Experimental games

advertisement
Experimental economics
Mylene Lagarde, LSHTM
Johannesburg, Nov 2007
HEPNet Workshop
Overview
1.
Background and characteristics of experimental
economics
Applications
2.
•
•
3.
Classic use: testing theories
Recent developments: social preferences
Designing an experiment
•
•
•
An example
Main features of an experiment
Some classic games
4.
Relevance for health systems research
5.
Possibilities and limitations of experimental
economics
1. Background and characteristics
of experimental economics
Origins : the willingness to test
economic assumptions
• Micro-economics relies heavily on a
number of assumptions
• Preferences/mechanisms are hard to observe
in natural environments
• Need to test and therefore potentially refute
these assumptions in a scientific way?
– Market classroom experiment : Chamberlain (1948) and his
student V. Smith (1962)
The influence of game theory
• Game theory attempts to model strategic
interactions
– How do agents choose strategies which will maximize
their return, given the strategies the other agents choose
• They use simplified structures for modeling
strategic behavior
– Games provide a formal modelling approach to social
situations in which decision makers interact with other
agents
• Application to many problems
– Political science, economics, etc.
Other major influences
• Psychology
– Psychologists have used experiments with participants
to test some assumptions on fundamental behaviours:
authority
• Behavioural economics
– Informing economic theory with lessons from
psychology
– Challenges neo-classic assumptions of economics
(rationality, self-interest, stability of preferences, risk
aversion, etc.)
Principles of experimental economics
• To create an (economic) situation in a neutral
environment (lab) which is:
– real
• Real people…
• … participating for real monetary stakes
• …following real rules
– simple
– controlled
• Motivations
• Initial information
• Interaction processes
– reproducible
• And observe behaviours and outcomes
A method of empirical investigation
1. Testing theories and establishing empirical
regularities as a basis for new theories
2. Elicitation of preferences: attitude towards
risk, fairness, altruism, trust, time
preferences
3. Testing policies: wind tunnel experiments
A young and blossoming field
The 2002 Nobel Prize in Economic Sciences
Daniel Kahneman
"for having integrated insights
from psychological research
into economic science,
especially concerning human
judgment and decision-making
under uncertainty"
Vernon L. Smith
"for having established
laboratory experiments as a
tool in empirical economic
analysis, especially in the
study of alternative market
mechanisms"
2. Some applications
•Classic use: testing theories
•Recent development:
• Field experiments
• Revealing social preferences
Main areas of application
• Financial economics
– Information diffusion and aggregation, price setting mechanisms,
irrationality, bubble formation
• Industrial economics
– Design of incentives and contracts (improving performance), auction
designs (electricity markets, air slot allocation, etc.)
• Environmental economics
– Regulations to use common resources (fishery), to design pollution
permit markets, etc. (public goods, externalities)
• Agricultural economics
– Real market experiments: food auction, Actual WTP for cereals, etc.
Main fields of research
• Individual choices
– inter-temporal choices, risk aversion, social preferences,
etc.
• Strategic Interactions (game theoretic hypotheses)
– contracts, conflicts/coordination, public goods, etc.
• Industrial organization
– Market efficiency
– Market power issues (development of collusion,
cooperation)
– Market design
Testing theories
• Neoclassical theory works (relatively) well for
markets where important assumptions hold (prop
rights well defined, perfect information, low
transactions costs)
• BUT not always….
• EE are simple tools which allow to
isolate/measure the reasons for a model’s failure
• Models and principles that survive the testing can
be used to address field questions
Testing regulations/policies
• Financial markets
– Testing how to regulate markets to prevent
irrational behaviour, the formation of bubbles,
etc.
• Auction design
– Testing one predictions of models with real
individuals in a controlled environment
• Environment economics
– How to regulate the use of rare resources
– Regulations and externalities (pollution
permits, etc.)
The spectrum of experiments
• Historically, EE has been applied in lab
experiments
– Participants = graduate students
• Computerized game or paper-based
• Neutral frame
– To test theories, policies
• Move lab games into the field: a new practice
– Paper based
– Participants are not students, but ‘real’ people
– Experiments are tailored to a field context/policy, and
less neutral
– Experiments are used as tools to measure some values
The importance of social preferences
• “Social preferences”
– “preferences”= choices made by individuals , tradeoffs btw different
collections of things they value (time, money, prestige, leisure, etc.)
– “social preferences”= how people rank different allocations of material
benefits to themselves and others
• Existence and importance of social preferences
– Social preferences have an impact on the enforcement of social norms:
abundant literature on the effects of fairness, inequality (absence of), trust
on economic behaviour, etc.
– Evidence of adverse effects of (monetary) incentives if social preferences
not accounted for (voluntary work, intrinsic motivation)
Revealing social preferences
• Difficult to identify them, even more to measure them
– How do you observe altruism ? Fairness ? Inequality aversion
(relative to each other)? Risk aversion ?
• Why use experiments to measure them?
– Surveys on trust, altruism may be biased
– Experiments are likely to suffer less from response bias to reveal
fundamental preferences because of monetary stakes
– Experiments allow quantification
• What types of preferences can experiments measure?
– Social preferences: norms of altruism, fairness, trust,
trustworthiness, aversion for inequity.
– Other preferences: risk aversion, time preferences.
Social preferences in 15 small scale
societies (1)
• Henrich et al. (AER 2005, Science 2006)
• Setting: project led by anthropologists, in
15 remote societies on 4 continents
Social preferences in 15 small scale
societies (2)
• Objectives: to investigate how social preferences vary
across cultural environments
• Methods
– Measures of altruism, fairness and aversion for inequality
– Measures of market integration and other aspects of socio-economic
life in these societies
• Results
– preferences are not exogenous (as assumed by neo-classic
economics) but shaped by the economic and social interactions of
everyday life
– Substantial variation among populations of aversion for inequality
which co-evolves with norms of fairness/altruism
– the canonical model of the self-interested material payoffmaximizing actor is systematically violated
Using EE to measure preferences (1)
• Karlan (AER 2005)
• Setting: Peru, clients of a micro-credit program
• Objectives: investigating the determinants of default payment
and saving
• Methods
– uses survey and experimental measures of social capital -trustworthiness
and willingness to cooperate;
– Observes borrowing behaviour over 1.5 year
• Findings
– Significant negative correlation between (magnitude of) trustworthiness
and (1) probability of default , (2) dropout of dropping out of the program
and (3) volume of savings
– Some individuals are just not trustworthy and will not repay their loans?
– TG can be a valid measurement tool for trustworthiness, less obvious for
‘trust’ which is too context specific
Using EE to measure preferences (2)
• Carpenter and Seki (2006) and Carpenter and Seki (2004)
• Setting: Japan, individuals involved in different degrees in
fishing activities and are exposed to different degrees of
competition
• Objectives:
– to investigate the level of competition in a work environment can
explain the propensity to cooperate
– to investigate whether the individual propensity to cooperate affects
productivity
• Methods
– uses experimental measures of cooperation
– Uses surveys to measure productivity and level of competition
• Findings
– The higher the degree of competition the less cooperative
individuals are
– The more cooperative individuals are, the higher their productivity
A new but growing field
• Many other examples of games used to measure social
preferences
– To see whether they help explain actual behaviours
– To use them as outcomes (e.g. impact of education program
on altruism amongst children)
• Currently implemented in Mexico: games piggybacking on a national survey
– “to provide new insights into the role of preference
heterogeneity in explaining economic decisions and enable
new tests of hypotheses that seek to explain family
behaviour”
3. Designing an experiment
• Let’s play a game !
• Main characteristics of the game
•Examples of other games
Basic rules
• Please try to follow all the rules:
– Listen and read carefully the instructions
– If you have questions, raise your hand and I will come and
answer it privately
– Do not talk with each other, even when the experimenters
are calculating the payoffs between rounds
• 100 ECU = US$ 1
Example with 3 players
• Players 1, 2 and 3 are given 200 ECU each
• Their decisions:
Player 1 invests 0 in A
Player 2 invests 100 in A
Player 3 invests 200 in A
Total invested in
A = 300
• Return on investment from A = 300 x 1.5=450
• Return for all three players= 450/3= 150
• Endowments at the end of round 1:
Player 1 has 200 +150 = 350
Player 2 has 200 -100 +150 = 250
Player 1 has 200 -200 +150 = 150
Steps
1. Write your ID number on your response sheet
2. Write the amount you want to invest in A
3. Detach the bottom section of your answer sheet,
fold it and place it in the box
4. Wait until you know the value of individual
payoffs
5. Calculate yourself the value of your endowment
at the end of the round
6. Play the next round
Public good game: interpretation
and results
• Interpretation:
– Propensity to cooperate (vs. free-ride) in a group, reciprocal
behaviours in groups
– Use of different types of regulations (external, communication) in
different contexts: team compensation, cooperative
production/maintenance, use of common resources
• Results:
– Prediction: each player will contribute nothing (individual welfare
vs. collective welfare with public goods bringing externalities)
– One-shot game: 50% contribution ; contributions diminish over
time (final period: contribution=0)
– What contributes to increase cooperation
• Communication
• Possibility of individual punishment
• Disclosure of free-riding
What would be different “for real”
• Smaller group
• Participants don’t make their own
calculations !
• You would be playing for real money
• Several “treatments” could be tested
Game procedures (1)
• Abstract setting
– Anonymous players, playing once, for money, without
communication
– Not lifelike, but baseline to observe effects of any
variant (communication, non-anonymity, etc.)
• Participants:
– college students recruited with minimal knowledge of
the task (e.g. Study on decision making)
– More “real” people that you may want to observe
• Organization:
– Ensure anonymity: private box, random drawing
– Ensure credibility: existence of recipients if not
present, real incentives paid at the end; DON’T LIE
Game procedures (2)
• Instructions to describe rules:
–
–
–
–
Plain abstract language (benchmark)
Use letters/neutral framing
Avoid some words “sharing”, “helping”, “trust”
Examples, quiz
• Design characteristics
– Multiple rounds vs. One-shot games
• Depends on the game
• If “equilibrium” situation, then multiple rounds to
allow learning (then rematching)
– Treatments vs. individual measurements
Game procedures (3)
• Giving interactions to participants in multiplayer games:
– Direct feedback
– “Strategy method”: players make choices conditional
on every single response of other player
• Treatments
– Several groups: 1 difference per group (rules, framing,
etc.)
– Allows to measure the effect
– Not necessary if objective= measurement of
preferences for all individuals
Components of an experiment
• Institution (rules of the game)
– Feasible actions
– Sequence of actions
– Information given
– Knowledge of participants
• Framing of instructions
– Abstract framing : basic incentives only
– Framing matching reality to get at social norms:
• Participation to maintenance of a water well
• Health worker, community, patient
– Interesting to compare abstract vs. real framing
Experimental variables to define
• Procedural design
–
–
–
–
Instructions
Illustrative examples and tests of understanding
Criteria for answering subjects’ questions
Trial periods
• Experiment characteristics
– Payment
– Recruiting subject pool, number of subjects
– Matching procedures when interaction
• Environment
– Date and place
– Computerized or paper-based
– Irregularities which occurred
Analysis of experiments
• Most analyses do not require sophisticated
methods
• Measures of social preferences
– Creation of proxies based on individual choices
• Descriptive analysis at group level
underlying
– Outcome variables under each treatment
– Influence of simple socio-economic variables
A few examples of other games
Dictator Game
X $$
PLAYER 1
% of X
= measure of
altruism
PLAYER 2
DG - interpretation and results
• Interpretation
– Measure of altruism; aversion for inequality
(knowledge that the other is poor), etc.
– Relates to charitable giving
• Results
– Prediction: no sharing BUT on average individuals give
20%
– Allocations increase with needy/deserving recipients
and decrease with anonymity
– Norms vary across settings/countries
– Game sensitive to setting, etc.
Ultimatum Game
X $$
PLAYER 1
% of X
PLAYER 2
AGREES:
Deal is made
REFUSES:
= measure of
preference for
equity/fairness
Both get
NOTHING
UG – interpretation and results
• Interpretation
– Measures of social norms of sanction
– Relates to pricing in a monopoly situation, or last minute settlement
• Results:
– Prediction: proposer sends the smallest possible unit (always
accepted by recipient because better than nothing)
– On average proposers offer 30-50%
– Offers below 20% are rejected half of the time
– Norms vary across settings and are well accepted (offers will be
readily made according to norms)
– Effects of competition:
• Btw proposers increases offers
• Btw recipients diminishes threshold of acceptance
– No effect of anonymity
Third Party Punishment Game
= altruism under
social norms
pressure
X $$
PLAYER 1
-Y$
% of X
SANCTION
WATCHES
-k$
= measure of
preference for
equity/fairness
PLAYER 3
X$
PLAYER 2
TPP – interpretation and results
• Interpretation
– Behaviour of punisher: Reveals aversion for
inequality (against a shared norm): people who
dislike it will take costly actions to reduce
inequality
• Results
– Prediction: No allocation from dictator ;
observer never punishes dictator
– The less dictator allocates, the higher is the
sanction
Trust game
= measure of trust
X $$
PLAYER 1
% of X
2 x (% of X)
PLAYER 2
Part of
2 x (% of X)
= measure of
trustworthiness
Trust – interpretation and results
• Interpretation
– Trust of investor and reciprocity of trustees
(trustworthiness)
– Exchange of services without possibility to bind the
second movers
• Results
– Prediction: (expected) repayment from trustee: 0 and
Entrusted/invested amount: 0
– On average investors give 50%, while trustees repay
slightly less than that
– The bigger the investment, the bigger the repayment
Bribery game
- $$
$$
Private
citizens
Expected cost of - Y $
being punished/
caught
Other
members
of society
$$$
$$
$B
- $$
Public
servants
$$
Expected cost of
being punished/
caught + cost of
producing the
service
Results from bribery games
• Very few examples, different versions/level
of complexity
• Barr and Serra:
– Experiments on students coming from different
countries
– positive correlation between internalized norms
(Transparency international index) and
corruption in the game
– Contradicts 1 other experiment
4. Possible applications to health
system issues
• Investigating corruption
• Revealing the influence of social preferences
• Health insurance
• Designing incentives, contracts, etc.
Trust and trustworthiness
• Studies have emphasized the importance of trust
– In the workplace
– For patients towards health workers
• Experimental games could:
– Provide a way to measure trust and trustworthiness in
different framed contexts
– See how trusting behaviour differ when played with
anonymous players/ nurses with doctors / doctors with
themselves, etc.
Experiments as measurement tools
• Quantification of social preferences
– altruism, aversion for inequity, trust, time preference,
etc.
• Use the variables to test their association with:
– health seeking behaviour (time preferences)
– Perceived quality of care (trust)
– Providers’ behaviours (different cooperative behaviours)
• Objective:
– Descriptive
– Prescriptive: link with other theories ; test new policies,
etc.
Health insurance
• When people are free to choose whether and how
they secure their wellbeing, do they make
predictable and rational decisions that are
consistent with the objective of reducing
vulnerability to health risks?
• Are there differences across individuals that can
explain decisions to enroll/not enroll?
– Impact of time preferences on decision to enroll in a
CBI ?
– Impact of risk aversion on propensity to enroll
Investigating corruption
• Barr et al. (2004): the public servant’s dilemma
• Setting: Ethiopia, nursing students
• Objectives:
– to investigate the determinants of corrupt behaviour
• Methods
– uses a designed experiment with 3 types of players (monitor, health
workers and community members)
• Findings
• Embezzlement of resources is affected by
– Earnings HWs (small effect)
– Observability of embezzlement behaviour and likelihood to be
punished
– Legitimate monitoring (Election of a member of the community vs.
random designation)
– Framing increases variance in behaviours suggesting different
professional norms
Plenty of other possible games !
• Games designed to reproduce an existing
environment
– More complex, specifically designed/tailored
– Modelling choices faced by actors in a particular situation
– Reproduce incentives and main characteristics of these
choices
– See which regulations may work better
• Objectives
– To better understand the determinants of behaviours
5. Possibilities and limitations of
experimental economics
• What experiments can / can’t do
• Advantages
• Limitations
Experiments are one of many
research tools
• Qualitative methods
• Surveys: Econometric/statistical analysis of
existing data
• Experimental methods
• Real-world pilots/trials
• Each has its relative strengths/weaknesses
– Complements, not substitutes
When are experiments useful?
• To reveal or measure social preferences (altruism,
trust, etc.)
• To understand behaviours/phenomena that can
hardly be observed in the reality
– Measure (relative) values
• If the data don’t exist, then create the data you
need
– If it’s never been done, how do you know what works?
– Test new policies/ incentives / types of regulatory
mechanisms
– Preliminary step, before actual pilot study in the field
Advantages of Exp Econ (1):
enhanced control
• It is known which variables are exogenous and
which are endogenous
• Many variables that cannot be directly observed in
the field can be observed in the lab.
– Quantification of reaction to incentives, cut-offs (risk
aversion)
• Subjects are randomly assigned to the treatment
conditions – rules out selection bias.
Advantages of Exp Econ (2):
avoiding some biases
• Motivation (payment)
– avoid potential biases of responses obtained through
classic survey instruments which can reflect a
“purchase of moral satisfaction”
• Language:
– neutral phrasing, avoid framing effects that influence
answers
• “Double blind”
– Some designs allow complete anonymity and avoids
experimenter effects and other-regarding concerns
Example: incentive program (P4P)
• Experiments could help
understand:
– Relative efficiency of different
types of rewards/sanctions on
an experimental outcome
– Design of incentive program:
• individual or group-based
and their impact on freeriding/gaming
• Impact of possibility to
observe efforts or not
• Experiments could not
answer:
– Magnitude of effect on
performance/quality of work, etc.
– Which value of incentive for a
particular task (WTP, DCE)
Example: health insurance
• Experiments could help
understand:
– Time preferences of different
groups of population
– Risk aversion of different
types of populations regarding
low and high probability events
• Experiments could not
answer:
– Which service package would
– Which value of premiums could
be acceptable (WTP, DCE with
service package as well)
– Which type of management
(community based, facility-based,
etc.)
Limitations of Exp Econ
• Internal validity: do the data permit causal
inferences?
– a research study has internal validity if the outcome is a function of
the variables that are measured/ controlled/ manipulated in the
study
– Therefore internal validity of experiments is a question of proper
experimental controls and correct data analysis.
• The method is not realistic and lack external
validity
– A research study has external validity if the results obtained would
apply to other similar programs or approaches.
– Problem of realism of “abstract” setting / simplistic modelling
– Problem of representativity/realism: are experimental subjects
representative for out of sample applications?
Responses to lack of realism
• Most economic models are unrealistic in the sense
that they leave out many aspects of reality to
simplify it...
• Simplicity as a virtue?
– because it enhances our understanding of the interaction
of relevant variables.
– Because it allows to get at underlying values/ reflexes
• Whether realism is important depends on the
purpose of the experiment.
– If testing a theory or understanding the failure of a
theory, then the evidence is important for theory
building but not for a direct understanding of reality.
– If EE is used to better understand real behaviours, then
efforts must be made to frame the experiments and
make them as close to real as possible
Ethical issues
• Transparency of motives
– Even if not being deceptive, usually not “honest” about
motives
– When deception is involved, potentially more serious.
• Inducing decisions
– The problem of the size of stakes / bribing
• Emotional distress
– subjects may learn something about themselves not like
(selfishness, unacknowledged racist preferences)
– Even more problematic when not anonymous (to study
effects of social norms)
Download