Abstract
We study criteria that compare mechanisms according to a property (e.g., Pareto efficiency or stability) in the presence of multiple equilibria. The multiplicity of equilibria complicates such comparisons when some equilibria satisfy the property while others do not. We axiomatically characterize three criteria. The first criterion is intuitive and based on highly compelling axioms, but is also very incomplete and not very workable. The other two criteria extend the comparisons made by the first and are more workable. Our results reveal the additional robustness axiom characterizing each of these two criteria.
Similar content being viewed by others
Notes
In Gerber and Barberà (2016), the solution concept is “iterated elimination of weakly dominated strategies” and the correspondence is the possibility of agenda manipulation. In Dasgupta and Maskin (2008), the solution concept is “truthful revelation” and the correspondence is a collection of five voting properties.
The relevant primitives for our criteria are the numbers of equilibria that yield an outcome that is (resp. not) selected by the correspondence. In contrast, the relevant primitives for these rules include the “hit rate”, i.e. the fraction of observations predicted by the theory, and the “area”, i.e. the fraction of potential outcomes predicted by the theory. Neither the “hit rate” nor the “area” are relevant primitives for our criteria. We provide here the intuition why the “area” is not a relevant primitive for our criteria. The following two mechanisms have different area but should be considered equivalent by our criteria. The first mechanism has a unique equilibrium that yields an outcome that is selected by the correspondence. The second mechanism has multiple equilibria, all of which yield outcomes that are selected by the correspondence.
In particular, the weak relation is transitive.
This property assumes that all equilibria count the same. This is a natural assumption if one believes that all equilibria are equally likely to occur.
Under our assumptions, the proportion is always well-defined. Indeed, we assume that solution concepts admit at least one equilibrium for each type profile. As a result, the denominateur of the proportion is never zero.
These three axioms are independent. Showing independence of Monotonicity is the most difficult part. We propose the criterion I2, which satisfies all these axioms except Monotonicity. Criterion I2 is based on the following function \(f:[0,1]\rightarrow [0,1]\) defined as \(f(x)=1-x\) for \(x \in \{0,1\}\) and \(f(x)=x\) for all \(x \in (0,1)\). That is, function f is strictly increasing for all \(x \in (0,1)\), but returns the smallest value for \(x=1\) and the greatest for \(x=0\). For any two \(F,F' \in {\mathcal {F}}\), we have \(F' \succeq _{I2} F\) whenever \(f\left( \frac{{F}'_1(y)}{{F}'_0(y)+{F}'_1(y)}\right) \ge f\left( \frac{{F}_1(y)}{{F}_0(y)+{F}_1(y)} \right) \) for all \(y \in Y\), and we have \(F' \succ _{I2} F\) if in addition the inequality is strict for some \(y^*\in Y\).
Consider the following complete order. For any two \(F,F' \in {\mathcal {F}}\), we have \(F' \succeq _{COMP} F\) whenever
$$\begin{aligned} \mathop {\sum }\limits _{y\in Y}\frac{F'_1(y)}{F'_0(y)+F'_1(y)}\ge \mathop {\sum }\limits _{y\in Y}\frac{F_1(y)}{F_0(y)+F_1(y)} \end{aligned}$$Observe that these three axioms do not jointly imply this order. Additional properties would be required, typically imposing some form(s) of anonymity.
These type profiles are irrelevant in the sense that the exact fraction of outcomes in X of each mechanism does not matter.
An assignment has a blocking pair if a student is assigned to a school that another student prefers to her assignment and the other student has higher priority at this school than the first student.
See Appendix 7.2 for the description of both mechanisms.
Note that with \(F^{DA}\) and \(F^{BOS}\) the functions respectively associated to unrestricted DA and BOS by C and X, we also have \(F^{DA}\succ _{PHO}F^{BOS}\). Indeed, in DA all students have a single dominant strategy which consist in ranking all their acceptable schools without switches. There is therefore only one undominated strategy profile in DA, and this profile is always stable. It is then sufficient to show that some of the many undominated strategy profiles in BOS are not stable.
References
Abdulkadiroğlu A, Che Y, Yasuda Y (2011) Resolving conflicting preferences in school choice: the “Boston mechanics” reconsidered. Am Econ Rev 101(1):399–410
Abdulkadiroglu A, Angrist J, Narita Y, Pathak PA (2019) Breaking ties: regression discontinuity design meets market design. Discussion paper, (2170)
Andersson T, Ehlers L, Svensson L-G (2014) Least manipulable envy-free rules in economies with indivisibilities. Math Soc Sci 69:43–49
Arribillaga RP, Massó J (2015) Comparing generalized median voter schemes according to their manipulability. Theoret Econ 11(2):547–586
Bonkoungou S, Nesterov A (2020) Reforms meet fairness concerns in school and college admissions. Working paper
Calsamiglia C, Haeringer G, Klijn F (2010) Constrained school choice: an experimental study. Am Econ Rev 100(4):1860–74
Chen Y, Kesten O (2017) Chinese college admissions and school choice reforms: a theoretical analysis. J Polit Econ 125(1):000–000
Combe J, Tercieux O, Terrier C (2017) The design of teacher assignment: theory and evidence. Working paper
Dasgupta P, Maskin E (2008) On the robustness of majority rule. J Eur Econ Assoc 6(5):949–973
Decerf B, Van der Linden M (2018a) In search of advice for participants in constrained school choice. SSRN Working Paper No. 3100311
Decerf B, Van der Linden M (2018b) Manipulability in constrained school choice. Available at SSRN 2809566
Dogan B, Ehlers L (2020a) Minimally unstable pareto improvements over deferred acceptance. SSRN working paper
Dogan B, Ehlers L (2020b) Robust minimal instability of the top trading cycles mechanism. SSRN working paper
Ergin H, Sönmez T (2006) Games of school choice under the Boston mechanism. J Public Econ 90(1–2):215–237
Fleurbaey M (2012) Social preferences for the evaluation of procedures. Soc Choice Welf 39:599–614
Gerber A, Barberà S (2016) Sequential voting and agenda manipulation. Theoret Econ 12:211–247 (Forthcoming)
Haeringer G, Klijn F (2009) Constrained school choice. J Econ Theory 144(5):1921–1947
Klijn F, Pais J, Vorsatz M (2013) Preference intensities and risk aversion in school choice: a laboratory experiment. Exp Econ 16:1–22
Pathak PA, Sönmez T (2013) School admissions reform in Chicago and England: comparing mechanisms by their vulnerability to manipulation. Am Econ Rev 103(1):80–106
Selten R (1991) Properties of a measure of predictive success. Math Soc Sci 21(2):153–167
Acknowledgements
We are very grateful to Martin Van der Linden for helpful comments and suggestions. We thank John Weymark who commented on a preliminary version of this work. We are grateful to one anonymous referee, one anonymous co-editor and the editor for suggestions that greatly helped improve the paper. We thank all the participants to the DEFIPP workshop and the 13th Meeting of the Society for Social Choice and Welfare for valuable comments and discussions. All remaining mistakes are of course ours.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix on school choice application
Appendix on school choice application
1.1 The school choice model
The model and notation are inspired from Haeringer and Klijn (2009). There are three students \(i_1\), \(i_2\) and \(i_3\) and three schools \(s_1\), \(s_2\) and \(s_3\), each endowed with one seat. Each student can be assigned to at most one school. Students have preferences over the schools they could be assigned to as well as the possibility of remaining unassigned (i.e., being self-matched). Each school has a strict priority ordering over the students. In this setting, a (school choice) problem is a pair \(\pi = (R, \trianglerighteq )\) where
-
1.
\(R {:=}(R_{i_1}, R_{i_2}, R_{i_3})\) is the (strict) preference profile of students over the three schools, and
-
2.
\({\trianglerighteq } {:=}(\trianglerighteq _{s_1}, \trianglerighteq _{s_2}, \trianglerighteq _{s_3})\) is the (strict) priority profile of schools over the three students.
The preference \(R_i\) of student i is a linear order over \(S \cup \{i\}\). If student i strictly prefers school s over school \(s'\), we write \(s~P_i~s'\). As usual, \(s~R_i~s'\) denotes a weak preference, allowing for \(s=s'\). We say that a school s is acceptable for a student i if \(s~P_i~i\) and unacceptable if \(i~P_i~s\). To avoid trivialities, we assume that all students find at least one school acceptable.
The priority \(\trianglerighteq _s\) of school s is a linear order over the three students. If student i has a higher priority than student j at school s, then \(i\vartriangleright _sj\) and we say that i is a competitor of j at school s. School s is a top-priority school for student i if i has no competitor at school s.
We denote by \(\Pi \) the domain of problems satisfying these assumptions.
An assignment is a function \(\mu : \{i_1,i_2,i_3\} \rightarrow \{s_1,s_2,s_3\} \cup \{i_1,i_2,i_3\}\) that matches every student with a school or with herself. We say that student i is assigned in the former case, and unassigned in the latter case.
An assignment is feasible if no two students are assigned to the same school.
Given any problem \(\pi \), an assignment \(\mu \) is stable if it satisfies each of the three following properties.
- Individual rationality::
-
For any student i, we have \(\mu (i) ~R_i~ i\).
- Non-wastefulness::
-
For any student i and any school s, if \(s~P_i~\mu (i)\), then \(\# \{ j \in I~|~\mu (j) = s\} = 1\).
- No justified-envy::
-
For any two students i and j, if \(\mu (j)~P_i~\mu (i)\), then j is a competitor of i at school \(\mu (j)\).
A (school choice) mechanism M is a function that associates every problem \(\pi \) in some domain \(\Pi ^M \subseteq \Pi \) of problems with a feasible assignment. We say that a mechanism is individually rational, non-wasteful or stable, if \(M(\pi )\) is individually rational, non-wasteful or stable for all \(\pi \in \Pi ^M\). As is common, when there is no ambiguity about \(\trianglerighteq \), we often use M(R) to denote the assignment selected by mechanism M.
We assume that the three schools report their priority ordering truthfully to the mechanism. A type profile y is a school choice problem \(\pi = (R, \trianglerighteq )\) (and thus \(Y=\Pi \)), and the players of mechanism M are the three students. For the two mechanisms that we consider, the strategy space \(S_i\) of each student i consists in the set of reported preference \(Q_i\) for which at least one school is unacceptable and at least one school is acceptable.
For any type profile y, the pair (M, y) defines a strategic form game for which students report a preference and the outcome is the assignment selected by M under the profile of reported preferences. Given (M, y), the strategy-space of student i is the set of all the preferences of i that are featured in at least one problem of \(\Pi ^M\). We call these strategies reported preferences. A reported profile is a list \(Q {:=}(Q_{i_1}, Q_{i_2}, Q_{i_3})\) of the reported preferences of all students.
The outcome of the game when students report Q is assignment M(Q). Student i evaluates this assignment according to her true preference \(R_{i}\). In particular, strategy \(Q_i\) is a (weakly) dominant strategy for student i if
In turn, strategy \(Q_i\) is a dominated strategy for student i if
and \(M_i(Q_i',Q_{-i}')~P_i~M_i(Q_i,Q_{-i}')\) for some \(Q_{-i}'\). A strategy is undominated if it is not dominated.
1.2 Two mechanisms
In this section we describe the two school choice mechanisms we compare, which are members of the class considered in Haeringer and Klijn (2009). We first describe \(BOS^2\), a constrained version of the Boston mechanism for which students are allowed to report preferences on two schools only.
- Input ::
-
A (reported) school choice profile.
- Round 1::
-
Students apply to the school they reported as their favorite school. Every school that receives more applications than its capacity starts rejecting the lowest applicant in its priority ranking, up to the point where it meets its capacity. All other applicants are definitively accepted at the schools they applied to, and capacities are adjusted accordingly.
- Round 2 ::
-
Students who are not yet assigned apply to the school they reported as their second favorite school. Every school that receives more new applications in round 2 than its remaining capacity starts rejecting the lowest new applicants in its priority ranking, up to the point where it meets its capacity. All other applicants are definitively accepted at the schools they applied to. The algorithm terminates and all students not yet assigned remain unassigned.
We now turn to \(DA^2\), a constrained version of the Deferred Acceptance mechanism for which students are allowed to report preferences on two schools only.
- Input ::
-
A (reported) school choice profile.
- Round 1::
-
Students apply to the school they reported as their favorite school. Every school that receives more applications than its capacity definitively rejects the lowest applicant in its priority ranking, up to the point where it meets its capacity. All other applicants are temporarily accepted at the schools they applied to (this means they could be rejected at a later point).
- Round 2: :
-
Students who were rejected in round 1 apply to the school they reported as their second favorite school. Every school considers the new applicants of round 2 together with the students it temporarily accepted. If needed, each school definitely rejects the lowest students in its priority ranking, up to the point where it meets its capacity. The algorithm terminates and all students not yet assigned remain unassigned.
1.3 Preliminary results on undominated strategies under \(DA^2\) and \(BOS^2\)
Propositions 1 and 2 require identifying undominated strategies under \(DA^2\) and \(BOS^2\). The following lemmas provide the necessary results for such identification. They are direct implications of characterization results taken from Haeringer and Klijn (2009) and Decerf and Van der Linden (2018a).
Lemma 2
If student i finds only one school acceptable, then reporting only this school is a dominant strategy under both \(BOS^2\) and \(DA^2\).
Proof
This is a straightforward implication of the characterization of dominant strategies in constrained BOS and constrained DA in Decerf and Van der Linden (2018b). □
Lemma 3
Assume that the most-preferred school of student i is a top-priority school for i. Under both \(BOS^2\) and \(DA^2\), (1) i has a dominant strategy and (2) i is assigned to her most-preferred school when she plays her dominant strategy.
Proof
This is a straightforward implication of the characterization of undominated strategies in constrained BOS and dominant strategies in constrained DA in Decerf and Van der Linden (2018a) (Propositions 2 and 4). □
Lemma 4
Assume that the second most-preferred school of student i is a top-priority school for i. Student i has a dominant strategy under \(DA^2\), which consists in reporting these two schools truthfully.
Proof
This is a straightforward implication of the characterization of dominant strategies in constrained DA in Decerf and Van der Linden (2018a) (Proposition 2). □
Let student i’s most-preferred top-priority school be the school that i prefers among the schools that are top-priority for i (if any).
Lemma 5
Assume that the most-preferred school of student i is not a top-priority school for i. Strategy \(Q_i\) is undominated under \(DA^2\) only if \(Q_i\) reports two schools, \(Q_i\) ranks these two schools according to i’s true preference and i weakly prefers these two schools over her most-preferred top-priority school.
Proof
Haeringer and Klijn (2009) (Proposition 4.2) show that a necessary condition for \(Q_i\) to be undominated under \(DA^2\) is that \(Q_i\) reports two schools and \(Q_i\) ranks these two schools according to i’s true preference. Decerf and Van der Linden (2018a) (Proposition 3) show that another necessary condition for \(Q_i\) to be undominated under \(DA^2\) is that i weakly prefers these two schools over her most-preferred top-priority school. □
Lemma 6
Strategy \(Q_i\) is undominated under \(BOS^2\) if and only if (i) the school reported first is i’s most-preferred top-priority school or (ii) the school reported first is not top-priority for i and \(Q_i\) reports two schools, one of which is strictly preferred to i’s most-preferred top-priority school.
Proof
This is a straightforward implication of the characterization of undominated strategies in constrained BOS in Decerf and Van der Linden (2018a) (Proposition 4). □
Rights and permissions
About this article
Cite this article
Decerf, B., Woitrin, F. Criteria to compare mechanisms that partially satisfy a property: an axiomatic study. Soc Choice Welf 58, 835–862 (2022). https://doi.org/10.1007/s00355-021-01376-1
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00355-021-01376-1