Group versus individualdecision-makingIs there a shift?

Attila AmbrusBen Greiner

Parag Pathak

Motivation• Many important decisions in group context:

committees, governing bodies, juries, partners, teams, families.

• If these decisions cannot be explained as aggregation of individual choices, need to rethink if we can take individuals as the unit of analysis.

Motivation (cont.)• Large literature in social psychology, and recent

literature on economics: do people behave differently in groups than in isolation?

• Intellective tasks (Laughlin, 1980): informational reasons why people behave differently.

• Nonintellective tasks (no normative criterion for evaluating decisions): no informational reason.

Motivation (cont.)• Stoner (1961): group shifts in lottery choices;

cautious shifts; later research finds risky shifts.

• Selfish shift: Pylyshyn et al. (1966); later research verified it in PD games, centipede games, ultimatum games, gift-exchange games, ultimatum games.

• Moscovini and Zavalloni (1969): group polarization/discontinuity effect.

General explanations for group shift• Social comparison theory (Levinger and Schneider

(1969)): people behave in group settings differently because motivated to perceive and present themselves in socially desireable way.

• Persuasive argument theory (Burnstein et al. (1973)): pool of arguments in one direction more persuasive; alternative: people with certain preferences more persuasive.

Explanations in particular contexts• In-group vs out-group sentiments (Tajfel et al

(1971), Charness, Rigotti and Rustichini(2007).

• Identifiability explanation/personal responsibility (Wallach et al. (1964))

• Eliaz, Ray and Razin (2005): group shift for non expected utility maximizers in lottery choices along lines of Allais paradox.

Our main point• Point out that whether group shift cannot be

mechanically determined based on comparing distributions of individual and group decisions.

• Need theory of how individual opinions aggregated at group level, since different theories predict different distributions of group choices, for given distribution of individual choices.

Main point (cont.)• Example: single-peaked preferences and majority

voting: median voter’s decision adopted by group (Moulin, 1980).

• If distribution of preferences in population asymmetric: population median differs from population mean – average group decisions differ from average individual decisions in systematic manner.

• Right question: is there shift wrt group decisions predicted by aggregation theory (median if single-peaked preferences)?

Experimental Design• Gift-exchange game (Fehr, Kirchsteiger & Riedl,

1993; Brandts and Charness, 2004)– First mover sends amount between 1 and 10 tokens to

second mover– Amount tripled on the way– Second mover observes amount, then decides about how

many tokens (1-10) to send to first mover– Amount tripled on the way

• Lottery choices (Holt & Laury, 2002)– Decision-maker chooses between pairs of lotteries, e.g.

A: $11.50 with 60% and $0.30 with 40%OR B: $ 6.00 with 60% and $4.80 with 40%

– 10 choices over the full range of percentages switching point between A and B measure of risk attitude

Experimental DesignN groups of 5 persons each

• Gift exchange• Gift exchange• Lotteries

--> Reshuffle• Gift exchange• Gift exchange• Lotteries

--> Reshuffle• Gift exchange• Gift exchange• Lotteries

6 individualfirst moverseach making

N gifts

each decision first individually, then jointly as a groupTreatment A: VotingTreatment B: Free discussion

Design (cont.)• Important that group and individual choices

solicited for same decision (both apply to same set of people).

• Charness et al. (2007), Chen and Li (2009), Sutter (2009) show individual decisions effected by applying only to the individual or to a group containing the individual.

Treatments I.• No deliberation: group members cannot talk

to each other, but see each other; group decision determined by public majority voting.

No possibility for persuasion, but social comparison theory might still apply.

Treatments II• Deliberation: unlimited group discussion; free

to decide how to make group decision. Here persuasive argument theory can be tested, too.

The two most popular treatments in literature; differ in multiple dimensions so we analyze them separately. Possibility of in-between treatment.

Hypotheses

• Mean hypothesis: β1= β2= β3= β4= β5

strong: β1= β2= β3= β4= β5= 1

• Median hypothesis: β1= β2= β4= β5= 0, β3 > 0strong: β1= β2= β4= β5= 0, β3= 1

• Shift hypothesis: α > 0

Results

groupsindividuals

Binary lottery choicesgroups

individualsGift exchange 2nd mover decisionsDifferent groupsParticipantsSessions

3301801650900

6636330180

331855 + 1130 + 6

41

DeliberationNo deliberation

Table 6 missing• See it in paper

Gift Exchange Results

6636N0.870.96R2

0.020.23*highest0.12-0.03fourth0.51***0.76***third0.39***0.02second0.020.01lowest-0.15-0.36intercept

Delib-eration

No deliberation

66360.880.99

plus Phase Fixed Effects

0.100.23*0.03-0.010.24***0.73***0.61***0.050.060.00

Delib-eration

No deliberation

Interpretation of results• Findings provide evidence against social

comparison theory in contexts we investigated.

• Findings in line with predictions of persuasive argument theory.

• Extreme group members don’t seem to have influence on group decisions; people close to the median can have an effect, to different extent in different directions.

Gift Exchange ResultsFigure 1: Comparison of Group Trust Game Decision to Mean and Median

0

1

2

3

4

5

6

7

8

9

10

0 1 2 3 4 5 6 7 8 9 10

Predicted Decision

median

mean

Lottery resultsShare of "risky" choices

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

1 2 3 4 5 6 7 8 9 10

Lottery choice

No deliberation - Individuals

No deliberation - Groups

In 96% casesMedian becomes group

Share of "risky" choices

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

1 2 3 4 5 6 7 8 9 10

Lottery choice

Deliberation - Individuals

Deliberation - Groups

Lottery results

Lottery results

1.001.000.971.001.001.00101.001.000.961.001.001.0090.910.940.851.001.000.9280.820.580.581.000.780.6370.880.150.270.780.330.3960.970.030.050.940.110.2251.000.000.021.000.000.0041.000.000.011.000.000.0031.000.000.011.000.000.0021.000.000.011.000.000.001

gr=medgroupsind.gr=medgroupsind.LotteryDeliberationNo deliberation

• Share of “risky” choices

Lottery Results: Gender

1620162016501650N0.810.810.810.81R2-Adj

0.000-0.002age-0.0060.00

5econ

0.050***0.047***maleIndividual decision

330N0.92R2-Adj

-0.12fifth *female0.08*fifth-0.39**fourth *female0.30***fourth-0.05third *female0.55***third0.68***second *female-0.02second-0.12first *female0.09first

Group decision

Summary and Conclusions• Importance of reexamining results on group shift,

relative to theories of preference aggregation.• Median hypothesis can explain at least part of the

group shift findings.• Can explain why different shifts observed in same

type of choice (risky vs safe shifts)• Can explain why group shifts tend to occur in

direction of original inclination of subjects (Brown, 1986), and why less likely to occur when two roughly equal sized sets of people predisposed in two directions (Sunstein, 2000).