Skip to main content
Log in

The “Boston” school-choice mechanism: an axiomatic approach

  • Research Article
  • Published:
Economic Theory Aims and scope Submit manuscript

Abstract

The Boston mechanism is a popular student-placement mechanism in school-choice programs around the world. We provide two characterizations of the Boston mechanism. We introduce two new axioms; favoring higher ranks and rank-respecting invariance. A mechanism is the Boston mechanism for some priority if and only if it favors higher ranks and satisfies consistency, resource monotonicity, and rank-respecting invariance. In environments where each type of object has exactly one unit, as in house allocation, a characterization is given by favoring higher ranks, individual rationality, population monotonicity, and rank-respecting invariance.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Similar content being viewed by others

Notes

  1. Related experiments are conducted by Pais and Pinter (2007) and Calsamiglia et al. (2010).

  2. Also see Abebe (2009).

  3. Roth (1991) explains how priority mechanisms in various regions of the U.K. entry-level labor market for new physicians failed in matching as unraveling prevailed.

  4. Seattle Public Schools, New Student Assignment Plan Transition Plan for 2010-11, January 2010 draft, retrieved from http://www.seattleschools.org/area/board/09-10agendas/010610agenda/nsaptransitionattachment.pdf on 01/27/2010.

  5. Pathak and Sönmez (2013) suggest that the school district’s desire to decrease student mobility may have played a role for this switch, as the Boston mechanism can force students to use conservative strategies such as listing their neighborhood schools first.

  6. Based on their findings, Abdulkadiroğlu et al. (2008) propose a mechanism that could be regarded as a hybrid of the deferred acceptance and the Boston mechanism.

  7. The models of Abdulkadiroğlu et al. (2008) and Miralles (2008) share several features. Most notably, the ordinal preferences of students over schools are homogeneous, and there are no intrinsic priorities in both models.

  8. Ehlers (2008) shows that Boston mechanism is difficult to manipulate when there is limited symmetric information.

  9. Chen and Kesten (2012) observe that there is a class of hybrid mechanisms between the Boston mechanism and the deferred acceptance mechanism that are being successfully used in China for college admissions. They formalize this class of hybrid mechanisms and inspect their properties.

  10. Ehlers and Klaus (2003, 2006) characterize the class of deferred acceptance mechanisms with acyclic priority structures Ergin (2002).

  11. See also an earlier contribution by Balinski and Sönmez (1999).

  12. Kesten (2009) inspects the reasons why the random priority mechanism lacks efficiency and shows that a modification to the random priority method would make it equivalent to the probabilistic serial mechanism.

  13. Manea (2009) provides sufficient conditions for asymptotic efficiency and inefficiency of the random priority mechanism.

  14. A matching \(\mu \) induces justified envy for a student \(i\) if there exists a school \(a\) such that \(a P_i \mu _i\) and \(i \succ _a j\) for some \(j \in \mu _a\).

  15. Note that no information about the utility functions of individuals is provided in the description of the problem: Rather, only ordinal preferences are primitive to the problem.

  16. The experiments of Chen and Sönmez (2006) were conducted in two environments, called designed and random environments. The proportions of subjects reporting their true preferences in the Boston and the deferred acceptance mechanisms were 14 and 72 % in the designed environment, and 29 and 56 % in the random environment, respectively.

  17. It is straightforward to see that favoring higher ranks implies that there are no \(c\) and \(i \in I_c\) such that \(i\) prefers \(c\) to her assignment and \(c\) has a vacant position.

  18. A mechanism \(\varphi \) is group strategy-proof if, for any preference profile \(P\), for any quota vector \(q\), for any group of agents \(J \subseteq I\), there exists no joint manipulation of these agents \(P^\prime _{J}\), such that for all \(j \in J, \varphi _j[P^\prime _J,P_{-J};q] R_j \varphi _j[P;q]\) and there exists some \(i \in J\) such that \(\varphi _i[P^\prime _J,P_{-J};q] P_i \varphi _i[P;q]\).

  19. As explained in Sect. 5.2, favoring higher ranks does not imply individual rationality in a setting in which the quota vector is fixed.

  20. This definition implies that for all \(P_i^{\emptyset }, \tilde{P}_i^{\emptyset } \in \mathcal P^\emptyset , \varphi \left[ P_i^{\emptyset },P_{-i};q\right] =\varphi \left[ \tilde{P}_i^{\emptyset },P_{-i};q\right] \).

  21. It follows that, for all \(P_i^{\emptyset }, \tilde{P}_i^{\emptyset } \in \mathcal P^\emptyset , \varphi \big [ P_{i}^{\emptyset } ,P_{-i};q_{\varphi _{i}\left[ P;q\right] }-1,q_{-\varphi _{i}\left[ P;q\right] }\big ]=\varphi \big [ \tilde{P}_{i}^{\emptyset } ,P_{-i};q_{\varphi _{i}\left[ P;q\right] }-1,q_{-\varphi _{i}\left[ P;q\right] }\big ]\).

  22. We stipulate \(\infty -1=\infty \), so consistency implies that when an unmatched student is removed from the problem, assignments for all remaining students are unchanged.

  23. Although we suspect that this lemma is well known, we were unable to find a reference.

  24. To see this point, note first that it is clear that a Boston mechanism in which every student is acceptable satisfies all the axioms in this alternative environment. To show that the set of axioms implies that the mechanism is a Boston mechanism, observe that the assumption that all students be qualified places a further restriction, but all the arguments that go through in our original setting still go through, so the conclusion follows.

  25. To see why both \(P_{j}\) and \(P_{j}^{\prime }\) top-rank \(c\), first observe that \(P_{j}^{\prime }\) top-ranks \(c\) because otherwise \(j\) receives \(\emptyset \) by individual rationality, which follows from favoring higher ranks. This implies that \(j \in I''\) and, since \(I'' = I'\) by the assumption maintained throughout this proof step, it follows that \(j\) top-ranks \(c\) at \(P_j\) as well.

  26. Note, however, that we should account for the difference in the settings by setting \(q_{c^{\prime }}=1\) for every \(c^{\prime }\in C\) here, while \(q_{c^{\prime }}=0\) for every \(c^{\prime }\ne c\) in the proof of Theorem 1. Formally, for each \(c\in C\), construct \(\succ _{c}\) as follows: Fix some \(P^{\left( 1\right) }\in \mathcal P ^{\left| I\right| }\left\langle c,I\right\rangle \) and let the top-priority student under \(\succ _{c}\), denoted by \(i_{c}^{1}\), be the student in \(\varphi _{c}[P^{\left( 1\right) }]\) (who is unique, if existent, since \(q_{c}=1\)). Iteratively continue, such that: For each \(\ell \ge 2\), fix some \(P^{\left( \ell \right) }\in \mathcal P ^{\left| I\right| }\left\langle c,I\setminus \{i_{c}^{1},\ldots ,i_{c}^{\ell -1}\} \right\rangle \) and let the \(\ell ^{\text {th}}\) priority student under \(\succ _{c}\), denoted by \(i_{c}^{\ell }\), be the student in \(\varphi _{c}[P^{\left( \ell \right) }]\) (who is unique, if existent, since \(q_{c}=1\)). If \(\varphi _{c}[P^{\left( \ell \right) }]=\emptyset \) for any \(\ell \ge 1\), then order all students in \(I \setminus \{i_c^1,\ldots ,i_c^{\ell -1}\}\) so that \(\emptyset \succ _c i\) for all \(i \in I \setminus \{i_c^1,\ldots ,i_c^{\ell -1}\}\). This procedure defines a priority order \(\succ _{c}\).

  27. The paragraph in Sect. F.1 supposes, in addition to favoring higher ranks, resource monotonicity and consistency instead of individual rationality and population monotonicity. However, consistency and resource monotonicity are used only to obtain individual rationality and population monotonicity. Therefore, the example is valid in the current context.

References

  • Abdulkadiroğlu, A., Che, Y., Yasuda, Y.: Expanding choice in school choice. Economic Research Initiatives at Duke Research Paper No. 20 (2008)

  • Abdulkadiroğlu, A., Che, Y., Yasuda, Y.: Resolving conflicting preferences in school choice: the “Boston mechanism” reconsidered. Am. Econ. Rev. 101 (1), 399–410 (2011)

    Google Scholar 

  • Abdulkadiroğlu, A., Pathak, P.A., Roth, A.E., Sönmez, T.: Changing the Boston School Choice Mechanism. Harvard University and Boston College, Working Paper (2006)

  • Abdulkadiroğlu, A., Sönmez, T.: House allocation with existing tenants. J. Econ. Theory 88, 233–260 (1999)

    Article  Google Scholar 

  • Abdulkadiroğlu, A., Sönmez, T.: School choice: a mechanism design approach. Am. Econ. Rev. 93, 729–747 (2003)

    Article  Google Scholar 

  • Abdulkadiroğlu, A., Pathak, P.A., Roth, A.E.: The New York City high school match. Am. Econ. Rev. Pap. Proc. 95, 364–367 (2005)

    Article  Google Scholar 

  • Abdulkadiroğlu, A., Pathak, P.A., Roth, A.E., Sönmez, T.: The Boston public school match. Am. Econ. Rev. Pap. Proc. 95, 368–372 (2005)

    Article  Google Scholar 

  • Abdulkadiroğlu, A., Pathak, P.A., Roth, A.E.: Strategy-proofness versus efficiency in matching with indifferences: redesigning the NYC high school match. Am. Econ. Rev. 99, 1954–1978 (2009)

    Google Scholar 

  • Abebe, R.: School assigning process criticized. The Harvard Crimson, November 25, http://www.thecrimson.com/article/2009/11/25/cambridge-school-choice/ (2009)

  • Afacan, M.O.: Alternative Characterizations of Boston Mechanism, Working Paper (2012)

  • Balinski, M., Sönmez, T.: A tale of two mechanisms: student placement. J.Econ. Theory 84, 73–94 (1999)

    Article  Google Scholar 

  • Bogomolnaia, A., Moulin, H.: A new solution to the random assignment problem. J. Econ. Theory 100, 295–328 (2001)

    Article  Google Scholar 

  • Calsamiglia, C., Haeringer, G., Klijn, F.: Constrained school choice: an experimental study. Am. Econ. Rev. 100, 1860–1874 (2010)

    Google Scholar 

  • Che, Y.K., Kojima, F.: Asymptotic equivalence of random priority and probabilistic serial mechanisms. Econometrica 78, 1625–1672 (2010)

    Article  Google Scholar 

  • Chen, Y., Kesten, O.: From Boston to Shanghai to Deferred Acceptance: Theory and Experiments on A Family of School Choice Mechanisms, Working Paper (2012)

  • Chen, Y.: Characterizing the Boston Mechanism, Working Paper (2012)

  • Chen, Y., Sönmez, T.: School choice: an experimental study. J. Econ. Theory 127, 202–231 (2006)

    Article  Google Scholar 

  • Dur, U.M.: A Characterization of the Top Trading Cycles Mechanism in the School Choice Problem, Working Paper (2012)

  • Ehlers, L., Klaus, B.: Allocation via Deferred Acceptance under Responsive Priorities. Mimeo (2009)

  • Ehlers, L., Klaus, B.: Resource-monotonicity for house allocation problems. Int. J. Game Theory 32, 545–560 (2003)

    Google Scholar 

  • Ehlers, L., Klaus, B.: Efficient priority rules. Games Econ. Behav. 55, 372–384 (2006)

    Article  Google Scholar 

  • Ehlers, L.: Truncation strategies in matching markets. Math. Oper. Res. 33, 327–335 (2008)

    Article  Google Scholar 

  • Ehlers, L., Erdil, A.: Efficient assignment respecting priorities. J. Econ. Theory 145, 1269–1282 (2010)

    Article  Google Scholar 

  • Erdil, A., Ergin, H.: What’s the matter with Tie-Breaking? Improving efficiency in school choice. Am. Econ. Rev. 98, 669–689 (2008)

    Article  Google Scholar 

  • Ergin, H.: Consistency in house allocation problems. J. Math. Econ. 34, 77–97 (2000)

    Article  Google Scholar 

  • Ergin, H.: Efficient resource allocation on the basis of priorities. Econometrica 70, 2489–2497 (2002)

    Article  Google Scholar 

  • Ergin, H., Sönmez, T.: Games of school choice under the Boston mechanism. J. Public Econ. 90, 215–237 (2006)

    Article  Google Scholar 

  • Featherstone, C., Niederle, M.: Ex Ante Efficiency in School Choice Mechanisms: An Experimental Investigation. Stanford University, Working Paper (2008)

  • Gale, D., Shapley, L.S.: College admissions and the stability of marriage. Am. Math. Mon. 69, 9–15 (1962)

    Article  Google Scholar 

  • Haeringer, G., Klijn, F.: Constrained school choice. J. Econ. Theory 144, 1921–1947 (2009)

    Article  Google Scholar 

  • Hylland, A., Zeckhauser, R.: The efficient allocation of individuals to positions. J. Polit. Econ. 87, 293–314 (1979)

    Article  Google Scholar 

  • Kesten, O.: On two competing mechanisms for priority-based allocation problems. J. Econ. Theory 127, 155–171 (2006)

    Article  Google Scholar 

  • Kesten, O.: Why do popular mechanisms lack efficiency in random environments? J. Econ. Theory 144, 2209–2226 (2009)

    Google Scholar 

  • Kesten, O.: School choice with consent. Q. J. Econ. 125 (3), 1297–1348 (2010)

    Google Scholar 

  • Kesten, O.: On two kinds of manipulation for school choice problems. Econ. Theory 51(3), 677–693 (2011)

    Google Scholar 

  • Kesten, O., Ünver, M.U.: A Theory of School Choice Lotteries. Carnegie Mellon University and Boston College, Working Paper (2009)

  • Klaus, B., Klijn, F.: Local and Global Consistency Properties for Student Placement, Working Paper (2011)

  • Kojima, F., Manea, M.: Axioms for deferred acceptance. Econometrica 78, 633–653 (2010a)

    Article  Google Scholar 

  • Kojima, F., Manea, M.: Incentives in the probabilistic serial mechanism. J. Econ. Theory 145, 106–123 (2010b)

    Article  Google Scholar 

  • Konishi, H., Ünver, M.U.: Games of capacity manipulation in the hospital-intern market. Soc. Choice Welf. 27, 3–24 (2006)

    Article  Google Scholar 

  • Manea, M.: Asymptotic ordinal inefficiency of random serial dictatorship. Theor. Econ. 4, 165–197 (2009)

    Google Scholar 

  • Maskin, E.: Nash equilibrium and welfare optimality. Rev. Econ. Stud. 66, 23–38 (1999)

    Article  Google Scholar 

  • Miralles, A.: School Choice: The Case for the Boston Mechanism. Boston University, Working Paper (2008)

  • Moulin, H.: Axioms of Cooperative Decision Making. Cambridge University Press, Cambridge (1991)

    Google Scholar 

  • Moulin, H.: Fair Division and Collective Welfare. The MIT Press, Cambridge (2004)

    Google Scholar 

  • Pais, J., Pinter, A.: School choice and information: an experimental study on matching mechanisms. Games Econ. Behav. 64, 303–328 (2007)

    Article  Google Scholar 

  • Pápai, S.: Strategyproof assignment by hierarchical exchange. Econometrica 68, 1403–1433 (2000)

    Article  Google Scholar 

  • Pathak, P.A., Sönmez, T.: Leveling the playing field: sincere and sophisticated players in the Boston mechanism. Am. Econ. Rev. 98, 1636–1652 (2008)

    Article  Google Scholar 

  • Pathak, P., Sönmez, T.: School Admissions reform in Chicago and England: comparing mechanisms by their vulnerability to manipulation. Am. Econ. Rev. 103 (1), 80–106 (2013)

    Google Scholar 

  • Pycia, M., Ünver, M.U.: Incentive Compatible Allocation and Exchange of Discrete Resources. UCLA and Boston College, Working Paper (2009)

  • Roth, A.E., Sotomayor, M.A.O.: Two-Sided Matching: A Study in Game-Theoretic Modeling and Analysis. Econometric Society Monographs, Cambridge (1990)

    Book  Google Scholar 

  • Roth, A.E.: A natural experiment in the organization of entry level labor markets: regional markets for new physicians and surgeons in the UK. Am. Econ. Rev. 81, 415–440 (1991)

    Google Scholar 

  • Roth, A.E.: Deferred acceptance algorithms: history, theory, practice, and open questions. Int. J. Game Theory 36, 537–569 (2008)

    Article  Google Scholar 

  • Sönmez, T., Ünver, M.U.: Matching, allocation, and exchange of discrete resources. In: Benhabib, J., Bisin, A., Jackson, M.O. (eds.) Social Economics, vol. 1A of Handbooks in Economics, pp. 781–852. Elsevier (2010b)

  • Sönmez, T.: Manipulation via capacities in two-sided matching markets. J. Econ. Theory 77, 197–204 (1997)

    Article  Google Scholar 

  • Sönmez, T., Ünver, M.U.: House allocation with existing tenants: a characterization. Games Econ. Behav. 69, 425–445 (2010a)

    Article  Google Scholar 

  • Takamiya, K.: Coalition strategy-proofness and monotonicity in Shapley-Scarf housing markets. Math. Soc. Sci. 41, 201–213 (2001)

    Article  Google Scholar 

  • Thomson, W.L.: Fair allocation rules. In: Arrow, K., Sen, A., Suzumura, K. (eds) Handbook of Social Choice and Welfare. North-Holland, Amsterdam, New York (forthcoming) (2008)

  • Thomson, W.L.: Monotonic Allocation Mechanisms: Preliminary Results. Mimeo (1978)

  • Thomson, W.L.: The fair division of a fixed supply among a growing population. Math. Oper. Res. 8, 319–326 (1983a)

    Article  Google Scholar 

  • Thomson, W.L.: Problems of fair division and the Egalitarian solution. J. Econ. Theory 31, 211–226 (1983b)

    Article  Google Scholar 

  • Thomson, W.L.: A theory of choice correspondences in economies with a variable number of agents. J. Econ. Theory 46, 237–254 (1988)

    Article  Google Scholar 

  • Velez, R.: Consistent Strategy-Proof Assignment by Hierarchical Exchange, Working Paper (2012)

Download references

Author information

Authors and Affiliations

Authors

Corresponding author

Correspondence to Fuhito Kojima.

Additional information

We are grateful to Eric Budish, John William Hatfield, Daisuke Hirata, Flip Klijn, Morimitsu Kurino, Herve Moulin, and seminar participants at Rice for comments. Bobak Pakzad-Hurson, Fanqi Shi, and Pete Troyan provided excellent research assistance. Ünver thanks NSF for the support of his research. An earlier draft of this paper was titled “The “Boston” School-Choice Mechanism.” We also thank the Editor and three anonymous referees for their excellent suggestions.

Appendices

Appendix A: Proof of Proposition 1

For contradiction, assume that matching \(\mu \) favors higher ranks but is not constrained Pareto efficient. Since \(\mu \) favors higher ranks, there exists no school \(c\) and student \(i\) such that \(i \in I_c, |\mu _{c}|<q_{c}\) and \(cP_{i}\mu _{i}\). This and the assumption that \(\mu \) is not constrained Pareto efficient imply that there exists a sequence of students \(i_{1},i_{2},\ldots ,i_{n}\) such that \(i_k \in I_{\mu _{i_{k+1}}}\), and \(\mu _{i_{k+1}}P_{i_{k}}\mu _{i_{k}}\) or equivalently \(P_{i_{k}}(\mu _{i_{k+1} })<P_{i_{k}}(\mu _{i_{k}})\) for all \(k=1,\ldots ,n\) (with the convention that \(n+1=1\)). Since \(\mu \) favors higher ranks, \(P_{i_{k+1}}(\mu _{i_{k+1} })\le P_{i_{k}}(\mu _{i_{k+1}})\) for all \(k=1,\ldots ,n\). Combining these inequalities, we obtain

$$\begin{aligned} P_{i_{1}}(\mu _{i_{1}})&\le P_{i_{n}}(\mu _{i_{1}})<P_{i_{n}}(\mu _{i_{n}})\le P_{i_{n-1}}(\mu _{i_{n}})<P_{i_{n-1}}(\mu _{i_{n-1}})\\&\le \dots \le P_{i_{1}} (\mu _{i_{2}})<P_{i_{1}}(\mu _{i_{1}}), \end{aligned}$$

a contradiction.

Appendix B: Proof of Lemma 1

Suppose that \(\varphi \) satisfies resource monotonicity and consistency, and fix \(P, q\), and \(i\) arbitrarily. If \(\varphi _{i}[P;q]=\emptyset \), then consistency implies

$$\begin{aligned} \varphi _{j}[P_{i}^{\emptyset },P_{-i};q]=\varphi _{j}[P;q]\,\quad {for all}\,j\not =i\, {and all}\,P_i^\emptyset \in \mathcal P ^\emptyset . \end{aligned}$$

Suppose \(\varphi _{i}[P;q]\ne \emptyset \). Then, by consistency, \(\varphi _{j}\left[ P_{i}^{\emptyset },P_{-i};q_{\varphi _{i}\left[ P;q\right] }-1;q_{-\varphi _{i}\left[ P;q\right] }\right] =\varphi _{j}\left[ P;q\right] \) for all \(j\not =i\) and all \(P_i^\emptyset \in \mathcal P ^\emptyset \). Hence, resource monotonicity implies that for all \(j\ne i\)

$$\begin{aligned} \varphi _{j}\left[ P_{i}^{\emptyset },P_{-i};q\right] R_{j}\varphi _{j}\left[ P_{i}^{\emptyset },P_{-i};q_{\varphi _{i}\left[ P;q\right] }-1,q_{-\varphi _{i}\left[ P;q\right] }\right] =\varphi _{j}\left[ P;q\right] . \end{aligned}$$

The above displayed relations show that \(\varphi \) satisfies population monotonicity.

Appendix C: Proof of Theorem 1

It is straightforward to see that the Boston mechanism for an arbitrary priority profile \(\succ \) satisfies all the axioms in the statement. Thus, we show the converse.

Based on Lemma 1, we invoke population monotonicity of \(\varphi \) in various parts of the proof. For any \(c\in C\) and \(I^{\prime }\subseteq I\), let \(\mathcal P ^{\left| I\right| }\left\langle c;I^{\prime }\right\rangle \) be the set of preference profiles such that all students in \(I^{\prime }\) rank \(c\) as the first choice and all other students rank \(\emptyset \) as the first choice. We prove the theorem in two parts.

1.1 Part 1: Construction of priority profile \(\succ \)

For each \(c\in C\), construct \(\succ _{c}\) as follows: Let \(q_{c}=1\) and \(q_{c^{\prime }}=0\) for all \(c^{\prime }\in C\setminus \{c\}\). Fix some \(P^{\left( 1\right) }\in \mathcal P ^{\left| I\right| }\left\langle c,I\right\rangle \) and let the top-priority student under \(\succ _{c}\), denoted by \(i_{c}^{1}\), be the student in \(\varphi _{c}[P^{1};q]\) if any (who is unique, if existent, since \(q_{c}=1\)). Iteratively continue, such that: For each \(\ell \ge 2\), fix some \(P^{\left( \ell \right) }\in \mathcal P ^{\left| I\right| }\left\langle c,I\setminus \{i_{c}^{1},\ldots ,i_{c}^{\ell -1}\} \right\rangle \) and let the \(\ell ^{\text {th}}\) priority student under \(\succ _{c}\), denoted by \(i_{c}^{\ell }\), be the student in \(\varphi _{c}[P^{\left( \ell \right) };q]\) if any (who is unique, if existent, since \(q_{c}=1\)). If \(\varphi _{c}[P^{\left( \ell \right) };q]=\emptyset \) for any \(\ell \ge 1\), then order all students in \(I \setminus \{i_c^1,\ldots ,i_c^{\ell -1}\}\) so that \(\emptyset \succ _c i\) for all \(i \in I \setminus \{i_c^1,\ldots ,i_c^{\ell -1}\}\). This procedure defines a priority order \(\succ _{c}\). It is an implication of the following claim that the construction of \(\succ _{c}\) is independent of the choice of preference \(P^{\left( \ell \right) }\in \mathcal P ^{\left| I\right| }\left\langle c,I\setminus \{i_{c}^{1},\ldots ,i_{c}^{\ell -1}\} \right\rangle \) in each step \(\ell \):

Claim 1

Let \(I^{\prime }\subseteq I\). Suppose \(\varphi _{c}[P;q]=i\) where \(P\in \mathcal P ^{\left| I\right| }\left\langle c,I^{\prime }\right\rangle \) and \(q_{c}=1,q_{c^{\prime }}=0\) for all \(c^{\prime }\ne c\). Then \(\varphi _{c}[P^{\prime };q^{\prime }]=i\) for all \(q^{\prime }\) with \(q_{c}^{\prime }=1\) and \(P^{\prime }\in \mathcal P ^{\left| I\right| }\left\langle c,I^{\prime \prime }\right\rangle \) with \(\{i\} \subseteq I^{\prime \prime }\subseteq I^{\prime }\).

Proof

We will prove the claim in three steps:

Proof Step 1:

First, we show the claim when \(q^{\prime }=q\) and \(I^{\prime \prime }=I^{\prime } \). Let \(P\in \mathcal P ^{\left| I\right| }\left\langle c,I^{\prime }\right\rangle \) and \(\varphi _{i}[P;q]=c\). Assume \(P_{j}^{\prime }\ne P_{j}\) for some \(j\in I\) and \(P_{k}^{\prime }=P_{k}\) for all \(k\ne j\) without loss of generality. Two cases are possible for the identity of \(j\), where Case 2 has also two sub-cases:

Case 1

\(j=i\): Then, since \(\varphi _{i}[P;q]=c\) and \(c\) is top-ranked both at \(P_{i}\) and \(P_{i}^{\prime }, P^{\prime }\, {m.t.}\,P\) at \(\varphi [P;q]\) and

$$\begin{aligned} U_{k}\left( P,\varphi [P;q]\right)&= U_{k}\left( P^{\prime },\varphi [P;q]\right) \!,\\ V_{k}\left( P,\varphi [P;q]\right)&= V_{k}\left( P^{\prime },\varphi [P;q]\right) \end{aligned}$$

for all \(k\in I\). Thus, \(P^{\prime }\, {r.r.m.t.}\,P\) at \(\varphi [P;q]\). Since \(\varphi \) satisfies rank-respecting invariance, we conclude that \(\varphi [P^{\prime };q]=\varphi [P;q]\) and hence \(\varphi _{i}[P^{\prime };q]=c\).

Case 2

\(j\ne i\) and \(\varphi _{j}[P^{\prime };q]\ne c\):

For any \(P_{j}^{\emptyset }\in \mathcal P ^{\emptyset }\), since \(\varphi \) satisfies population monotonicity, \(\varphi _{i}[P_{j}^{\emptyset } ,P_{-j};q]R_{i}\varphi _{i} [P;q]=c.\) Since \(P_{i}\) top-ranks \(c\), we obtain

$$\begin{aligned} \varphi _{i}[P_{j}^{\emptyset },P_{-j};q]=c. \end{aligned}$$
(1)

Suppose for contradiction that \(\varphi _{i}[P^{\prime };q]\ne c\). Then, since \(\varphi \) favors higher ranks, \(i \in I_c\), and \(\varphi _j[P;q] \ne c\), there exists \(i^{\prime }\ne i,j\) such that \(\varphi _{i^{\prime }}[P^{\prime };q]=c\). Thus, \(P_{i^{\prime }} =P_{i^{\prime }}^{\prime }\) top-ranks \(c\) as \(\varphi \) favors higher ranks (otherwise \(i^{\prime }\) receives \(\emptyset \) by individual rationality) and, by population monotonicity,

$$\begin{aligned} \varphi _{i^{\prime }}[P_{j}^{\emptyset },P_{-j}^{\prime };q]=c. \end{aligned}$$
(2)

Relations (1) and (2) contradict each other since \(i^{\prime }\ne i, q_{c}=1,\) and \(P_{-j}^{\prime }=P_{-j}\).

Case 2-(ii)

\(j\ne i\) and \(\varphi _{j}[P^{\prime };q]=c\):

Then, since both \(P_{j}\) and \(P_{j}^{\prime }\) top-rank \(c\),Footnote 25 \(P\, {m.t.}\,P^{\prime }\) at \(\varphi [P^{\prime };q]\) and

$$\begin{aligned} U_{k}\left( P,\varphi [P';q]\right)&= U_{k}\left( P^{\prime },\varphi [P';q]\right) \!,\\ V_{k}\left( P,\varphi [P';q]\right)&= V_{k}\left( P^{\prime },\varphi [P';q]\right) \end{aligned}$$

for all \(k\in I\). Thus, \(P\, {r.r.m.t.}\,P^{\prime }\) at \(\varphi [P^{\prime };q]\). Since \(\varphi \) satisfies rank-respecting invariance we conclude \(\varphi _{j}[P;q]=c\), a contradiction to \(j\ne i, \varphi _{i}[P;q]=c\) and \(q_{c}=1\).

Proof Step 2:

Given the preceding argument, population monotonicity of \(\varphi \) implies that the claim holds for all cases when \(q^{\prime }=q\) and \(\{i\} \subseteq I^{\prime \prime }\subseteq I^{\prime }\), noting that \(c\) is the top-ranked school at \(P_{i}.\)

Proof Step 3:

: Finally, we will show that the claim holds for all \(q^{\prime }\) with \(q_{c}^{\prime }=1\) and \(P^{\prime }\in \mathcal P ^{\left| I\right| }\left\langle c,I^{\prime \prime }\right\rangle \) with \(\{i\} \subseteq I^{\prime \prime }\subseteq I^{\prime }\). Thus, assume that \(q^{\prime }\) satisfies \(q_{c} ^{\prime }=1\) and \(P^{\prime }\in \mathcal P ^{\left| I\right| }\left\langle c,I^{\prime \prime }\right\rangle \) with \(\{i\} \subseteq I^{\prime \prime }\subseteq I^{\prime }\). By Proof Step 2,

$$\begin{aligned} \varphi _{i}[P^{\prime };q]=c. \end{aligned}$$
(3)

Since \(\varphi \) satisfies resource monotonicity and \(q_{c}^{\prime }\ge q_{c}\) for all \(c\in C\) (as \(q_{c}^{\prime }=q_{c}=1\) and \(q_{c^{\prime }}^{\prime } \ge 0=q_{c^{\prime }}\) for all \(c^{\prime }\ne c\)),

$$\begin{aligned} \varphi _{i}[P^{\prime };q^{\prime }]R_{i}^{\prime } \varphi _{i}[P^{\prime };q]. \end{aligned}$$
(4)

Relations (3) and (4) imply

$$\begin{aligned} \varphi _{i}[P^{\prime };q^{\prime }]R_{i}^{\prime }c. \end{aligned}$$
(5)

Since \(c\) is top-ranked under \(P_{i}^{\prime }\) by assumption, relation (5) implies

$$\begin{aligned} \varphi _{i}[P^{\prime };q^{\prime }]=c, \end{aligned}$$

completing the proof. \(\square \)

1.2 Part 2: Proof that \(\varphi =\psi ^{\succ }\)

Let \(\succ =(\succ _c)_{c \in C}\) be the priority order profile constructed in Part 1. For any given preference profile \(P\) and quota profile \(q\), we will show that \(\varphi [P;q]=\psi ^{\succ }[P;q]\). Construct the following student sets and quotas:

For any \(c\in C\cup \left\{ \emptyset \right\} \), define

$$\begin{aligned} I_{c}(0)&=\emptyset ,\\ J_{c}(0)&=\emptyset ,\\ q_{c}(0)&=q_{c}. \end{aligned}$$

For any \(\ell \ge 1\), given \((I_{c}(0),J_{c}(0),q_{c}(0))_{c\in C\cup \left\{ \emptyset \right\} },\ldots ,(I_{c}(\ell -1),J_{c}(\ell -1),q_{c}(\ell -1))_{c\in C\cup \left\{ \emptyset \right\} }\), recursively define

$$\begin{aligned} I_{c}(\ell )&= \left\{ i\in I_c \setminus \left( { \bigcup \limits _{d\in C\cup \left\{ \emptyset \right\} }} \bigcup _{\ell ^{\prime }=1}^{\ell -1}J_{d}(\ell ^{\prime })\right) :P_{i}(c)=\ell \right\} ,\\ J_{c}(\ell )&= \left\{ i\in I_{c}(\ell ):\varphi _{i}[P;q]=c\right\} ,\\ q_{c}(\ell )&= q_{c}(\ell -1)-|J_{c}(\ell -1)|. \end{aligned}$$

\(I_{c}(\ell )\) is the set of qualified students who rank \(c\) as their \(\ell ^{\text {th}}\) choice and have not received any higher-ranked school. \(J_{c}(\ell )\) is the set of students in \(I_{c}(\ell )\) who receive seats at \(c\) under \(\varphi \left[ P;q\right] \). \(q_{c}(\ell )\) denotes the number of seats remaining after assigning seats to students who rank \(c\) as the \((\ell -1)^{\text {st}}\) choice or higher. Let \(\succ _{\emptyset }\in \Pi \) be an arbitrary priority order. Individual rationality of \(\varphi \) (following from resource monotonicity of \(\varphi \)) implies \(I_\emptyset =I\), and it suffices to show the following claim.

Claim 2

For all \(\ell \ge 1,\,J_{c}(\ell )=\arg \max _{I_{c}(\ell ),q_{c}(\ell )}\succ _{c}\) for all \(c\in C\cup \left\{ \emptyset \right\} \), where

$$\begin{aligned} \arg \max _{I_{c}(\ell ),q_{c}(\ell )}\succ _{c}:=\left\{ i\in I_{c}(\ell ):\left| \left\{ j\in I_{c}(\ell ):j\succeq _{c}i\right\} \right| \le q_{c}(\ell )\right\} , \end{aligned}$$

is the set of (at most) \(q_{c}(\ell )\) students who have the highest priorities at \(c\) among those in \(I_{c}(\ell )\).

Proof

Let \(c\in C\cup \left\{ \emptyset \right\} .\) Fix \(\ell \ge 1\). If \(|I_{c}(\ell )|\le q_{c}(\ell )\), then \(\arg \max _{I_{c}(\ell ),q_{c}(\ell )}\succ _{c}=I_{c}\left( \ell \right) \). Then \(J_{c}(\ell )=I_{c}(\ell )\) because \(\varphi \) favors higher ranks. Hence, the conclusion \(J_{c}\left( \ell \right) =\arg \max _{I_{c}(\ell ),q_{c}(\ell )}\succ _{c}\) holds.

Thus, assume \(|I_{c}(\ell )|>q_{c}(\ell )\). Suppose for contradiction that the conclusion \(J_{c}\left( \ell \right) =\arg \max _{I_{c}(\ell ),q_{c}(\ell )}\succ _{c}\) does not hold. Since \(\varphi \) favors higher ranks, it follows that \(|J_{c}(\ell )|=q_{c}(\ell )\), and hence, there exist \(i\in \arg \max _{I_{c} (\ell ),q_{c}(\ell )}\succ _{c}\) such that \(\varphi _{i}[P;q]\ne c\) and \(j\in I_{c}(\ell )\setminus \left( \arg \max _{I_{c}(\ell ),q_{c}(\ell )}\succ _{c}\right) \) such that \(\varphi _{j}[P;q]=c\).

Let \(P^{\prime }=(P_{i},P_{j},P_{-i,j}^{\emptyset })\) for some \(P_{-i,j} ^{\emptyset }\in \left( \mathcal P ^{\emptyset }\right) ^{\left| I\right| -2}.\) Let \(q^{\prime }=(q_{c^{\prime }}^{\prime })_{{c^{\prime }}\in C}\) be defined by

$$\begin{aligned} q_{c^{\prime }}^{\prime }=q_{c^{\prime }}-\left| \{k\in I\setminus \{i,j\} : \varphi _{k}[P;q]=c^{\prime }\} \right| \end{aligned}$$

for each \(c^{\prime }\in C\). By construction of \(I_{c}\left( \ell \right) ,\) since \(i\in I_{c}\left( \ell \right) \), she has not been matched to a higher-choice school than \(c\). Thus, \(cP_{i}\varphi _{i}[P;q]\). Since \(\varphi \) is consistent,

$$\begin{aligned} \varphi _{j}[P^{\prime };q^{\prime }]&= \varphi _{j}[P;q]=c,\\ \varphi _{i}[P^{\prime };q^{\prime }]&= \varphi _{i}[P;q]\quad \text {and hence,}\ cP_{i}\varphi _{i}[P^{\prime };q^{\prime }]. \end{aligned}$$

Consider preference \(P_{i}^{\prime \prime },P_{j}^{\prime \prime }\) such that \(c\) is top-ranked at \(P_{i}^{\prime \prime }\) and \(P_{j}^{\prime \prime }\) and relative rankings of all other schools are unchanged from \(P_{i}^{\prime }\) and \(P_{j}^{\prime }\), respectively. Then \(P^{\prime \prime }:=(P_{i}^{\prime \prime },P_{j}^{\prime \prime },P_{-i,j}^{\prime })\, {r.r.m.t.}\,P^{\prime }\) at \(\varphi [P^{\prime };q^{\prime }]\). Thus, since \(\varphi \) satisfies rank-respecting invariance, \(\varphi _{i}[P^{\prime \prime };q^{\prime }]\ne c\). This is a contradiction since Claim 1 and the assumption that \(i\) has a higher priority than \(j\) at \(c\) imply \(\varphi _{i}[P^{\prime \prime };q^{\prime }]=c\). \(\square \)

Claim 2 completes the proof of the Theorem.

Appendix D: Proof of Theorem 2

The Boston mechanism for an arbitrary priority profile clearly satisfies all the axioms in the statement. Let \(q\) be the quota vector with \(q_{c}=1\) for all \(c \in C\), which is fixed throughout the current analysis.

1.1 Part 1: Construction of priority profile \(\succ \)

Construct \(\succ _{c}\) in an analogous manner to the one in the proof of Theorem 1.Footnote 26 It is an implication of the following claim that the construction of \(\succ _{c}\) is independent of the choice of preference \(P^{\left( \ell \right) }\in \mathcal P ^{\left| I\right| }\left\langle c,I\setminus \{i_{c}^{1},\ldots ,i_{c}^{\ell -1}\} \right\rangle \) in each step \(\ell \):

Claim 3

Let \(I^{\prime }\subseteq I\). Suppose \(\varphi _{c}[P]=i\) where \(P\in \mathcal P ^{\left| I\right| }\left\langle c,I^{\prime }\right\rangle .\) Then \(\varphi _{c}[P^{\prime }]=i\) for all \(P^{\prime } \in \mathcal P ^{\left| I\right| }\left\langle c,I^{\prime \prime }\right\rangle \) with \(\{i\} \subseteq I^{\prime \prime }\subseteq I^{\prime }\).

Proof

The proof is omitted since it is identical to Steps 1 and 2 of the proof of Claim 1. \(\square \)

1.2 Part 2: Proof that \(\varphi =\psi ^{\succ }\)

Let \(\succ =(\succ _c)_{c \in C}\) be the priority order profile constructed in Part 1. For any given preference profile \(P\), we will show that \(\varphi [P]=\psi ^{\succ }[P]\). For each \(c \in C \cup \{ \emptyset \}\) and \(\ell =1,2,\ldots ,\) define \(I_{c}(\ell ), J_{c}(\ell )\), and \(q_{c}(\ell )\) as in Part 2 of the Proof of Theorem 1. Recall that \(I_{c}(\ell )\) is the set of students who rank \(c\) as their \(\ell ^{\text {th}}\) choice and have not received any higher-ranked school (except those who cannot be matched to \(c\) under any preference profile), \(J_{c}(\ell )\) is the set of students in \(I_{c} (\ell )\) who receive seats at \(c\) under \(\varphi \left[ P \right] \), and \(q_{c}(\ell ) \in \{ 0, 1 \}\) denotes the number of seats in \(c\) remaining after assigning seats to students who rank \(c\) as the \((\ell -1)^{\text {st}}\) choice or higher. Let \(\succ _{\emptyset }\in \Pi \) be an arbitrary priority order. Individual rationality of \(\varphi \) implies \(I_\emptyset =I\), and it suffices to show the following claim.

Claim 4

For all \(\ell \ge 1, J_{c}(\ell )=\arg \max _{I_{c}(\ell ),q_c(\ell )}\succ _{c}\) for all \(c\in C\cup \left\{ \emptyset \right\} \), where

$$\begin{aligned} \arg \max _{I_{c}(\ell ),q_{c}(\ell )}\succ _{c}:=\left\{ i\in I_{c}(\ell ):\left| \left\{ j\in I_{c}(\ell ):j\succeq _{c}i\right\} \right| \le q_{c}(\ell )\right\} , \end{aligned}$$

is the set of (at most) \(q_{c}(\ell )\) students who have the highest priorities at \(c\) among those in \(I_{c}(\ell )\).

Proof

Let \(c\in C\cup \left\{ \emptyset \right\} \) . Fix \(\ell \ge 1\). If \(|I_{c}(\ell )|\le q_c(\ell )\), then \(\arg \max _{I_{c}(\ell ),q_c(\ell )}\succ _{c}=I_{c}\left( \ell \right) \). Then \(J_{c}(\ell )=I_{c}(\ell )\) because \(\varphi \). Hence, the conclusion \(J_{c}\left( \ell \right) =\arg \max _{I_{c}(\ell ),q_{c}(\ell )} \succ _{c}\) holds.

Thus, assume \(|I_{c}(\ell )|>q_c(\ell )\). Suppose for contradiction that the conclusion does not hold. Since \(\varphi \) favors higher ranks, it follows that \(|J_{c}(\ell )|=1\), and hence, there exist \(i\in \arg \max _{I_{c} (\ell )}\succ _{c}\) such that \(\varphi _{i}[P]\ne c\) and \(j\in I_{c}(\ell )\setminus \left( \arg \max _{I_{c}(\ell ),q_c(\ell )}\succ _{c}\right) \) such that \(\varphi _{j}[P]=c\).

Consider preference profile \(P^{\prime }\) such that (1) at \(P_{k}^{\prime }\) for every student \(k \in I_{c}(\ell ), c\) is top ranked and relative rankings of all other schools are unchanged from \(P_{k}\), and (2) preferences of all other students are unchanged. Since \(q_{c}=1\), it follows that \(P^{\prime }\, {r.r.m.t.}\,P\) at \(\varphi [P]\). Thus, since \(\varphi \) satisfies rank-respecting invariance, \(\varphi _{j}[P^{\prime }]= c\).

Let \(P^{\prime \prime }=(P_{i}^{\prime },P_{j}^{\prime },P_{-i,j}^{\emptyset })\) for some \(P_{-i,j}^{\emptyset }\in \left( \mathcal P ^{\emptyset }\right) ^{\left| I\right| -2}.\) Since \(\varphi \) satisfies population monotonicity,

$$\begin{aligned} \varphi _{j}[P^{\prime \prime }]R_{j}^{\prime }\varphi _{j}[P^{\prime }]=c. \end{aligned}$$

This and the assumption that \(c\) is top-ranked at \(P_{j}^{\prime }\) imply

$$\begin{aligned} \varphi _{j}[P^{\prime \prime }]=c. \end{aligned}$$

Since \(q_{c}=1\), the above relation implies

$$\begin{aligned} \varphi _{i}[P^{\prime \prime }]\ne c, \end{aligned}$$

which is a contradiction since Claim 3 and the construction of \(\succ \) that gives \(i\) a higher priority than \(j\) at \(c\) imply \(\varphi _{i}[P^{\prime \prime }]=c\). \(\square \)

Claim 4 completes the proof of the Theorem.

Appendix E: A welfare maximization interpretation of the Boston mechanism

For each problem \(\left[ P;q\right] ,\) the solution of the following linear assignment program, which maximizes the sum of the induced utilities, is equivalent to the solution of the Boston mechanism induced by a priority profile \(\succ \): First, define for each \(c \in C, I_c=\{i \in I : i \succ _c \emptyset \}\) as the set of acceptable students for school \(c\).

$$\begin{aligned}&\max _{\left[ z_{i,c} \right] } {\sum \limits _{i\in I,c\in C\cup \left\{ \emptyset \right\} }} z_{i,c}u_{i}\left( c\right) \\&\quad \text {subject to}\\&\quad i \not \in I_c \Rightarrow z_{i,c} =0\quad \quad \forall i \in I, c \in C . \\&\quad 0 \le z_{i,c}\quad \quad \forall i\in I,c\in C\cup \left\{ \emptyset \right\} \\&\quad {\sum \limits _{c\in C\cup \left\{ \emptyset \right\} }} z_{i,c} =1\quad \quad \forall i\in I\\&\quad {\sum \limits _{i\in I}} z_{i,c} \le q_{c}\quad \quad \forall c\in C\cup \left\{ \emptyset \right\} , \end{aligned}$$

where \(q_{\emptyset }=\infty \) and \(\left( u_{i}\right) _{i\in I} \subseteq (\mathbb R _{+}^{\left| C\right| \cup \left\{ \emptyset \right\} })^{|I|}\) are utility functions consistent with the given preferences \(P\) (i.e., for any \(i \in I\) and \(c,d \in C \cup \{\emptyset \}\), we have \(u_i(c)>u_i(d)\) if and only if \(c \, P_i \, d\)) satisfying

$$\begin{aligned} \left| I\right| u_{i}\left( c\right)&< u_{j}\left( d\right) \quad \quad \forall i,j\in I\text { and }c,d\in C\cup \left\{ \emptyset \right\} \text { with }P_{j}\left( c\right) <P_{i}\left( d\right) \\ \left| I\right| u_{i}\left( c\right)&< u_{j}\left( c\right) \quad \quad \forall c\in C\cup \left\{ \emptyset \right\} \text {with } P_{i}\left( c\right) =P_{j}\left( c\right) \text { and }\succ _{c}\left( j\right) <\succ _{c}\left( i\right) \end{aligned}$$

In the solution \(\left[ z_{i,c}\right] _{i\in I,c\in C\cup \left\{ \emptyset \right\} }\), which is unique and where \(z_{i,c}\in \left\{ 0,1\right\} \) for all \(i\in I\) and \(c\in C\cup \left\{ \emptyset \right\} \), school \(i\) is matched with school \(c\) if and only if \(z_{i,c}=1\).

Appendix F: Independence of axioms for remaining cases

The main text has presented examples showing that the axioms in the characterizations are independent for all but a few values of \(|I|\) and \(|C|\). This section completes the investigation by considering all other cases.

1.1 Axioms for Theorem 1

A mechanism violating only consistency: Suppose \(\left| I\right| \le 2\): If \(\left| I\right| = 1\), then consistency is vacuously satisfied by any mechanism. If \(\left| I\right| =2\), then favoring higher ranks implies consistency. To see this first note that, for any \(i \in I\) and \(j \ne i, \varphi _i[P;q]\) is the most preferred school in \(\{c \in C \cup \{ \emptyset \}: i \in I_{c}, q_{c}- \mathbf 1 _{\varphi _j[P;q]=c} \ge 1 \}\) by favoring higher ranks, where \(\mathbf 1 _{\varphi _j[P;q]=c}=1\) if \(\varphi _j[P;q]=c\) and \(0\) otherwise. By inspection, favoring higher ranks implies that \(\varphi _i[P_i,P^\emptyset _j;(q_{c}-\mathbf 1 _{\varphi _j[P;q]=c})_{c \in C}]\) is the most preferred school in \(\{c \in C \cup \{ \emptyset \}: i \in I_{c}, q_{c}- \mathbf 1 _{\varphi _j[P;q]=c} \ge 1 \}\), showing consistency. Thus, there is no mechanism that violates consistency while favoring higher ranks.

A mechanism violating only resource monotonicity: Suppose that \(|I|=1\) or \(|C|=1\): If \(|I|=1\), then favoring higher ranks implies resource monotonicity. To see this point observe that the unique agent, denoted \(i\), receives her most preferred school in \(\{c \in C: i \in I_c, q_c \ge 1\}\). Since this set is increasing in each \(q_c\) in the set inclusion sense, resource monotonicity follows.

If \(|C|=1\), then favoring higher ranks, consistency, and rank-respecting invariance imply resource monotonicity. To show this, first recall that favoring higher ranks implies individual rationality (since \(I_\emptyset =I\) and \(q_\emptyset =\infty \)). Now suppose, for contradiction, a mechanism \(\varphi \) satisfies favoring higher ranks, consistency, and rank-respecting invariance, while violating resource monotonicity. Then there exists a student \(i \in I\), preference profile \(P\), and a quota \(q_c\) of the unique school \(c\) such that

$$\begin{aligned}&\varphi _i[P;q_c-1]=c, \end{aligned}$$
(6)
$$\begin{aligned}&\varphi _i[P;q_c] = \emptyset ,\end{aligned}$$
(7)
$$\begin{aligned}&c \, P_i \, \emptyset . \end{aligned}$$
(8)

By relationships (6)–(8) and favoring higher ranks, \(|\varphi _c[P;q_c]|=q_c\), and hence, there exists a student \(j \ne i\) such that

$$\begin{aligned} \varphi _j[P;q_c]&= c,\end{aligned}$$
(9)
$$\begin{aligned} \varphi _j[P;q_c-1]&= \emptyset . \end{aligned}$$
(10)

By consistency of \(\varphi \) and relation (9), it follows that

$$\begin{aligned} \varphi _i[P^\emptyset _j,P_{-j};q_c-1]=\varphi _i[P;q_c]. \end{aligned}$$
(11)

Relations (7) and (11) imply

$$\begin{aligned} \varphi _i[P^\emptyset _j,P_{-j};q_c-1]=\emptyset . \end{aligned}$$
(12)

Meanwhile relation (10) implies that \((P^\emptyset _j,P_{-j}) \, {r.r.m.t.} \, P\) at \(\varphi [P;q_c-1]\). Thus, by rank-respecting invariance of \(\varphi \),

$$\begin{aligned} \varphi _i[P^\emptyset _j,P_{-j};q_c-1]=\varphi _i[P;q_c-1]. \end{aligned}$$
(13)

Then, by relations (6) and (13), we obtain

$$\begin{aligned} \varphi _i[P^\emptyset _j,P_{-j};q_c-1]=c. \end{aligned}$$
(14)

Relations (12) and (14) contradict each other, showing that \(\varphi \) is resource monotonic.

A mechanism violating only rank-respecting invariance: Suppose that \(\left| C\right| =1\) or \(|I|=1\): If \(|C|=1\), then consistency, resource monotonicity, and favoring higher ranks imply rank-respecting invariance. To see this point, first recall that these properties imply individual rationality and population monotonicity. Let \(\varphi \) be a mechanism for that these axioms hold and for each \(i \in I\) let

$$\begin{aligned} \begin{array}[c]{ll} P'_i &{} P_i'' \\ \hline c &{} \emptyset \\ \emptyset &{} c \end{array} \end{aligned}$$

Fix a preference profile \(P\) arbitrarily. Since \(\varphi \) satisfies individual rationality,

$$\begin{aligned} \varphi _i[P;q]=c \Rightarrow P_i=P_i' \end{aligned}$$
(15)

Moreover, if \(\varphi _i[P;q]\) is the top-ranked school for \(i\) at \(P_i\), then the only monotonic transformation of \(P_i\) at \(\varphi _i[P;q]\) is \(P_i\) itself. By this fact and (15), if \(\tilde{P}\) is a monotonic transformation of \(P\) at \(\varphi [P;q]\), then for any \(i \in I\), either

  1. 1.

    \(\tilde{P}_i=P_i\), or

  2. 2.

    \(P_i=P_i', \varphi _i[P;q]=\emptyset \), and \(\tilde{P}_i=P_i''.\)

For any \(i\) such that Case 1 above applies, population monotonicity of \(\varphi \) and Cases 1 and 2 imply \(\varphi _i[\tilde{P};q] \, R_i \varphi _i[P;q]\). So, if \(\varphi _i[P;q]=c\), then \(\varphi _i[\tilde{P};q]=c\). Since \(\varphi \)favors higher ranks, it follows that \(\varphi _i[P;q]=\emptyset \) implies \(\varphi _i[\tilde{P};q]=\emptyset \). Thus, we conclude \(\varphi _i[P;q]=\varphi _i[\tilde{P};q]\). For any \(i\) such that Case 2 above applies, individual rationality of \(\varphi \) implies \(\varphi _i[\tilde{P};q]=\emptyset =\varphi _i[P;q]\). Therefore, we conclude \(\varphi [\tilde{P};q]=\varphi [P;q]\), showing rank-respecting invariance.

If \(|I|=1\), then favoring higher ranks implies rank-respecting invariance. To show this, let \(i\) be the unique student in \(I\) and \(C_i=\{c \in C \cup \{ \emptyset \} : i \in I_c, q_c \ge 1\}\). Since \(i\) is the unique student, favoring higher ranks imply that \(\varphi _i[P_i;q]\) is the school that is top-ranked by \(P_i\) within \(C_i\). Any monotonic transformation \(P'_i\) of \(P_i\) at \(\varphi _i[P_i;q]\) leaves the top-ranked school in \(C_i\) unchanged (namely \(\varphi _i[P_i;q]\)), so \(\varphi _i[P_i';q]=\varphi _i[P_i;q]\) by favoring higher ranks. This shows that \(\varphi \) satisfies rank-respecting invariance.

1.2 Axioms for Theorem 2

A mechanism violating only favoring higher ranks: Suppose that \(|C|=1\) or \(|I|=1\): If \(|C|=1\), then any mechanism \(\varphi \) that satisfies individual rationality and rank-respecting invariance favors higher ranks. To see this, let \(C=\{c\}\) and assume for contradiction that \(\varphi \) does not favor higher ranks. Then there exists \(i \in I_c\) (so \(\varphi _i[P]= c\) for some \(P\)) such that \(c P'_i \varphi _i[P']\) for some \(P'\) where either \(\varphi _c[P']=\emptyset \) or \(\varphi _c[P']=j\) with \(P'_i(c)<P'_j(c)\). The latter is a contradiction to individual rationality of \(\varphi \), because \(P'_i(c)<P'_j(c)\) and \(|C|=1\) imply \(\emptyset P'_j c\). Thus, assume \(\varphi _c[P']=\emptyset \). Since \(q_c=1\) and \(C=\{c\}\), it follows that \(\varphi _j[P]=\varphi _j[P']=\emptyset \) for all \(j \ne i\). Also note that \(P_i=P_i'\) because \(\varphi _i[P]=c\) implies \(c P_i \emptyset \) by individual rationality of \(\varphi , c P'_i \emptyset \) from before, and \(|C|=1\). Hence, \((P_i,P_{-i}^\emptyset ) {r.r.m.t.} P\) at \(\varphi [P]\) and \((P_i,P_{-i}^\emptyset ) {r.r.m.t.} P'\) at \(\varphi [P']\) for any \(P_{-i}^\emptyset \in \mathcal ( P^\emptyset )^{|I|-1}\). Thus, rank-respecting invariance implies \(c=\varphi _i[P]=\varphi _i[P_i,P_{-i}^\emptyset ]=\varphi _i[P']=\emptyset \), a contradiction.

If \(|I|=1\), then rank-respecting invariance implies favoring higher ranks. To see this point suppose for contradiction that a mechanism \(\varphi \) violates favoring higher ranks while satisfying rank-respecting invariance. Then, for the unique student \(i\), there exist \(P_i,P_i'\) such that

$$\begin{aligned} \varphi _i[P_i'] P_i \varphi _i[P_i]. \end{aligned}$$
(16)

Then a preference \(P_i''\) such that

$$\begin{aligned} \begin{array}{c} P''_{i}\\ \hline \varphi _i[P_i']\\ \varphi _i[P_i]\\ \vdots \end{array} \end{aligned}$$

satisfies \(P_i'' \, {r.r.m.t.} \, P_i\) at \(\varphi [P_i]\) and \(P_i'' \, {r.r.m.t.} \, P_i'\) at \(\varphi [P_i']\). By rank-respecting invariance, we obtain

$$\begin{aligned} \varphi [P_i]=\varphi [P_i'']=\varphi [P'_i], \end{aligned}$$

a contradiction to relation (16).

A mechanism violating only population monotonicity: Suppose that \(|C|=1\) or \(|I| \le 2\): If \(|I|=1\), then population monotonicity is vacuously satisfied. If \(|I|=2\), then individual rationality and favoring higher ranks imply population monotonicity. To see this point, let \(I=\{1,2\}\) and \(P_2^\emptyset \in \mathcal P ^\emptyset \). Consider an arbitrary preference profile \(P\). Since \(\varphi \) is individually rational, \(\varphi _1[P] \, R_1 \, \emptyset \). If \(\varphi _1[P]=\emptyset \), then individual rationality of \(\varphi \) implies that \(\varphi _1[P_1,P_2^\emptyset ] \,R_1 \,\emptyset =\varphi _1[P]\), satisfying the conclusion of population monotonicity. Thus, suppose \(\varphi _1[P] \ne \emptyset \) and, for contradiction, that \(\varphi _1[P] \, P_1 \, \varphi _1[P_1,P^\emptyset _2]\). Then, by favoring higher ranks, \(\varphi _2[P_1,P^\emptyset _2]=\varphi _1[P]\). However, \(\emptyset \,P_2^\emptyset \, \varphi _1[P]\) since \(P_2^\emptyset \) top-ranks \(\emptyset \), which contradicts individual rationality of \(\varphi \). By a symmetric argument, the conclusion of population monotonicity holds for matchings of student \(2\), showing that population monotonicity holds.

If \(|C|=1\), then individual rationality and rank-respecting invariance imply population monotonicity. To see this point, first note that \(\varphi _i[P] \, R_i \, \emptyset \) for all \(i\) and \(P\) since \(\varphi \) is individually rational. Hence, if \(\varphi _i[P]=\emptyset \), then \(\varphi _i[P^\emptyset _j,P_{-j}] R_i \emptyset =\varphi _i[P]\) for any \(i \in I, j \ne i\) and \(P^\emptyset _j \in \mathcal P ^\emptyset \); thus, the conclusion of population monotonicity holds. So suppose \(\varphi _i[P]=c \ne \emptyset \). Then, since \(|C|=1\) and \(q_c=1, \varphi _k[P]=\emptyset \) for all \(k \ne i\). Thus, for any \(j \ne i\) and \(P^\emptyset _j \in \mathcal P ^\emptyset \), we have \((P^\emptyset _j,P_{-j}) \, {r.r.m.t.} \, P\) at \(\varphi [P]\). By rank-respecting invariance of \(\varphi \), we obtain \(\varphi _i[P^\emptyset _j,P_{-j}]=\varphi _i[P]\), showing that the conclusion of population monotonicity holds. These arguments show the claim.

A mechanism violating only rank-respecting invariance: Suppose that \(|I|=1\) or \(|C|=1\): If \(|I|=1\) or \(|C|=1\), then the paragraph on rank-respecting invariance in Sect. F.1 shows that there exists no mechanism that satisfies favoring higher ranks, individual rationality, and population monotonicity, and yet violates rank-respecting invariance.Footnote 27

Rights and permissions

Reprints and permissions

About this article

Cite this article

Kojima, F., Ünver, M.U. The “Boston” school-choice mechanism: an axiomatic approach. Econ Theory 55, 515–544 (2014). https://doi.org/10.1007/s00199-013-0769-8

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: https://doi.org/10.1007/s00199-013-0769-8

Keywords

JEL Classification

Navigation