4. Public Good Games Experiments

Question 1

General results of public goods games

Answer 1

About 40-60% cooperation in first rounds but there is decay over time

Question 2

What are possible reasons for why people contribute in public goods games?

Answer 2

-error, confusion
-strategic reasons, repeated game effects, reputation
-warm glow, altruism (unconditional cooperation)
-conditional cooperation

Question 3

Which experiment is used to measure error as a reason for public good game contribution? Describe the set up

Answer 3

Keser 1996
Design with interior solution
Token 1-13: payoff from keeping> payoff from public good
Token 14-20: payoff from keeping< individual return from PG
NE: contribute 7 tokens, keep 13

Question 4

Results of Keser 1996

Answer 4

People over contribute but by the end of 25 periods are very close to NE. If we suppose errors are equally likely in both directions, then on average they should cancel out. This shows contribution isn’t likely due to errors

Question 5

Describe set up of Andreoni 1995

Answer 5

Idea- subtract out the incentives for kindness, leaving confusion as only explanation for cooperation
3 treatments
Regular- standard PG
Rank- subjects play PG but get paid based on their rank- transforms into zero sum game (no gains from cooperation) free riding still dominant strategy
RegRank- same as regular but subjects also receive info about their rank- rank as a level of kindness

Question 6

Results of Andreoni 1995

Answer 6

On average 75% are cooperative and half of these are confused about incentives while half understand free riding but choose to cooperate out of some form of kindness. The focus on learning in experimental research should shift to include studies of preferences of cooperation

Question 7

Describe the set up of Andreoni 1988

Answer 7

Repeated PG game with partners and strangers with a restart after 10 rounds. Attempt to disentangle the difference between learning and strategy

Question 8

Results of Andreoni 1988

Answer 8

Strangers contribute more than partners. Jump in both treatments when restated suggesting contributions are strategic. Note results in stranger treatment on only one independent observation

Question 9

What did Croson 1996 (replication of Andreoni 1988) find?

Answer 9

Cooperation higher in partners and there is a jump in both after the restart

Question 10

Describe the set up of Weimann 1994

Answer 10

PGG where each player plays against 4 fictitious characters whose contributions were made by the experimenter to see whether people are conditional cooperators

Question 11

Weimann 1994 results

Answer 11

In E6 these phantoms invested 89.75% to PG- mean investment was 53% falling over rounds
In E7 phantoms invested 15.75%- mean investment was 33.4% falling over rounds

Question 12

Definition of deception

Answer 12

Intentional misinformation of subjects and use of computers/ confederates without revealing this to subjects

Question 13

Definition of deliberate ambiguity

Answer 13

Witholding info about research hypotheses, full range of experimental conditions or some experimental details

Question 14

What do Charness Et Al 2022 consider grey areas of deception?

Answer 14

-surprise restart
-sub group rematching
-deliberate reliance on misinterpretation
-unexpected data use
-unknown/unpaid participation

Question 15

Arguments against deception

Answer 15

-it can often be avoided
-use of deception slows down methodological innovation
-loss of control and loss of internal validity, contamination of subject pool

Question 16

Quote from Hey 1998 on deception

Answer 16

If the subjects don’t believe what the experimenter tells them then the experimenter no longer knows what is being tested in an experiment

Question 17

How can virtual players be used without deception?

Answer 17

Houser & Kurzbon 2002 disentangle the role of confusion and social preferences by replacing real players with pre programmed computers and tell subjects

Question 18

Describe the set up of Croson 2000

Answer 18

-repeated linear PG games
-subjects are asked to estimate others contributions
-beliefs were incentivised. The better they estimate the higher the additional money reward

Question 19

Results of Croson 2000

Answer 19

High correlation between beliefs and individual behaviour. Belief elicitation has an influence on contribution behaviour: lower contributions with belief elicitations

Question 20

Pros and cons of paying accuracy of beliefs

Answer 20

+subjects have an incentive to take the task seriously and to state correct beliefs
-subjects might have an incentive to make behaviour predictable
-subjects might think about the decision situation differently
-hedging where subjects falsely state pessimistic belief to insure against the risk of true belief being wrong

Question 21

Set up and results of Gachter & Renner 2010

Answer 21

Impact of incentivised vs non incentivised belief elicitation in a related PGG
-belief accuracy is slightly higher when beliefs are incentivised
-the distribution of beliefs as well as the relationship between contributions and beliefs are unaffected
-eliciting incentivised beliefs increases contribution levels relative to benchmark treatment without belief elicitation

Question 22

Strategy method

Answer 22

Essentially a contribution table where subjects answer for every option possible in terms of a contribution from an imaginary other player

Question 23

Pros and cons of strategy method

Answer 23

+data for all info sets
+more info about a player and their motivation
+lower costs
-decisions are “cold” - hypothetical
-subjects have to consider all possible subgames not only those which arise in games

Question 24

Describe set up of Bayer, Renner & Sausgruber 2013

Answer 24

Two treatments
-standard PGG with no info feedback on others contributions or others payoffs- impedes herding
-learning treatment that induces subject confusion. Subjects are asked to choose a number between 0-20 and learn their payoff after each period

Question 25

Results of Bayer, Renner, Sausgruber 2013

Answer 25

-contribution patterns are qualitatively very similar
-contribution rates 54%/56% in standard/learning so confusion created no systematic upwards bias
-steeper decline in standard
-strong patterns of conditional cooperation in standard. 48% of subjects exhibit +ve correlation between own and group members last contribution (only 5% in learning). Note subjects must infer others contributions from their own payoffs, these subjects can’t just imitate others

Question 26

Utility function of public good games

Answer 26

Ui= (ei-ci) + alpha(sum of all contributions)

Where alpha is marginal per capita return

Question 27

Possible downside to Keser 1996

Answer 27

The dominant strategy is contribute 7/20 tokens which means there is more space to make errors above this level than below. Having a dominant strategy of 10/20 would be better

Question 28

Critical remarks of Andreoni 1995

Answer 28

-A pure altruist might still want to contribute in Rank treatment since they would rather other people get a higher payoff than them.
-rank treatment might be overall complicated which induces confusion

Question 29

Critical remark of Andreoni 1988

Answer 29

Play was suspended after only 3 rounds following the restart. He claimed “had the budget for subjects been bigger, this would have been unnecessary. Such deceptive practices are, under less restrictive circumstances not recommended”

Question 30

Critical remark of Croson 1996?

Answer 30

In strangers treatment there were groups of 4 matched from a room of 12 so it is very likely strangers will be matched together again

Question 31

What is perfect stranger matching?

Answer 31

Probability of being re-matched with same person is zero. Rules out reputational concerns but lab size restricts number of periods

Question 32

Results of Fischbacher, Gachter & Fehr 2001

Answer 32

Free riders and conditional cooperators dominate but there are also hump shaped contributors.
Contributions stem from heterogeneity of preferences

Question 33

Conclusions of Bayer, Renner & Sausgruber (2013)

Answer 33

-simple learning can’t generate the kind of contribution dynamics commonly attributed to the existence of conditional cooperators.
-cooperative behaviour and its decay observed in PGG is not a pure artefact of confusion and learning