Abstract
This paper studies an optimal assignment problem of heterogenous students to schools with a particular kind of preference complementarity: peer effects, defined by the average ability of those in the same school. The tractability of the problem allows us to characterize the optimal assignment mechanism, which has a simple “(stochastic) pass-fail” structure. Its shape is mainly determined by the convexity/concavity of the attainment function, interpreted as the preference for/against having diverse-ability students in different schools. We also provide comparative statics as to when more or less mixture of heterogenous ability types would be desirable.
Similar content being viewed by others
Notes
See Winkler (1988) for a general treatment of the idea of the extreme points of moment sets. Our simplifying assumptions allow us to provide more direct proofs, but his results should be useful in more general environments.
Recently, the literature proposes more elaborated models of peer effects, for example, allowing for heterogeneity and network structure (see, for example, Epple and Romano 2011, and the references therein). We focus on the simple and classical specification, in order to highlight the main insights in the optimal design problem. Even in such simple models, not much has been known about the optimal assignment. Nevertheless, studying the optimal assignment with such elaborated models of externality would be interesting for future research.
See also Che et al. (2022) for an equilibrium model with prestige-seeking agents.
The case of asymmetric schools (e.g., one school is exogenously given a better amenity than the other school) would add an interesting, though somewhat orthogonal, dimension to the problem. See Sect. 7.2.
In a short run, it seems difficult to drastically change each school’s capacity, even if the planner knew that such a change would be socially beneficial. The possibility of changing capacities is an important problem from a longer-run perspective, where a student’s u does not only depend on his ability \(\theta \) and the mean ability \(x_s\) of a school s, but also on its capacity \(Q_s\). Such a model may be analyzed as a generalization of the current problem, and is left for future research.
Supermodularity is one of the standard assumptions in the related literature (see Board 2009, for example), while in the education literature, whether this sort of supermodularity is a relevant assumption has been discussed. In our paper, in order to highlight its main point about the effect of externality on the shape of the optimal allocation, we focus on this standard assumption in the literature.
The case with \(g''\le 0\) is just its mirror image, and briefly discussed later.
As such, the incentive compatibility condition here corresponds to the equilibrium condition in the previous section. However, it may also be interpreted as a sort of a fairness concern. Specifically, the condition requires that a student with type \(\theta \) does not envy another student with type \(\theta '\). This fairness interpretation could particularly be relevant in the context where the principal can observe \(\theta \) and enforce any student allocation subject to the envy-freeness constraint.
See Börgers et al. (2015), for example.
Recall our (innocuous) restriction that \(x_1\ge x_0=1-x_1\), implying \(x_1\ge \frac{1}{2}\).
It could be done either fully randomly, or more selectively as long as the selection criterion is uncorrelated with \(\theta \) (e.g., based on locations, demographic information, etc.).
Recall that School 0 comprises the failed students only, while School 1 comprises a mixture of the failed and passed students. First, School 0 would be better off by having some passed students, but no passed student has such an incentive. School 1 is already best-off by taking all the passed students. By just knowing pass/fail labeling, it is impossible for them to find a mutually better match.
Again, this rationing part may be done either fully randomly, or based on the selection criterion that is uncorrelated with \(\theta \).
It seems reasonable to guess that the degree of complementarity between the school amenity level and the peer effect would play the key role.
This includes the possibility that \(\int _0^y (\hat{\Pi }(\theta )-\Pi ^*(\theta ))d\theta \) is non-decreasing for all \(y\in (y_1,y_2)\), which corresponds to the case with \(z=y_2\); similarly, the possibility that \(\int _0^y (\hat{\Pi }(\theta )-\Pi ^*(\theta ))d\theta \) is non-increasing for all \(y\in (y_1,y_2)\) (with \(z=y_1\)).
References
Arnott R, Rowse J (1987) Peer group effects and educational attainment. J Public Econ 32(3):287–305
Bhaskar D, Sadler E (2020) Resource allocation with positive externalities. Discussion paper
Bizzotto J, Vigier A (2023) School-based statistical discrimination. Discussion paper
Board S (2009) Monopolistic group design with peer effects. Theor Econ 4(1):89–125
Börgers T, Krahmer D, Strausz R (2015) An Introduction to the Theory of Mechanism Design , no. 9780199734023 in OUP Catalogue. Oxford University Press, Oxford
Carroll G, Segal I (2019) Robustly optimal auctions with unknown resale opportunities. Rev Econ Stud 86(4):1527–1555
Che Y-K, Hahm DW, Kim J, Kim S-J, Tercieux O (2022) Prestige seeking in college application and major choice. Discussion paper
Dubey P, Geanakoplos J (2010) Grading exams: 100,99,98,... or A, B, C? Games Econom Behav 69(1):72–94
Dworczak P (2020) Mechanism design with aftermarkets: cutoff mechanisms. Econometrica 88(6):2629–2661
Epple D, Romano R (1998) Competition between private and public schools, vouchers, and peer-group effects. Am Econ Rev 88(1):33–62
Epple D, Romano RE (2011) Chapter 20—Peer effects in education: a survey of the theory and evidence, vol 1 of handbook of social economics. North-Holland, pp 1053–1163
Figueroa N, Skreta V (2009) The role of optimal threats in auction design. J Econ Theory 144(2):884–897
Figueroa N, Skreta V (2011) Optimal allocation mechanisms with single-dimensional private information. Rev Econ Des 15(3):213–243
Goldlücke S, Tröger T (2020) The multiple-volunteers principle. Discussion paper
Haile PA (2003) Auctions with private uncertainty and resale opportunities. J Econ Theory 108(1):72–110
Jehiel P, Moldovanu B (2001) Efficient design with interdependent valuations. Econometrica 69(5):1237–1259
Jehiel P, Moldovanu B, Stacchetti E (1996) How (not) to sell nuclear weapons. Am Econ Rev 86(4):814–829
Jehiel P, Moldovanu B, Stacchetti E (1999) Multidimensional mechanism design for auctions with externalities. J Econ Theory 85(2):258–293
Kushnir A, Zubrickas R (2019) Optimal income taxation with endogenous prices. Discussion paper
Moldovanu B, Sela A, Shi X (2007) Contests for status. J Polit Econ 115:338–363
Myerson R (1981) Optimal auction design. Math Oper Res 6(1):58–73
Ostrovsky M, Schwarz M (2010) Information disclosure and unraveling in matching markets. Am Econ J: Microecon 2(2):34–63
Paloyo AR (2020) Chapter 21—Peer effects in education: recent empirical evidence. In: Bradley S, Green C (eds) The economics of education, 2nd edn. Academic Press, London, pp 291–305
Pycia M, Yenmez MB (2022) Matching with externalities. Rev Econ Stud 90:948–974
Rayo L (2013) Monopolistic signal provision. B.E. J Theor Econ 13(1):1–32
Roth AE, Sotomayor MAO (1990) Two-sided matching: a study in game-theoretic modeling and analysis. Econometric society monographs. Cambridge University Press, Cambridge
Rothschild C, Scheuer F (2013) Redistributive taxation in the Roy model. Q J Econ 128(2):623–668
Rothschild C, Scheuer F (2016) Optimal taxation with rent-seeking. Rev Econ Stud 83(3):1225–1262
Sarkisian R, Yamashita T (2022) Mechanism design with expectation-based allocation externality. Discussion paper
Sasaki H, Toda M (1996) Two-sided matching problems with externalities. J Econ Theory 70(1):93–108
Winkler G (1988) Extreme points of moment sets. Math Oper Res 13(4):581–587
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.
We are grateful to Vasiliki Skreta, Gabriel Carroll, Philippe Jehiel, Andy Skrzypacz, Bart Harstad, Pawel Gola, Konrad Mierendorff, Fuhito Kojima, and workshop and seminar participants at UCL, Stonybrook and Oslo, for earlier versions of the paper. We are also grateful to David Francisco Morales Ruiz and Seyed Kourosh Khounsari for their research assistance. Takuro acknowledges funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (Grant Agreement No 714693), ANR under Grant ANR-17-EURE-0010 (Investissements d’Avenir program), and JSPS KAKENHI (JP23K01311).
Proofs
Proofs
1.1 Proof of Proposition 1
Formally, this is a corollary of Theorem 2, and hence the proof is omitted.
1.2 Proof of Proposition 2
Let \(\underline{q},\overline{q}:[0,1]\rightarrow [0,1]\) be such that \(\underline{q}(\theta )\equiv \frac{1}{2}\) and \(\overline{q}(\theta )=1_{\left\{ \theta >\frac{1}{2}\right\} }\). That is, \(\underline{q}(\cdot )\) corresponds to the full pooling mechanism, and \(\overline{q}(\cdot )\) corresponds to the maximum segregation mechanism. Note that \(x_1=\frac{1}{2}\) given \(\underline{q}(\cdot )\), and \(x_1=\frac{3}{4}\) given \(\overline{q}(\cdot )\).
Given any \(x_1^*\in (\frac{1}{2},\frac{3}{4})\) fixed, let \(q(\cdot )\) be such that:
where \(\omega \) satisfies \(x_1=\frac{1}{2}\omega +\frac{3}{4}(1-\omega )\) (hence \(\omega \in (0,1)\)). It is readily checked that \(q(\cdot )\) is non-decreasing (and hence satisfies the incentive compatibility with an appropriate \(p(\cdot )\) function), and \(x_1\) given the assignment \(q(\cdot )\) is \(x_1^*\).
1.3 Proof of Theorem 1
Recall the original problem:
Given any \(x_1\in [\frac{1}{2},\frac{3}{4}]\) fixed, consider the restricted problem where maximization is only within the class of \(q(\cdot )\) that attains \(x_1\) as the peer-effect index. Observe that the objective is:
where \((\text {Const})\) is a term which does not depend on \(q(\cdot )\) (though it depends on \(x_1\); thus, this term can be ignored (only) in the subproblem). Therefore, the subproblem is essentially to maximize \(\int _0^1 q(\theta )g(\theta )d\theta \) among all \(q(\cdot )\) that attains \(x_1\).
Let \(q^*(\cdot )\) be given by:
where y is uniquely determined in order to satisfy \(x_1=2(\frac{2y-1}{2y}\int _0^y\theta d\theta +\int _y^1\theta d\theta )\).
Take any non-decreasing \(q(\cdot )\) which attains \(x_1\) as its peer-effect index. It suffices to show that the distribution of the students’ types in School 1 given \(q^*(\cdot )\) is a mean-preserving spread of that given \(q(\cdot )\), because then, convexity of g implies that \(\int _0^1 q^*(\theta )g(\theta )d\theta \ge \int _0^1 q(\theta )g(\theta )d\theta \).
Let \(\Pi ^*(\theta )\) be the cdf of the students’ types in School 1 given \(q^*(\cdot )\):
and similarly, let \(\Pi (\theta )\) be the one given \(q(\cdot )\):
Applying the definition of mean-preserving spread (Rothschild and Stiglitz, Machina and Pratt), it suffices to show:
for all \(y\in [0,1]\), where the equality hods with \(y=1\). First, if \(y=1\), then:
which is indeed the case, because both \(q^*(\cdot )\) and \(q(\cdot )\) attain the same peer-effect index.
It suffices to show that there exists \(z\in [0,1]\) such that (i) \(\Pi ^*(\theta )-\Pi (\theta )\ge 0\) for all \(\theta <z\), which implies:
and (ii) \(\Pi ^*(\theta )-\Pi (\theta )\le 0\) for all \(\theta >z\), which implies:
For the sake of contradiction, suppose that there exist \(z'<z''\) such that \(\Pi ^*(z')< \Pi (z')\) and \(\Pi ^*(z'')> \Pi (z'')\). First, that \(\Pi ^*(z'')> \Pi (z'')\) implies \(z''\le y\), because for any \(\theta >y\), we have:
Because \(\Pi ^*(\theta )\) is linear for \(\theta <y\), we obtain:
where \(\Pi ^*(0)=\Pi (0)=0\). This means that \(\Pi \) is not globally convex, which is a contradiction, because \(\Pi \) is an integral of a non-decreasing function (which must be globally convex).
Now, given that \(q^*(\cdot )\) is optimal in the subproblem for each given level of \(x_1\), the optimal choice of y must clearly satisfy the condition in the statement, and hence we omit this part of the proof.
1.4 Proof of Theorem 2
The case where \(y^*=1\) (i.e., \(q^*(\cdot )\) exhibits full pooling) is trivial, and thus, we focus on the case where \(y^*<1\).
In what follows, first, consider the case where \(y^*\) is interior (i.e., \(y^*\in (\frac{1}{2},1)\)). The first-order condition with respect to y implies:
at \(y=y^*\), where:
and \(x_1(y)=2(\int _0^y\frac{2y-1}{2y}\theta d\theta +\int _y^1\theta d\theta )\). Note that \(x_1'(y)\le 0\).
Let \(\tilde{f}\) be the same as f except that h is replaced by \(\tilde{h}\):
It suffices to show:
at \(y=y^*\), because this implies that there is some \(\hat{y}\le y^*\) that solves the maximization problem with \(\tilde{h}\), as desired.
We have:
For the first term:
where
by the mean-value theorem and convexity of \(\psi \). Noticing that \(\frac{\partial }{\partial y} \Big [ \frac{2y^*-1}{2y^*}\int _0^{y^*} g(\theta )d\theta +\int _{y^*}^1\,g(\theta )d\theta \Big ]\le 0\), we conclude the first term is non-positive.
For the second term:
where
because of convexity of \(\psi \). Because \(x_1'(y^*)\le 0\) and \(\frac{2y^*-1}{2y^*}\int _0^{y^*} g(\theta )d\theta +\int _{y^*}^1\,g(\theta )d\theta \ge 0\), we conclude that the second term is non-positive.
1.5 Proof of Theorem 3
Let \(\hat{q}(\cdot )\) be as follows:
where \(y_1,y_2\) are chosen so that \(y_1< y_2\) and:
Note that \(y_1,y_2\in (0,1)\) exist as long as \(x_1^*<\frac{3}{4}\) (i.e., \(q^*(\cdot )\) is not maximally segregating).
The planner’s payoff difference between under \(\hat{q}(\cdot )\) and under \(q^*(\cdot )\) is:
Because \(g''>0\) and \(h(x_1)\ge h(1-x_1)\), it suffices to show that \(\hat{q}(\cdot )\) is a mean-preserving spread of \(q^*(\cdot )\).
Let \(\Pi ^*(\theta )\) be the cdf of the students’ types in School 1 given \(q^*(\cdot )\):
and similarly, let \(\hat{\Pi }(\theta )\) be the one given \(\hat{q}(\cdot )\).
Applying the definition of a mean-preserving spread, it suffices to show:
for all \(y\in [0,1]\), where the equality hods with \(y=1\). First, if \(y=1\), then:
which is indeed the case, because both \(\hat{q}(\cdot )\) and \(q^*(\cdot )\) attain the same peer-effect index.
For any \(y\le y_1\), we have:
because for \(\theta \le y_1\), we have \(\hat{\Pi }(\theta )-\Pi ^*(\theta )=\int _0^\theta (1-q^*(z))dz>0\).
For any \(y\ge y_2\), we have:
because for \(\theta \ge y_2\), we have \(\hat{\Pi }(\theta )=1-2\int _\theta ^1 1dz\) and \(\Pi ^*(\theta )=1-2\int _\theta ^1 q^*(z)dz\ge 1-2\int _\theta ^1 1dz\).
For any \(y\in (y_1,y_2)\), in order to show that
it suffices to show that there exists \(z\in [y_1,y_2]\) such that \(\int _0^y (\hat{\Pi }(\theta )-\Pi ^*(\theta ))d\theta \) is non-decreasing in y for \(y<z\) and non-increasing in y for \(y>z\) (recall that the expression is non-negative for \(y=y_1\) and \(y=y_2\)).Footnote 15
Indeed:
where, for \(y\in (y_1,y_2)\), \(\hat{\Pi }(y)\) is constant and \(\Pi ^*(y)\) is non-decreasing. Therefore, there exists \(z\in [y_1,y_2]\) such that \(\int _0^y (\hat{\Pi }(\theta )-\Pi ^*(\theta ))d\theta \) is non-decreasing in y for \(y<z\) and non-increasing in y for \(y>z\).
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Sarkisian, R., Yamashita, T. Optimal student allocation with peer effects. Rev Econ Design (2024). https://doi.org/10.1007/s10058-024-00349-x
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10058-024-00349-x