culture and Bargaining

• ”Bargaining and Market Behavior in Jerusalem, Ljubljana,

Pittsburch, and Tokyo”, Roth et al (1991).

• Classic paper for cross-cultural research. Discusses in depth and

develops methodology to deal with challenges in cross-cultural experiments.

• Compares behaviour in four cities in different countries (no

definition of ”culture” and number of countries too small to test for what elements of culture are important).

• ”Co-operation, Reciprocity and Punishment in Fifteen

Small-scale Societies”, Henrich et al (2001)

• Compares behaviour in 15 hunter-gatherer and nomadic

societies.

• Explains differences by economic structure of societies

(culture=economic/social institutions?).

2/35

Roth et al (1991)

Roth et al (1991)

• Ultimatum game played in four locations: Jerusalem, Ljubljana,

Pittsburgh, Tokyo:

• Subjects were all university students.
• 10 rounds of stranger-matching as either proposer or receiver.
• Anonymous, learn results of own negotiation after each round.
• Paid one random round.
• Challenges of cross-cultural experiments:
• Experimenter effects: uncontrolled procedural or personal

differences among experimenters.

• Language effects: differences caused by translations (even if

instructions in same language, some words may have different connotations).

• Currency effects: same numbers ⇒ different purchasing power,

same purchasing power ⇒ different numbers (scale, rounding).

3/35

Roth et al (1991)

Experimenter Effects: Solutions

• Each experimenter ran at least one session in Pittsburgh:
• Enabled better coordination of operational procedures.
• Data could be used to test for experimenter vs between-country

effects: any experimenter effect should show up in Pittsburgh data as well as in cross-country data.

• Alternative: have the same experimenter present at all sessions

(e.g. Cameron et al, 2009). But may be language problems, or reaction to presence of foreigner.

4/35

Roth et al (1991)

Translation Effects: Solutions

• Translations:
• The experimenter responsible is a national of ”other” country

and linguistically and culturally fluent in American English.

• Phrases chosen that could be faithfully translated (i.e. avoid

words with no equivalents, or heavy connotations).

• Control for translation differences:
• Also ran a market experiment using same vocabulary as

bargaining experiment.

• Claim: if between-country differences found in bargaining but

not in market game, this places ”upper bound” on effect of translation.

• Alternative: ”back translation” where second translator

independently translates the instructions back into original language which can be compared directly to original to identify important differences.

5/35

Roth et al (1991)

Currency Effects: Solutions

• Purchasing power:
• Potential payoff in Pittsburgh sessions ranged form $10 to $30.
• In other countries payoff chosen to be on the high side of

purchasing power equivalent of $10.

• If results from other countries fall outside range of results in

Pittsburgh then strong evidence for other factors.

• Numerical issues: pie of size 1000 tokens in all countries

(different token/local currency exchange rates), with proposals limited to multiples of 5.

6/35

Roth et al (1991)

Results: Proposer behaviour 1

• Modal offer in first round: 500 in all countries.
• Modal offer(s) in tenth round:
• 400 in Israel.
• 400 and 450 in Japan.
• 500 in Yugoslavia and United States.
• Downward trend in Israel and Japan, no significant trend in

Yugoslavia or US.

1I will not discuss statistical tests here. Some of the assumptions are a

bit dodgy, such as treating all observations in given round as independent despite stranger-matching (strictly speaking only first-round observations are independent).

7/35

Roth et al (1991)

Results: Responder behaviour

• Overall proportion of rejected offers similar across countries

(large differences in final round, but large fluctuations and no

• bvious trend).
• Pairwise between-country comparisons of acceptance rates by
• ffer size (see next slide) suggests ”low-offer” countries (i.e.

Israel and Japan) have higher rates of acceptance for all offers.

• Modal offer in tenth round consistent with proposer maximising

expected earnings given responder behaviour.

8/35

Roth et al (1991)

Results: Responder behaviour

9/35

Roth et al (1991)

Discussion

• Bargaining behaviour in all countries differs markedly from SPE

prediction.

• Significant differences between countries in both proposer and

responder behaviour.

• Authors say differences not related to ”toughness” in bargaining:

in tougher bargaining one would expect lower offers, but higher rates of rejection.

• Differences appear to be related to responders’ differing ideas

about what is a ”reasonable offer.” In unfamiliar environment it takes time for proposers to identify what is socially acceptable.

• Criticisms:
• City or country differences?
• ”God-of-the-gaps” definition of culture: cross-country

differences that can’t currently be explained by other factors.

10/35

Henrich et al (2001)

In Search of Homo Economicus: Behavioral Experiments in 15 Small-Scale Societies, Henrich et al (2001)

• Ultimatum game played in 15 (17) small-scale societies:
• Unlike previous studies, subjects NOT university students.
• Subjects play only one game in one role.
• Anonymous.
• Size of pie roughly equal to one or two day’s wages (mostly paid

in money, but sometimes tobacco or other goods).

• Controls for variations in implementation not so good as

previous paper for practical reasons (e.g. few Sangu herders are linguistically and culturally fluent in American English), but

• Experiments run from identical protocols.
• Stake sizes as similar as possible.
• Experimenter effects tested where possible (none found).

11/35

Henrich et al (2001) 12/35

Henrich et al (2001) 13/35

Henrich et al (2001)

Results: General observations

• No society behaves in line with SPE prediction:
• Lowest mean offer: 0.26; lowest modal offer: 0.15.
• Typically some non-zero offers rejected (interestingly there are

some groups which never rejected offers, but it is not clear that they would not have rejected smaller offers - need strategy method).

• Much wider variation in behaviour than in previous WEIRD2

studies:

• Means: 26-58% (cf. weird 40-50%).
• Modes: 15-50% (cf. weird 50% in first round).
• Rejection behaviour:
• Frequent rejections of offers over 50% (Au and Gnau).
• Zero rejection of low offers (Tsimane).
• Offers not income maximising given rejection probabilities (but

unlike Roth et al. no opportunity for learning).

2Western Educated Industrialized Rich Democracies, see Henrich,

Heine, and Norenzayan, 2010.

14/35

Henrich et al (2001)

Results: Relationship between proposer behaviour and economic structure

• Payoff to Cooperation (PC): how important is cooperation in

economic production (low: Machiguenga who are economically independent at family level; high: Lamelara hunt whales in 12+ person canoes).

• Market integration (MI): how much do people rely on market

exchange in daily life.

• PC and MI measured using indices constructed by general

discussion in research team before data was gathered.

• Both PC and MI positively related to group mean offers (highly

significant), jointly explaining 68% of variation.

• Robust to inclusion of individual measures (sex, age, wealth...),

none of which were significant.

15/35

Henrich et al (2001) 16/35

Henrich et al (2001) 17/35

Henrich et al (2001)

Discussion

• When faced with novel situation, subjects seek cues from

previous related experiences.

• Among Au and Gnau accepting gifts commits one to reciprocate

and establishes subordinate status, hence rejection of high offers.

• Sharing of results of hunt standard amongst Ache despite no

associated punishment, hence high offers and lack of rejections.

• Authors say their results suggest that preferences are not

exogenous, but shaped by social and economic interactions... questions calculation of welfare effects of policy changes that do not take this into account (e.g. introducing markets may reduce cooperative behaviour or result in less concern for others e.g. Armin Falk’s mice).

18/35

Biology and Bargaining

Biology and Bargaining

• ”Heritability of Ultimatum Game Responder Behavior”, Wallace

et al (2007).

• The effect of testosterone on behaviour in the UG:
• Burnham (2007).
• Eisenegger et al (2010).
• Zethraeus et al (2009).
• ”Diminishing Reciprocal Fairness by Disrupting the Right

Prefrontal Cortex”, Knoch et al (2006).

19/35

Wallace et al (2007)

Wallace et al (2007)

• To what extent is responder behaviour3 in the UG influenced by

social/environmental and genetic factors?

• Standard approach to answer this question is a Twin Design:
• Monozygotic twins share identical genes; Dizygotic twins have

imperfectly correlated genes.

• Both types of twins share the same environment.
• A greater correlation in behaviour between monozygotic twins

than for dizygotic twins implies a genetic influence on behaviour.

• Variation in behaviour can be broken down into genetic,

common environmental, and non-shared environmental effects.

• This study uses 71 DZ and 258 MZ pairs of twins from the

Swedish Twin Registry at the Karolinska Institutet.

• All subjects made decision as Proposer, then decisions whether

to accept or reject every multiple of 10% (”strategy method”).

3Almost no deviation from 50-50 split in proposer behaviour in

Sweden.

20/35

Wallace et al (2007)

Wallace et al (2007): Findings

No statistically significant difference between MZ and DZ acceptance thresholds.

21/35

Wallace et al (2007)

Wallace et al (2007): Findings continued

• Spearman rank correlations:
• Monozygotic twins: 0.39
• Dizygotic twins: -0.04
• Difference highly significant (P < 0.01, two sided)
• Point estimates of standard threshold model to decompose

contributions to variation in behaviour:

• Genetic contribution: 42%
• Common environmental contribution: 0%
• Non-shared environmental contribution: 58%
• Conclusion: genetic influences are important determinants of

rejection behaviour in the ultimatum game.

22/35

Testosterone and the Ultimatum Game

Burnham (2007): High-testosterone men and UG responder behaviour

• Individual testosterone levels are under genetic influence -

perhaps an explanation for Wallace et al findings?

• Testosterone linked to agression, status seeking,

reputation-management system, willingness to engage in conflict.

• If men interpret low UG offers as a challenge, those with higher

testosterone more likely to reject.

• All subjects (N=26) made decision as Proposer, then decisions

whether to accept or reject either possible offer ($5 or $25 out

• f $40).
• Testosterone level measured from saliva on three

non-experimental days at same time as UG decisions (14.00 - when rate of change of diurnal testosterone cycle is lowest).

23/35

Testosterone and the Ultimatum Game

Burnham (2007): Findings

• Subjects who rejected $5 (N = 6) had higher testosterone levels
• n average than those who accepted (N = 20), P < 0.01.
• Proposers offering $25 had higher testosterone levels on average

than those offering $5, but not difference not significant.

24/35

Testosterone and the Ultimatum Game

Eisenegger et al (2010)

• Burnham (2007) identified correlation but not causation - need

randomized manipulation of testosterone level.

• Double-blind experiment, between-subject design, pie=10 Swiss

Francs.

• Subjects make decisions as either Proposer or Receiver
• Offer either 5, 3, 2, or 0 Swiss Francs.
• Receiver accepts or rejects (not strategy method).
• Three decisions with stranger matching (no feedback).
• Testosterone or placebo administered sub-lingually.
• 121 female subjects (mean age=25 years).
• Only female because required quantity and time for effect of

treatment currently unknown for males

• Subjects screened to ensure not pregnant, psychotic, etc.

25/35

Testosterone and the Ultimatum Game

Eisenegger et al (2010)

• Competing hypotheses for Proposers:
• ”Folk” hypothesis: T increases aggressive, egoistic behaviour (T

lowers offers).

• ”Social status” hypothesis: T increases concern for social

status4 and thus avoidance of rejection, which is a threat to status (T increases offers).

• Also ask for belief about whether treated with T or placebo:
• Social status hypothesis implies T increases offers.
• Belief that one has been treated with T could be associated

with lower offers if subjects believe Folk hypothesis (excuse for selfish behaviour?).

4Which may be the underlying cause of aggressive behaviour, etc.

26/35

Testosterone and the Ultimatum Game

Eisenegger et al (2010): Results

• Main results are from ANOVA model relating subject’s mean
• ffer to actual treatment and belief.5
• Subjects who received T made higher offers (T: $3.90, Placebo:

$3.40, ANOVA, N=60, p=0.031).

• Subjects who believed they received T made lower offers (bT:

$3.08, bPlacebo: $3.88, ANOVA, N=60, p=0.006).

• No significant correlation between treatment and beliefs, nor

interaction effect.

• No significant correlation effects on responder behaviour.

5Justification for this model and further robustness tests are given in

”Supplementary Information”.

27/35

Testosterone and the Ultimatum Game

Zethraeus et al (2009)

• Double-blind experiment, between-subject design, pie=SEK 400.
• Subjects (203 Swedish post-menopausal women) were treated

with testosterone, estrogen, or placebo for four weeks.

• All subjects make decisions as Proposer and Receiver

(acceptance threshold, i.e. strategy method).

• Testosterone levels 4 times higher in treated group.
• No significant effect found of testosterone (or estrogen) in UG

behaviour6 (or risk preferences, behaviour in a dictator and trust game).

6Again very little deviation from 50-50 split in Proposer decision.

28/35

Testosterone and the Ultimatum Game

Discussion

• Contradictory results in studies could result from:
• Different experimental procedures (playing one or both roles,

strategy method, etc.)

• Different subject pools (testosterone may have different effects

in males and females, which may depend on age).

• Zethraeus et al (2009) conclude that:
• Previously identified correlation between endogenous

testosterone levels and behaviour are due to some other underlying biological factor.

• Results may be spurious, with lack of countervailing evidence

due to publication bias (i.e. 20 studies are run, and only one finds significance at %5 level, but this is the only ”interesting” result, so only one to be published.)

29/35

Knoch et al (2006)

Knoch et al (2006)

• Sanfey et al (2003) found that the right dorsolateral prefrontal

cortex of UG responders was activated more by unfair offers.

• But is the R DLPFC crucial for determining response to

fair/unfair behaviour?

30/35

Knoch et al (2006)

Knoch et al (2006)

• Brain activity can be disrupted by repetitive transcranial

magnetic stimulation (rTMS): Electric current passed through metal coil placed on scalp to produce magnetic field which affects firing of neurons.

• Each subject (n=52) played as responder in UG 10 times in one
• f three treatments: rTMS of left DLPFC, rTMS of right

DLPFC, or ”sham” rTMS.

• Proposer could offer either 4 or 10 out of 20 Swiss Francs.
• Subjects also asked about perceived fairness of offers on a scale

from 1 to 7.

• In 10 games Proposer chooses offer; in 10 computer randomly

chooses for Proposer - tests ”negative reciprocity” vs ”inequity aversion.”

31/35

Knoch et al (2006)

Human offer condition

32/35

Knoch et al (2006) 33/35

Knoch et al (2006)

Computer offer condition

34/35

Knoch et al (2006)

Discussion

• Disruption of right DLPFC reduces fair behavioural responses

but not fairness judgements.

• Subjects in left DLPFT and sham treatments took longer to

respond to unfair than fair offers, but this effect not present in right DLPFC treatment.

• Conflict between selfish and fairness related motives takes time

to resolve. Right DLPFC important in over-riding selfish behaviour in favour of fairness.

• Much smaller treatment difference for computer generated offers

suggests right DLPFC important for reciprocity motives rather than simple inequity aversion.

35/35