L19 - Experimental Economics: Introduction Flashcards
How can we further test game theory?
- (Traditionally) game theory is usually based on mathematical introspection rather than careful empirical observation.
- Traditional econometric data studies are not very effective in this context either.
- Why? Game theory is extremely sensitive to assumptions concerning individuals’ information, values, choices, order of moves etc, and so any standard empirical study will find it hard to control for all these necessary factors.
- Instead, laboratory experiments offer a possibility of testing game theory.
- They started in the 1960’s but only became popular in the 1980’s and 1990’s. Vernon Smith won a Nobel for experiments in 2002
What happens in economic experiments?
- Related to experiments in psychology, experiments in economics monitor how subjects make decisions or play games (using computer software) in a laboratory context or online.
- Unlike psychology though, the subjects are monetarily rewarded depending upon their behaviour. The data is then analysed with techniques similar to standard econometrics
- Beyond simply testing a theory, lab experiments are best used to test between competing theories, to find the specific conditions under which theories fail (stress-testing), to estimate behavioural parameters (like 𝛽, see later lecture) and to find new empirical regularities
Advantages of laboratory experiments?
- The key advantage is the ability to carefully control for relevant factors.
- An experimenter can control subjects’ information, choice sets and possible payoffs, in a way that would be extremely difficult by other methods.
- This allows experiments to provide a careful test of theories. Experiments can be carefully designed to match the assumptions of the theory. They can also control for outside factors to help provide a clear interpretation of the results.
- In particular, this gives high internal validity – the ability to make clean, causal conclusions from the results.
- Laboratory experiments also allow results to be replicated more easily
Disadvantages of laboratory experiments?
- Despite their benefits, laboratory experiments can be criticised.
- These criticisms often focus on external validity – the ability to transfer the conclusions to a wider setting outside the lab.
- Specific problems may occur from the fact that, unlike the real world,
- i) the lab situations are often abstract and simplified
- ii) the subjects are often students
- iii) there are often low incentives to behave optimally (not paid a lot)
- iv) there is often little opportunity for learning
- v) there is no market selection –> suboptimal behaviour isn’t filter out
- For v): If agents behave sub-optimally in the market, they will exit or not survive, making optimal behaviour more likely to be observed in real markets
What are other types of economic experiments?
- In response to these criticisms, some other forms of experiments with higher external validity are becoming very popular.
- Within such experiments, researchers will often utilise explicit randomisation of different treatments onto different groups to carefully measure the different effects.
- Natural Experiments: arise when some form of controlled setting occurs naturally in a way that allows a researcher to carefully test a hypothesis. E.g. a random, exogenous or policy variation. Min wage variation –> America tested the affect of different minimum wages across different states
- Field Experiments: arise when an experiment is designed such that subjects are undertaking tasks within their normal real-world environment, often without even knowing about the experiment. E.g. random variation in advertising.
How is a Ultimatum game played?
- Player 1 and Player 2 have to divide X between themselves.
- P1 moves first and proposes that P2 gets a share, m. P2 observes m and then either rejects or accepts.
- If P2 accepts, then the players receive the proposed shares. Player 1 receives (X-m) and Player 2 receives m. If P2 rejects, then both players get zero.
During lab experiments what was the actual empirical outcome of the Ultimatum Game?
What are the possible explanations for the results of the Ultimatum Game experiments?
These results could be explained in three ways:
i) Poor experimental design (low internal validity)
ii) Subject risk-aversion or fear of rejection - might explain high offers, but not receivers’ rejection rates.
iii) The results are genuine. Contrary to self-interest, individuals may have a concern for others’ payoffs or some sense of ‘fairness’.
Below, I hope to persuade you that i) poor experimental design cannot be a valid explanation and that iii) must be part of the best explanation.
Why isn’t risk aversion a reason for the results of the Ultimatum Game experiment?
- First, note that one can distinguish between the explanations of ii) risk aversion and iii) fairness.
- Consider a variation of the game, called the Dictator Game. Here P1 simply decides on a division and the players receive this split. P2 does not have an action. No fear of rejection for P1.
- Forsythe et al (1994) show that average offers in the Dictator game are around 20%. This is lower than in the Ultimatum game, but it is not 0% as one would predict under self-interest
- Hence, explanation ii) risk aversion cannot explain all the results. Subjects may indeed care about others’ payoffs or some form of ‘fairness’
Why is Poor Experimental Design not an explanation for the results of the Ultimatum Game experiment?
Criticism 1
Criticism 1: Subjects had no time to learn as they would in a real situation.
The predictions only refer to a one-shot game, so allowing the subjects to play repeatedly to allow learning is difficult.
However, researchers allow subjects to play the game repeatedly with different opponents each time in an attempt to preserve the one-shot nature while encouraging subject learning. The offers and rejections only decrease slightly
Why is Poor Experimental Design not an explanation for the results of the Ultimatum Game experiment?
Criticism 2
Criticism 2: Subjects did not have sufficient incentives to play the predicted strategies.
Normally the funds for experiments are indeed limited.
However, by conducting experiments in developing countries, researchers can offer stakes with much larger purchasing power.
E.g. Cameron (2000) allows X to be equal to one month’s average wage in Indonesia. The results show little change in either in rejections or demands.
Why is Poor Experimental Design not an explanation for the results of the Ultimatum Game experiment?
Criticism 3
- Criticism 3: The subjects were not perfectly anonymous and so they deviated from the predictions because of other dynamic considerations.
- This is a very important issue
- Why might non-anonymity affect player’s behaviour?
- potential social stigma from being selfish
- able to develop a reputation in lab for rejecting low offers to induce higher offers later
- Hoffman et al (1994) used a ‘double blind setting’ for the Dictator game where offers are given in envelopes containing $1 bills or similar pieces of blank paper entered anonymously into box before being handed out to respondents. Not even the experimenter could know what a subject offered. The results show much lower offers, 10% average, but they are still positive.
How do gender, race, culture and other subject characteristics influence fairness?
- Females reject less often. Subjects offer more to men and are more likely to reject women’s offers. Black subjects offer more generously but also reject more often. •
- Photos of subjects rated in attractiveness. Top and bottom 10% used with second group. Men no difference. Women offer 5% more to attractive men.
- Murnighan and Saxon (1998) tested children ranging from kindergarten to 3rd and 6th grade. Used either sweets or money (though the children were not paid). Less rejections than adults. Average offers increased in age. Young children behave very close to theory – do we learn to be altruistic/fair?
- as children get older they are offer more (learn social norms of fairness)
- Subjects who major in economics offer less and reject more often. No difference in first or senior years suggests behaviour is not learned at uni.
How did Culture influence the Ultimatum Games Results?
- Henrich et al (2001, 2002) used subjects from primitive, remote communities in areas such as Africa and S. America.
- In contrast to most standard results, many behave near the theoretical prediction.
- For example, with Machiguenga farmers in Peru, the average offer was 26% and all but one offer was accepted.
- This community are socially disconnected. Low sense of altruism? Show low levels of cooperative activity (e.g. hunting).
- Roth et al (1991) also found significant cultural differences…
What was the aim of the classic experimental paper Roth et al . (1991)?
This paper aims to:
i) provide another test of the predictions for the Ultimatum (U) game,
and to learn more about the U game results by:
ii) comparing these against the results of a related Market (M) game
iii) comparing the results across four different cultures USA, Japan, Slovenia and Israel.
How did the experiments work in the Roth et. al (1991)?
- The subjects were randomly assigned to the U game or the M game.
-
U game:
- subjects randomly assigned a P1 or a P2 role. P1 had to divide $10, using increments of $0.05. P2 accepted or rejected.
- SPNE predicts offers of 0 or $0.05. All positive offers should be accepted.
-
M game:
- subjects randomly assigned into one seller and 9 buyers. The buyers simultaneously submit a single bid for a ‘good’ of $10 in increments of $0.05. The buyer with the highest bid, b, earns a payoff of (10 – b), the rest gain nothing. The seller accepts or rejects b.
- SPNE predicts a winning bid of $9.95-10.00. The bidders compete by increasing their bids until the point where is no further marginal benefit. The seller should always accept a positive winning bid
- Note the similarity in game equilibria – unequal payoffs amongst subjects
