Radical Redefinition of Liability for Misrep on Net Social Site - US v Drew
Experimental & Behavioral Economics · decisions Top choice misrep. % of misrep. MD 261 49 18.8%...
Transcript of Experimental & Behavioral Economics · decisions Top choice misrep. % of misrep. MD 261 49 18.8%...
Experimental & Behavioral Economics
Lecture 12: Matching experiments.
Rustamdjan Hakimov, Dorothea Kübler Summer term 2016
Matching • Matching is a subfield of market design
• Markets without money
• In matching problems an agent is paired with an indivisible good (or another agent), and the allocation depends not only on the preferences of the former but also on preferences or a selection criterion form the latter.
• Examples: medical labor market, school choice, college admissions, kidney exchange, lawyers assignments to courts, appointments allocation…
Matching problems
• Matching problems: two-sided and one-sided
• Two-sided problem: marriage market (NRMP, University admissions like in US…)
• One-sided problem: agents have preferences and can be strategic, while objects have priorities and are not strategic (School choice)
School choice • History in US. Everyone has to go to own
district school – Local tax –based financing
– Huge quality differences
– Socio-economic segregation, including racial
• Starting from late 70s: School choice – Parents submit list of their preferences to an
authority
– Based of priorities of the schools allocation of students is done (Sibling, walking zone priorities, otherwise random)
Boston mechanism (BOS)
• All applications are sent to the first choice of students
• If there are more applicants than schools’ capacity, student's of the lowest priority are rejected. Those who are not rejected are ASSIGNED to a corresponding school.
• Quotas of schools are updated.
• Example:
Priorities of schools
A B C
i_1 i_2 i_3 i_2 i_3 i_2
i_3 i_1 i_1
Preferences of students
i_1 i_2 i_3
B A B
A C A C B C
Boston Step 1 A B C i_2 i_1
i_3
Final assignments
A B C i_2 i_3
Boston mechanism (BOS)
• All applications are sent to the first choice of students
• If there are more applicants than schools’ capacity, student's of the lowest priority are rejected. Those who are not rejected are ASSIGNED to a corresponding school.
• Quotas of schools are updated.
• Example:
Priorities of schools
A B C
i_1 i_2 i_3 i_2 i_3 i_2
i_3 i_1 i_1
Preferences of students
i_1 i_2 i_3
B A B
A C A C B C
Boston Step 2 A B C i_1
Final assignments
A B C i_2 i_3
Boston mechanism (BOS)
• All applications are sent to the first choice of students
• If there are more applicants than schools’ capacity, student's of the lowest priority are rejected. Those who are not rejected are ASSIGNED to a corresponding school.
• Quotas of schools are updated.
• Example:
Priorities of schools
A B C
i_1 i_2 i_3 i_2 i_3 i_2
i_3 i_1 i_1
Preferences of students
i_1 i_2 i_3
B A B
A C A C B C
Boston Step 3 A B C i_1
Final assignments
A B C i_2 i_3 i_1
Deferred acceptance mechanism (DA or GS)
• All applications are sent to the first choice of students
• If there are more applicants than schools’ capacity, student's of the lowest priority are rejected. Those who are not rejected are TENTATIVELY assigned to a corresponding school.
• Applications of rejected students are sent and considered TOGETHER with tentatively accepted students.
• Example:
Priorities of schools
A B C
i_1 i_2 i_3 i_2 i_3 i_2
i_3 i_1 i_1
Preferences of students
i_1 i_2 i_3
B A B
A C A C B C
DA Step 1 A B C i_2 i_1
i_3
Tentative assignments
A B C i_2 i_3
Deferred acceptance mechanism (DA or GS)
• All applications are sent to the first choice of students
• If there are more applicants than schools’ capacity, student's of the lowest priority are rejected. Those who are not rejected are TENTATIVELY assigned to a corresponding school.
• Applications of rejected students are sent and considered TOGETHER with tentatively accepted students.
• Example:
Priorities of schools
A B C
i_1 i_2 i_3 i_2 i_3 i_2
i_3 i_1 i_1
Preferences of students
i_1 i_2 i_3
B A B
A C A C B C
DA Step 2 A B C i_2 i_3 i_1
Tentative assignments
A B C i_1 i_3
Deferred acceptance mechanism (DA or GS)
• All applications are sent to the first choice of students
• If there are more applicants than schools’ capacity, student's of the lowest priority are rejected. Those who are not rejected are TENTATIVELY assigned to a corresponding school.
• Applications of rejected students are sent and considered TOGETHER with tentatively accepted students.
• Example:
Priorities of schools
A B C
i_1 i_2 i_3 i_2 i_3 i_2
i_3 i_1 i_1
Preferences of students
i_1 i_2 i_3
B A B
A C A C B C
DA Step 2 A B C i_1 i_3 i_2
Final assignments
A B C i_1 i_3 i_2
Top Trading Cycles mechanism (TTC)
• All students at the top of priority of a school “own” all seats of the school
• Every student point to a student who owns her favorite school
• There is at least one cycle, trades are implemented, quotas are adjusted and new “owners” are assigned.
• Example:
• TTC step 1.
Priorities of schools
A B C
i_1 i_2 i_3 i_2 i_3 i_2
i_3 i_1 i_1
Preferences of students
i_1 i_2 i_3
B A B
A C A C B C
Final assignments
A B C i_2 i_1
(i_1, A) (i_2, B)
(i_3, C)
Top Trading Cycles mechanism (TTC)
• All students at the top of priority of a school “own” all seats of the school
• Every student point to a student who owns her favorite school
• There is at least one cycle, trades are implemented, quotas are adjusted and new “owners” are assigned.
• Example:
• TTC step 2.
Priorities of schools
A B C
i_1 i_2 i_3 i_2 i_3 i_2
i_3 i_1 i_1
Preferences of students
i_1 i_2 i_3
B A B
A C A C B C
Final assignments
A B C i_2 i_1 i_3
(i_3, C)
Summary of predictions of mechanism
BOS DA TTC
Strategy-proof No Yes Yes
Stable (no justified envy) No Yes No
Pareto efficient Yes No Yes
Chen and Sönmez (2006) • Compare DA, TTC and BOS in the setup of school
choice
• One-shot
• Information: only induced preferences and district school. No information about preferences. Random priorities in non-district schools
• Two markets: designed and random, but adjustment for trivial decision
• 36 students, 7 schools, total number of seats equals 36
Results. Chen and Sönmez (2006)
Truth-telling in BOS is lower than in TTC and DA. However, the truth-telling rates in DA and TCC are far from universal
Results. Efficiency. Chen and Sönmez (2006)
Designed GS>TTC>BOS Random BOS>GS>TTC
Information in school choice. Pais and Pinter (2008)
• Compare DA, TTC and BOS in the setup of school choice under different information
• 5 teachers, 3 schools with 5 positions
• 4 informational treatments: – Zero information (only own preferences)
– Low information (Chen and Sönmez, 2006)
– Partial information (priorities of other students)
– Full information (full preference and priorities profiles)
Results. Pais and Pinter (2008)
How do we help people to report truthfully? Guillen and Hing (2013)
• Advice in TTC
• Individual decision making setup –playing against computer
• 4 schools, one seat each, 4 students
• Third party (newspaper or forum) advice by 4 treatments: – Baseline (no advice)
– Right advice: The mechanism is designed so that truthful reporting maximizes your chances of getting favored schools.
– Wrong advice: Since the top schools will have many applicants you should be realistic and apply to schools where you are likely to gain acceptance.
– Both
Results. Guillen and Hing (2013)
Is there demand effect?
Field experiment. Guillen and Hakimov (2016)
• Topic allocation problem (School choice problem) (smartphone, TV set and Scanner)
• Top Trading Cycles mechanism: strategy-proof
• Treatments:
1. Detailed mechanism description (MD)
2. Property description and advice (PD)
3. Both (MPD)
Results. Guillen and Hakimov (2016)
Treatment Sample
size
Top choice
misrep. % of misrep.
Non-
trivial
decisions
Top
choice
misrep.
% of misrep.
MD 261 49 18.8% 167 47 28.1%
PD 106 6 5.7% 63 6 9.5%
MPD 113 10 8.8% 74 9 12.1%
Total 480 65 13.5% 304 62 20.3%
Misrepresentation rates by treatments
Strong positive effect of advice. In line with Braun et al. (2014)
Results. Guillen and Hakimov (2016)
Tentative topic
N
Number of misrepresentation
s of the top choice
Number of students
affected by TTB
Proportion of truth
MD TV set 93 31 30 66.67%
PD TV set 40 3 3 92.50%
MPD TV set 40 9 8 77.50%
How do we help people to report truthfully? Ding and Schotter (2016a)
• 4 types of subjects (preferences)
• 3 types of objects
• Type 1 priority in A
• Type 2 priority in A
• Type 3 priority in B
• BOS16, BOS10, GS16
• Some subject can chat for 5 minutes between Phase 1 and Phase 2.
How do we help people to report truthfully? Ding and Schotter (2016a)
• Strategy changes (Effect of chatting)
Intergenerational advice and learning in DA. Ding and Schotter (2016b)
• 4 types of subjects (preferences)
• 3 types of objects
• Type 1 priority in A
• Type 2 priority in A
• Type 3 priority in B
• BOS, GS
• Repeated play versus intergenerational advice
Intergenerational advice and learning in DA