Show Summary Details

Page of

Printed from Oxford Research Encyclopedias, Economics and Finance. Under the terms of the licence agreement, an individual user may print out a single article for personal use (for details see Privacy Policy and Legal Notice).

date: 08 December 2021

School Choice and Accountabilityfree

School Choice and Accountabilityfree

  • Simon BurgessSimon BurgessCentre for Market and Public Organisation, University of Bristol; IZA
  •  and Ellen GreavesEllen GreavesSchool of Economics, University of Bristol; Institute for Fiscal Studies

Summary

School choice and accountability are both mechanisms initially designed to improve standards of education in publicly provided schools, although they have been introduced worldwide with alternative motivations such as to promote equality of access to “good” schools. Economists were active in the initial design of school choice and accountability systems, and continue to advise and provide evidence to school authorities to improve the functioning of the “quasi-market.”

School choice, defined broadly, is any system in which parents’ preferences over schools are an input to their child’s allocation to school. Milton Friedman initially hypothesized that school choice would increase the diversity of education providers and improve schools’ productivity through competition. As in the healthcare sector and other public services, “quasi-markets” can respond to choice and competition by improving standards to attract consumers. Theoretical and empirical work have interrogated this prediction and provided conditions for this prediction to hold. Another reason is to promote equality of access to “good” schools and therefore improve social mobility. Rather than school places being rationed through market forces in the form of higher house prices, for example, school choice can promote equality of access to popular schools. Research has typically considered the role of school choice in increasing segregation between different groups of pupils, however, due to differences in parents’ preferences for school attributes and, in some cases, the complexity of the system.

School accountability is defined as the public provision of school-performance information, on a regular basis, in the same format, and using independent metrics. Accountability has two functions: providing incentives for schools, and information for parents and central authorities. School choice and accountability are linked, in that accountability provides information to parents making school choices, and school choice multiplies the incentive effect of public accountability. Research has studied the effect of school accountability on pupils’ attainment and the implications for teachers as an intermediate mechanism.

Subjects

  • Health, Education, and Welfare Economics
  • Labor and Demographic Economics

A Framework for Studying School Choice and Accountability

Education is an important determinant of later outcomes, for example employment, health, and well-being. Public provision is therefore common, due to positive externalities and economies of scale. The quality of public education is consequently of central importance to governments spending public funds. School choice and accountability are two mechanisms commonly used to improve school effectiveness, where an “effective” school is one that generates causal improvements in student outcomes (Rothstein, 2006).

A defining feature of a school system is the way in which pupils are assigned to particular schools. While typically there are enough public school places overall to meet demand, schools are not equally attractive to parents or equally effective, and so how pupils are assigned to schools matters for education outcomes. School choice is one such assignment mechanism. Defined broadly, school choice is any system in which parents’ preferences over schools are an input to their child’s allocation to school.1 “School choice” is interpreted and implemented differently across countries, states, and districts. Friedman (1955) initially conceived of the “denationalization” of education to broaden the availability of choice for parents. By providing a subsidy to parents to attend any private school (now coined a voucher) they could “express their views about schools directly,” rather than through residential location or political channels. Friedman was initially skeptical that government could provide greater freedom to choose between public schools. Since 1955, however, the scope and interpretation of school choice has expanded dramatically. Examples include parents submitting a rank-ordered list of school choices to a central authority, households receiving a voucher to attend a private school, and households being able to opt out of a “default” or “neighborhood” school. The right for parents to express a preference for their child’s school was established in England in 1988, while in other countries, coordination of school applications is in its infancy, or yet to emerge.2

How does school choice induce improvements in school effectiveness? Overall, schools must be incentivized to compete to attract pupils/parents through raising their quality. From the demand side, the conditions for this to hold are that: parents care about school effectiveness, they are able to identify the most effective schools, and they are able to choose the most effective schools. School accountability feeds into the second of these conditions, providing information for parents to select their school(s) of preference. Why would parents not be able to choose their most preferred or most effective schools? A large body of research studies the importance of the design of the assignment mechanism that allocates pupils to schools according to their choices and schools’ capacity constraints and admission priorities. Different assignment mechanisms have different properties, such as whether they induce parents to make truthful school choices. School admission priorities, such as proximity, can also limit the likelihood of gaining admission for some households, and therefore the likelihood that they make that school choice.

On the supply side, there must be spare capacity in the system so that changes in school choices lead to changes in school allocations (and therefore funding for schools). In the absence of funding incentives, there must be reputation, career, or altruistic concerns for school leaders to drive school improvements. Schools must also have autonomy in school management/organization to deliver improvements. An initial rationale for school choice was that schools would specialize and diversify to meet diverse parent preferences (Friedman, 1955), although this has not been the focus of empirical research. Finally, it must be rational for schools to attract pupils through improvements in quality rather than alternative strategies such as “cream-skimming”—selecting pupils to improve school accountability measures—or marketing strategies. More generally, Hatfield et al. (2016) articulate that it must be in a school’s interest to attract more pupils, specifically that the “quality” of its students does not worsen if the number of school choices increases.

School choice can also affect the composition of schools. Active diversity programs within a system of school choice can lead to more integrated public schools, but the majority of research has concentrated on the potential for school choice to lead to more segregated schools. The primary mechanism is through parents’ heterogeneous preferences. For example, in a coarse, stylized example, segregation increases if all higher-educated parents choose a highly performing school while all lower-education parents choose a less academic school.

School accountability feeds into school improvement not only through providing information to parents (as discussed in relation to school choice above) but also through providing information to schools themselves and school authorities. Alongside information, school accountability provides incentives. In a principal–agent setting, and in the absence of other strong incentives, the public provision of performance information can exert pressure on school leaders of low-performing schools. Sanctions or rewards can be explicit, such as the replacement of school leaders, or implicit, such as raising applications to the school.

This review summarizes theoretical and empirical research organized around these themes. The material is set out as follows. First, the “rules of the game” given by the school place allocation mechanism. Second, the components of a school choice process: the preferences and choices of parents (observed through residential demand and direct school choices) and how these are affected by information provision. Third, the incentives for schools and teachers provided by the accountability system, and the implications for pupil attainment. After providing evidence on each of these research areas, the overall effect of school choice on outcomes of interest, principally market-level attainment and sorting, is examined. Before concluding, the evidence that schools have a causal effect on pupil attainment and wider indicators of well-being is assessed.

Allocation Mechanisms

Where the school choice system invites parents to express a preference for their child’s school, it is common that a central authority allocates pupils to schools. Typically, parents submit a rank-ordered list (ROL) of schools, which the central authority uses, in combination with schools’ priorities, to assign pupils to schools. Central coordination itself is important, increasing the number of applications and students’ welfare from the school match (Abdulkadiroğlu, Pathak, & Roth, 2005; Abdulkadiroğlu et al., 2017). This section will briefly review the choice of assignment mechanism, or algorithm, used to assign pupils to schools, which is relevant for the rest of this review. See Abdulkadiroğlu and Sönmez (2013), and Pathak (2017) for more detailed reviews of theoretical and practical considerations in market design.

Economists have been active in theory and practice to shape the allocation mechanisms that assign students to schools (Abdulkadiroğlu, Pathak, & Roth, 2005; Abdulkadiroğlu, Pathak, Roth, & Sönmez, 2005; Abdulkadiroğlu et al., 2009; Pathak, 2017). The choice of allocation mechanism (or algorithm) is important, as it affects whether parents choose their most preferred schools and whether the eventual outcome is Pareto efficient (Chen & Sönmez, 2006). In their seminal paper, Abdulkadiroğlu and Sönmez (2003) formalize three criteria to evaluate school choice algorithms. First, ex post efficiency: This is a measure of how the assignment respects preferences. Second, absence of justified envy: No student should be admitted to a school when another student (who prefers the school) has priority and is not accepted. Third, strategy-proofness: There is no better strategy for parents than to list their most preferred schools. This final property maps to the framework discussed in the previous section, because to provide incentives for school improvement parents should nominate their most preferred school(s). Strategy-proofness is also important, as empirical work from around the world has shown that parents make strategic errors if truth-telling is not a weakly dominant strategy under immediate acceptance (“Boston”) mechanisms (e.g., Ajayi, 2013, in Ghana; Lai et al., 2009, in Beijing; Lucas & Mbiti, 2012, in Kenya) and a significant proportion of parents act strategically (Agarwal & Somaini, 2018, in Cambridge, the United States; Calsamiglia et al., 2020, in Barcelona; Gortázar et al., 2020, in Madrid; He, 2017, in Beijing; Kapor et al., 2020, in Connecticut; Pathak & Sönmez, 2013, in Chicago) and have a vested interest in the algorithm’s continuation (Pathak & Sönmez, 2008). This creates an unequal playing field, as more advantaged households are more likely to understand the strategic incentives. Additional advantages of strategy-proofness are that parents’ true preferences are respected, and that it is therefore easier for central authorities to advise parents and monitor supply and demand for schools. Revealing true preferences is a weakly dominant strategy in two common mechanisms, Student Proposing Deferred Acceptance (SPDA; Gale & Shapley, 1962; also called Student Optimal Stable Matching) and Top Trading Cycles (TTC).

Refinements show that truth-telling is not optimal in some circumstances even with an SPDA mechanism, for example when parents can make only limited nominations (Calsamiglia et al., 2010; Haeringer & Klijn, 2009). Fack et al. (2019) overcome this limit to truth-telling by formalizing the concept of “stability”: Households’ school choices will reflect true preferences from the set of schools that have a positive probability of admission, “skipping the impossible.” There is also evidence that individuals make mistakes in their ROL even under a truth-telling allocation mechanism, which is summarized by Hassidim et al. (2017). For example, for higher education track/funding choice in Israel, Hassidim et al. (2016) find that around 20% of ROLs obviously misrepresent true preferences, by omitting options that are attached to funding over identical options without funding, or reversing the ranking of these options. Other examples cited within Hassidim et al. (2017) are consistent with “skipping the impossible,” although there may be other strategy at play, too.

There are trade-offs between manipulable and non-manipulable allocation mechanisms used in practice, however, in that manipulable mechanisms allow parents to express the strength of their preferences (Abdulkadiroğlu et al., 2015).3 Kapor et al. (2020) state that the optimal assignment mechanism depends on whether “applicants’ ability to express cardinal preferences through strategic play in the Boston mechanism outweighs the welfare costs of strategic mistakes due to misunderstandings about the mechanism or lack of information about demand conditions.” In this context, Kapor et al. (2020) find that imperfect information leads higher welfare under deferred acceptance than immediate acceptance algorithms.

Hatfield et al. (2016) are the first to consider the implications of the allocation mechanism for schools’ incentives to improve performance. The key insight is that schools must not be penalized for becoming more popular by attracting a “worse” set of students. Hatfield et al. (2016) conclude that no allocation mechanism commonly used (SPDA, Boston, or TTC) respects this condition. In larger markets, however, SPDA approximately respects this condition, while “Boston and the TTC mechanisms provide incentives for schools to make themselves less attractive to “less desirable” students.”

Overall, theoretical and empirical work point to the importance of centralizing school admissions to reduce inefficiency, and choosing an assignment algorithm that is transparent and easy to explain in addition to desirable formal properties. In his review of implementations of school choice in practice, Pathak (2017) argues that “what really matters for school choice market design are basic insights about straight forward incentives, transparency, avoiding inefficiency through coordination of offers and well-functioning aftermarkets, and influencing inputs to the design, including applicant decision-making and the quality of schools.”

The following sections (“Parents’ Preferences for Schools” and “Evidence From School Choices”) review research that infers the characteristics of schools that parents value from their school choices. The allocation mechanisms described here imply that parents’ school choices do not always reflect their true preferences for schools, which researchers must address in this empirical work.

Parents’ Preferences for Schools

Parents’ preferences for schools, particularly whether they value school effectiveness, is a crucial determinant of whether school choice has the potential to improve educational standards and/or increase segregation. Preferences for other school attributes are also important, however. For example, Barseghyan et al. (2019) show that schools’ incentives to improve productivity are weakened or eliminated if households have strong preferences for the peer group in the school. This section summarizes the literature for what we can infer about parents’ preferences for school attributes through their “revealed preference.” First, from residential demand—increased demand around the “best” schools. Second, from school choices. Finally, this section summarizes evidence on how the provision of information affects school choices, which is an important policy consideration.

Evidence From Residential Demand

There is a large empirical literature, beginning with Black (1999), on the willingness to pay for local school quality. Using boundary discontinuities, Black finds that house prices respond to local school quality. Typically, households are willing to pay a premium of around 3–4% for access to a one standard deviation increase in school average test scores. (For detailed summaries, see Black & Machin, 2011; Gibbons & Machin, 2008; Nguyen-Hoang & Yinger, 2011) Bayer et al. (2007) disentangle household preferences for school and neighborhood attributes, while taking into account endogenous sorting across neighborhoods. They show that previous estimates of willingness to pay for higher school quality are upward biased if households also have preferences for more affluent neighbors. Bibler and Billings (2020) find evidence of a higher premium for households with a strong preference for school quality and time constraints.

Exploiting reforms, Machin and Salvanes (2016) use a 1997 reform, which abolished neighborhood schooling in Norway to estimate the change in willingness to pay for houses close to above-average schools, finding a fall in the house price premium by 50%. Ries and Somerville (2010) use a rezoning reform in Vancouver in 2001 and find that house prices respond to changes in school quality only in the top quartile of the distribution.

A group of papers study the effect of school accountability classifications of schools on local prices. Figlio and Lucas (2004) identify the effect of “school grades” in Florida, over the effect of publicly available test score data. Using property and neighborhood fixed-effects, they find that new “school grades” were capitalized into house prices, but the effect was damped once the variability of grading over time was observed by households. Fiva and Kirkebøen (2011) find consistent evidence using the same identification strategy in Oslo, Norway, where a measure of “intrinsic school quality” was released in 2005. Hussain (2020) finds consistent evidence in response to established independent “Ofsted” inspections in England. Like Ries and Somerville (2010), Hussain (2020) also finds that the demand-side response increases with school quality: Prices around lower quality schools (as proxied by test performance and social composition) are not affected by the revelation of inspection information.

There is evidence that school quality is capitalized into house prices even where school quality information is not disclosed. Harjunen et al. (2018) study Helsinki, Finland, where no standardized test or inspection data are available to parents, but, using the boundary-discontinuity research design common in the literature, the house price premium is similar in magnitude to contexts where school quality is observable. In this context, the socioeconomic composition of the school is found to drive demand, rather than school effectiveness itself.

Evidence from across this literature suggests that households value the peer composition in addition to (or as a proxy for) school quality. Imberman and Lovenheim (2016) explicitly explore the characteristics of schools that parents’ value through residential demand. Following the introduction and publicity of school value-added measures by the Los Angeles Times, using a difference-in-differences framework, Imberman and Lovenheim (2016) find no evidence that value-added measures are capitalized into local house prices. This evidence suggests that parents do not respond to “school effectiveness,” but rather to measures of school quality that conflate peer-group composition and absolute measures of test scores. In contrast, through a carefully executed boundary-discontinuity approach, Gibbons et al. (2013) find that house prices respond by a similar magnitude to published measures value-added and academic attainment (as a proxy for peer group) for primary/elementary schools in England. Gibbons et al. (2013) conclude that

The most plausible explanation that is consistent with our results is that parents value both academic effectiveness and composition aspects of school quality because they are interested in their own child’s academic progress, as well as the social status of their child’s peers.

The different conclusions may be due to the context, for example, established performance indicators in England versus new performance indicators in Los Angeles. The following section will also consider whether parents value school effectiveness over and above peer composition, and also finds a mixed picture.

Evidence From School Choices

There are a number of empirical challenges to estimate preferences for school attributes. Early research use conditional logits (and variants) and rank-order lists (ROLs) of parents’ submitted school choices to estimate parents’ preferences for school attributes. The first challenge is whether ROLs reflect strategic choices rather than true preferences. Second, it is generally impossible to know the pool of schools that parents consider when making their choices. Finally, as admissions to popular schools are rationed, it is difficult to disentangle preferences from constraints.

A ROL may reflect true preferences for school attributes; the “revealed preference argument” is that the researcher can infer preferences by observing the types of schools that parents choose (Agarwal & Somaini, 2020). Papers in this strand of the literature must justify why ROLs are treated as truthful, through the context studied or method adopted. This section first summarizes the literature that assumes truthful reports on ROLs, before discussing subsequent research that allows ROLs to be strategic choices for all or some households.

Consistent findings across this strand of the literature are that parents value school “quality,” typically defined as test scores (Abdulkadiroğlu et al., 2017, 2020; Ajayi & Sidibe, 2020; Bertoni et al., 2020; Beuermann et al., 2018; Borghans et al., 2015; Burgess et al., 2015; Glazerman & Dotter, 2017; Harris & Larsen, 2019; Hastings et al., 2009; Oh & Sohn, 2019; Ruijs & Oosterbeek, 2019; Walker & Weldon, 2020) and value proximity (Abdulkadiroğlu et al., 2020; Bertoni et al., 2020; Beuermann et al., 2018; Borghans et al., 2015; Burgess et al., 2015; Glazerman & Dotter, 2017; Harris & Larsen, 2019; Hastings et al., 2009; Walker & Weldon, 2020). The school composition is also an important factor, in ability composition (Abdulkadiroğlu et al., 2020; Beuermann et al., 2018), social composition (Abdulkadiroğlu et al., 2017; Burgess et al., 2015; Glazerman & Dotter, 2017), and racial composition, with preferences for “own group” (Abdulkadiroğlu et al., 2017; Glazerman & Dotter, 2017; Hastings et al., 2009). There is also evidence from this strand of literature that parents are aware of supply-side constraints. There are strong preferences for the home-school, which gives priority for admission, all else equal (Hastings et al., 2009), indicators for previous oversubscription (Ruijs & Oosterbeek, 2019), and probability of admission (Beuermann et al., 2018; Walker & Weldon, 2020).

Exploring heterogeneity in preferences across parent types, there is consistent evidence that more advantaged households (defined by income or education) have stronger preferences for school “quality” (Borghans et al., 2015; Burgess et al., 2015; Harris & Larsen, 2019; Hastings et al., 2009; Walker & Weldon, 2020) and according to prior test score or academic track (Abdulkadiroğlu et al., 2017; Ajayi & Sidibe, 2020; Beuermann et al., 2018; Hastings et al., 2009; Ruijs & Oosterbeek, 2019). Hastings et al. (2009) and Walker and Weldon (2020) also find stronger preferences for school “quality” for non-White parents. Glazerman and Dotter (2017) provide the only example of variation in preferences for proximity according to race, however.

Harris and Larsen (2019) study school choices in New Orleans after Hurricane Katrina, described as “arguably the most competitive school market ever developed in the United States.” They study preferences for a wide range of school characteristics (extra-curricular activities, wrap-around care [free and paid], and school size), which reduces the possibility of omitted variable bias. Strong preferences for typically unobserved school attributes such as these “can have a substantive influence on the estimated demand and preference parameters.” This study, in contrast to previous literature, finds that parents value school effectiveness.

The earliest and still widely cited paper in this strand of the literature is Hastings et al. (2009), who use school choice data from Charlotte-Mecklenburg, North Carolina, to estimate a mixed-logit demand model for schools. The allocation mechanism (giving priority to first choices within priority groups) may have induced strategic listing, which is a limitation the authors acknowledge and present evidence against. They find that demand at high-performing schools is more responsive to increases in mean test scores than demand at low-performing schools. Their model also implies a “mobile,” more affluent group of families exerting pressure on school performance, and a less mobile, less affluent group essentially going to the local school.

Studying primary school choices in England, Burgess et al. (2015) define a set of schools for each family in the data that they could almost surely access. By comparing households of higher and lower socioeconomic background with the same choice sets, they demonstrate that two-thirds of the observed variation in choices for academic quality are driven by constraints in admission, rather than preferences. Walker and Weldon (2020) explicitly model the admission probability to secondary/high schools in England as a subjective probability, given multiple sources of uncertainty. That is, “preferences” for the probability of admission are estimated in a random utility model alongside preferences for school quality and distance. This approach is also taken by Beuermann et al. (2018). Also for England, Bertoni et al. (2020) find positive preferences for autonomous converter “Academy” schools. Demand is particularly high for academies with high standards that were previously oversubscribed, suggesting that academy status is a signal for quality.

Abdulkadiroğlu et al. (2020) and Beuermann et al. (2018) study whether parents value school attributes such as causal effects on academic and non-academic outcomes, which are estimated from their data. In New York, under a deferred acceptance algorithm, Abdulkadiroğlu et al. (2020) find that preferences are correlated with peer quality and effectiveness. But, as more effective schools enroll higher ability students, the correlation between preferences and effectiveness, conditional on peer quality, is zero.

Beuermann et al. (2018) adopt a similar approach but find different results in Barbados, under a deferred acceptance algorithm. They study whether parents choose schools that have positive causal effects, studying whether educational outputs are multi-dimensional. First, they estimate individual public schools’ causal impacts on a range of outcomes in Trinidad and Tobago: “high-stakes test scores, low-stakes test scores, dropout, teen motherhood, teen arrests, and labor market participation.” Second, these estimated impacts are linked to parents’ school ROLs to explore whether parents choose schools with positive causal effects on these multiple outcomes. In contrast to Abdulkadiroğlu et al. (2020), Beuermann et al. (2018) find that “parents of high-achievers can and do disentangle schools that causally improve test scores from schools with strong average performance.” In addition, parents in this context “have strong preferences for schools that reduce crime and increase labor market participation—impacts that are only weakly correlated with school impacts on tests.” In this context, and including a wide range of school attributes typically unobserved by the researcher, parents value school effectiveness, across a number of dimensions.

Several papers model preferences as strategic responses to the context and assignment mechanism (Agarwal & Somaini, 2018; Calsamiglia et al., 2020; Fack et al., 2019). Fack et al. (2019) study school choice in Paris, France, where the students are assigned to schools after submitting their ROLs (length 8) using the school-proposing deferred acceptance algorithm. Rather than assume truth-telling, Fack et al. (2019) formalize the concept of “stability” or “non-justified envy”: Each student is matched to her/his most preferred school within their specific constraints, “skipping the impossible” by not ranking schools with zero probability of admission. That is, Fack et al. “apply the revealed preferences argument conditional on the set of schools that each student is eligible for” (Agarwal & Somaini, 2020). Calsamiglia et al. (2020), studying Barcelona, Spain, recognize that the allocation mechanism will affect only strategic households: Non-strategic households will always list their most preferred schools. Calsamiglia et al. (2020) estimate both household preferences and the distribution of strategic types in a parametric model “in which strategic agents solve for the optimal report in an immediate acceptance mechanism using backward induction from lower- to higher-ranked choices.” Rather than assuming households choose the school that maximizes utility, Agarwal and Somaini (2018) assume that households maximize the expected utility from a lottery over assignments to different schools, where “the lottery implied by a rank-order list consists of the probabilities of getting assigned to each of the schools on that list.” In turn, these probabilities depend on the student’s priority type, listed schools and randomly generated tie-breaker, and competition from other students (their priority type and listed schools). In relation to the previous body of research, Agarwal and Somaini (2018) note that “Our empirical results indicate that treating preferences as truthful is likely to result in biased estimates in markets where students face stiff competition for their preferred schools.”

The results from these papers are generally in line with the first strand of papers that relied on truthful reports or modeled aspects of the strategic nature of choices, such as the probability of admission or oversubscription in the previous year. Parents care about school “quality,” again proxied by test scores (Agarwal & Somaini, 2018; Akyol & Krishna, 2017; Calsamiglia et al., 2020; Fack et al., 2019). Again, this is increasing with parents’ characteristics that are correlated with advantage: increasing with prior test scores in Paris (Fack et al., 2019), with parents’ education in Barcelona (Calsamiglia et al., 2020), and “competitiveness” is valued more by paid-lunch students in Cambridge, the United States (Agarwal & Somaini, 2018). In Barcelona, higher-education parents are also less price sensitive (Calsamiglia et al., 2020). Parents value proximity, modeled as convex with discontinuous jumps at 500 m and 1 km in Barcelona (Calsamiglia et al., 2020) and the peer group (Fack et al., 2019). Additional findings are the importance of siblings (Agarwal & Somaini, 2018; Calsamiglia et al., 2020); language immersion programs, which are valued more by Spanish and Portuguese speakers (Agarwal & Somaini, 2018); and larger schools/more resources (Calsamiglia et al., 2020). In addition, these papers unanimously find that a subset of parents respond to the strategic incentives in their school choice environment (Agarwal & Somaini, 2018; Akyol & Krishna, 2017; Calsamiglia et al., 2020; Fack et al., 2019).

Overall, this strand of the literature is consistent in that distance to school and some measure of academic quality are important factors for parents. Whether this finding implies that school choice provides incentives for schools to improve educational standards depends on whether parents value raw levels of academic attainment or value-added/school-effectiveness. Two papers that include a broad range of school characteristics (Harris & Larsen, 2019, and Beuermann et al., 2018) find that parents do value school effectiveness, suggesting some promise. Parents also value the school composition, however, which may shift schools’ incentives to compete on this dimension rather than educational standards, if permitted (Abdulkadiroğlu et al., 2020). Heterogeneity in parents’ preferences is evident, which may reflect parents’ preferences or constraints in access or information. The following section reviews evidence for whether information provision affects school choices.

The Effect of School Quality Information Provision on School Choices

The previous section described estimated preferences for school attributes, with a common finding that more advantaged households typically value school “quality” more. Hastings and Weinstein (2008) note that

If lower-income families face higher costs of gathering and interpreting statistics on academic achievement, they may choose schools based on easier-to-determine characteristics such as proximity, instead of school test scores.

A small but growing number of papers explore the effect of information provision on school choices, through experiments at the household, school or market-level. This section will summarize the evidence that information provision can alleviate imperfect information and influence parents’ school choices, informing whether heterogeneity in school choices across households is due to differences in preferences or access to information. Neilson et al. (2019) state that “Taken together, the empirical evidence to date indicates that information interventions do have the potential to change behavior but that policy details can matter quite a lot.” Relevant policy details include the complexity of the choice environment and properties of the allocation mechanism (truth-revealing or otherwise), which were described in the “Allocation Mechanisms” section.

Ajayi et al. (2020) present results from a student-level randomized controlled trial in Ghana, which provided comprehensive and accessible information about (the complex) secondary school choice to the treatment groups. The information provided was internalized, and changed students’ application decisions. The intervention did not improve students’ overall outcomes in the transition to secondary school, however, which the authors conclude is likely to be due to constraints such as expense and/or distance to preferred schools. In an earlier paper also in Ghana, Ajayi et al. (2017) find that information can increase the involvement of guardians in the senior high school decision process, through a school-level randomized controlled trial.

In the first study in this literature, using a mix of field and natural experiments (induced by “No Child Left Behind” accountability measures), Hastings and Weinstein (2008) find that the provision of additional information on school characteristics shifts school choices towards non-guaranteed and higher-attaining schools, particularly where a high-attaining school is in close proximity.

Neilson et al. (2019) show that an experimental intervention providing personalized information on elementary schools in Chile changes families’ choices towards more ambitious schools. They show that this affects outcomes, too—student scores are 0.2 SD higher five years later. They also set out a structural model to assess the systemic response to national rollout of the intervention; they note that capacity constraints are important, but show that typically supply-side responses are positive for school effectiveness. In contrast, in non-experimental evidence from Chile, where schools are identified as school-effectiveness “winners” from within a homogeneous group of schools (approximating a value-added measure), Mizala and Urquiola (2013) find that this signal does not significantly affect schools’ market outcomes.

Corcoran et al. (2018) run an intervention in New York City. Focused on more disadvantaged neighborhoods, the treatment is provision of a simple, short list of nearby schools with above-average performance. The findings were positive, leading to more students in higher performing schools; this came about more by the students avoiding low graduation rate schools than applying to high-rate schools.

In the only market-level experiment in Pakistan, Andrabi et al. (2017) show that the introduction of report cards improved parents’ knowledge of local schools’ quality. This leads to improved enrolment and learning—average test scores increasing by 0.11 standard deviations. Private schools responded by reducing fees, and lower quality private schools were more likely to exit the market.

School choice is increasingly researched and conducted online. Lovenheim and Walsh (2018) find that online search behavior is influenced by the school choice environment, suggesting that parents seek out information under certain incentives. Theoretical work explores endogenous information provision under alternative school choice environments (Bade, 2015; Harless & Manjunath, 2018). Glazerman (2017) reviews the “choice architecture” of “school shopping websites” of 14 U.S. school districts, finding substantial variation, for example, in how schools are sorted, whether a map of schools is given, what school attributes are included, and how data are presented. Glazerman et al. (2020) explore whether such variation in design features affects understanding, satisfaction, and ease of use for low-income households, in a low-stakes experiment with hypothetical schools. They conclude that design decisions can “affect how people interpret that information and evaluate schools relative to one another.” This is in contrast to earlier (off-line) information provision, where there is no marginal gain from simplified and ordered ranking of schools (Hastings & Weinstein, 2008). Through an online experiment with a nationally representative U.S. sample, Valant and Newark (2020) find that “perceptions of school quality are heavily influenced by parent comments even when these comments appear alongside official ratings.” This research area appears to be moving quickly, following the technological developments in how school choice information is provided to parents.

Overall, evidence suggests that the provision of relevant information to parents induces changes in school choices. Most of this research is from experimental settings, which may have limited external validity. The context is also changing rapidly with more information available online. The following section will consider the wider effects of school accountability, typically studied at scale.

School Accountability

We define school accountability as the public provision of school-performance information, on a regular basis, in the same format, and using independent metrics. Typically, the school-performance data will include measures of pupil achievement, which ideally derive from a common, comparable, externally marked exit exam. This has two intended functions: incentives and information. In a principal–agent setting, and in the absence of other strong incentives, the public provision of performance information can exert pressure on school leaders of low-performing schools. Sanctions or rewards can be explicit, such as the replacement of school leaders, or implicit, such as falling applications to the school. Second, accountability provides information for school improvement, perhaps for school authorities to act on, and certainly for parents to consider in their school choice.

The major focus for researchers has been the implications of school accountability for pupil achievement. Of course, this achievement comes via teachers, so there has also been substantial investigation of teachers’ reactions to accountability. Finally, as with every performance management system, there is inevitably scope for gaming (and cheating) the metrics and this has also been the subject of study. While almost all the research has focused on high/secondary schools, Bassok et al. (2019) note that in the 2010s, most U.S. states have ratings of childcare settings. They show similar effects: Lower-rated childcare programs see enrolment fall in response to the introduction of accountability, and respond with relative improvements.

What Effect Does the Accountability System Have on Pupil Performance?

Researchers face two main difficulties in trying to establish the causal impact of accountability systems on pupil performance. First, finding an adequate control group for the counterfactual is difficult, as typically a performance management reform covers a whole area. Second, it is difficult to evaluate individual components of accountability systems, as typically a multifaceted reform is introduced all at once (Figlio & Ladd, 2008).

These problems are well illustrated by the case of the first national school accountability framework in the United States, the No Child Left Behind (NCLB) Act, which came into force nationwide in 2002. This mandated annual testing of primary school pupils in math and reading, and required the state to measure and publicize the proficiency scores of the pupils for each school. Measures were published both for sociodemographic subgroups as well as averaging over all pupils in the school, and Dee (2020) notes that this strong focus on subgroups “was one of its most universally admired features.”

Studies of NCLB comprise the majority of the published research on school accountability, though all such papers have identification challenges. Another important issue, particularly with small subgroups, is the importance of measurement error and the scope for simple random variation to influence school accountability ratings (Kane & Staiger, 2002). The early evidence is usefully summarized in Figlio and Loeb (2011), and they note that all papers have to manage “the difficulty of isolating the effect of NCLB from other concurrent changes.” Dee and Jacob (2009) use the federal introduction of NCLB and compare states that had implemented a system of school accountability before NCLB. They found that NCLB had no impact on reading scores and a 0.15 pupil-level standard deviation impact on math scores. Hanushek and Raymond (2005) use state-level accountability, pre-NCLB, and adopt a state-level fixed-effects model, and find a positive effect of around 0.2 of a (state-level) standard deviation on test scores. Other studies exploit discontinuities in school accountability ratings and adopt a regression discontinuity approach. They show that schools receiving low ratings subsequently showed positive conditional impacts on pupil achievement gains, with strong and substantial effects (Chiang, 2009; Figlio & Rouse, 2006; Rockoff & Turner, 2010; Rouse et al., 2013).

Figlio and Loeb (2011) conclude that “taken as a whole, the body of research on implemented programs suggests that school accountability improves average student performance in affected schools.” A retrospective of the lessons from the past 30 years of school accountability in the United States concludes similarly that NCLB was effective in changing performance (Dee, 2020). Dee (2020) is less optimistic about its successor, Every Student Succeeds Act (ESSA), with its greater flexibility and diversity diluting the single-minded focus on student performance.

Other studies in other countries have cleaner identification strategies. In England and Wales, the school accountability program was established in 1992. Burgess et al. (2013) exploit a policy experiment that changed school accountability in Wales but not in England. Using a difference-in-differences analysis, Burgess et al. (2013) find significant and robust evidence that this reform markedly reduced school effectiveness in Wales. The impact is sizeable, 0.23 of a (school-level) standard deviation, equivalent to 0.09 of a pupil-level standard deviation. In this study, the significant heterogeneity shows a much stronger effect on attainment of low-achieving pupils.

Two other studies have evaluated the introduction of school accountability. Nunes et al. (2015) show that the publication of school rankings significantly affects parents’ school choices and eventual enrolment in Portugal. For the Netherlands, Koning and Van der Wiel (2012) show that the lowest ranked schools raised performance substantially (up to 12% of a standard deviation of school average test scores) after the publication of school quality scores.

There is consensus in this literature that accountability measures raise student performance, and typically more for low-performing pupils. There are fewer studies showing how this is achieved. The following section will focus on teachers’ career choices as one mechanism. In the United States, teachers are also found to change their teaching practices, for example spending more of the school day on instruction (Rouse et al., 2013) and work harder but also narrow the curriculum (Reback et al., 2014).

In addition, Craig et al. (2015) show that the ratings are reinforced by school district administrators, who reward high-performing schools with more funds. Standardized testing is also important: Bergbauer et al. (2018) use data from PISA across six waves and 59 countries, and find a positive impact of standardized testing on student outcomes, the effect being larger in low-performing countries.

There have been few studies of the long-run consequences of accountability. One valuable exception is a study of Texas public schools, under accountability from the 1990s (Deming et al., 2016). Results show different effects of accountability according to school quality: There are long-run positive effects from accountability systems focused on schools at risk of failing a low threshold, but accountability relative to a high threshold target showed no student gains, and may in fact have caused harm.

Effect on Teacher Career Choices

The introduction of test-based accountability would generally imply changes in teachers’ working conditions. For example, life as a teacher in schools newly publicly recognized as high-performing might be more rewarding, while in schools seen as low-performing, teacher performance may be much more closely inspected. This might potentially affect decisions on where to work, on teacher decisions to quit or be replaced, on whether to join the profession at all, and on teacher sorting between schools.

Clotfelter et al. (2004) show that the introduction of accountability in North Carolina reduced teacher retention rates, particularly in low-performing schools. This did not lead to lower qualifications of the teachers coming in to replace the leavers, however, so the overall impact on students is unclear. Elacqua et al. (2019) show a similar effect: After the introduction of a new accountability program in Chile, between-school mobility among less effective teachers in low-performing schools increased; however, there was no increase in mobility out of the profession as a whole.

Gjefsen and Gunnes (2020) study the introduction of school accountability in Oslo, Norway, in 2003. While they, too, find substantially increased teacher mobility in response to the provision of school quality information, by contrast they found that most teachers who changed jobs actually left the profession entirely. Again, this is likely to be transiently detrimental for students, as these teachers were typically replaced by high-ability teachers. In Florida, Feng et al. (2018) explore teacher mobility in more detail across the school-performance distribution. They show that teacher turnover is not generally affected by their school’s performance grade, but this is not true for schools designated as “failing.” Counterproductively, in such schools, it is the most effective teachers who disproportionately leave; even in this case, though, student scores improved.

Within schools, Boyd et al. (2008) show that high-stakes testing and accountability also altered the allocation of teachers to grades in New York. Teachers also face greater work pressure from accountability. Reback et al. (2014) show that accountability pressure from NCLB lowers teachers’ perceptions of job security and causes untenured teachers in high-stakes grades to work longer hours than their peers.

Overall, the evidence suggests that school accountability increases scrutiny to some degree on teachers’ work, but has not made major differences to career decisions. While teacher turnover does appear to increase somewhat with accountability, particularly at less effective schools, this is not inherently damaging to students. Studies differ in estimates of which teachers leave, but in most cases it is not clear that there is a negative impact on student outcomes.

Unintended Consequences

Schools may undertake strategies to game the system under accountability, in addition to or instead of efforts to improve pupils’ learning. These behavioral distortions can take many forms, from concentration of teacher time and effort, to outright cheating in exams.

It has been generally established that schools will tend to focus their resources on whatever is tested: the subjects that are tested, the topics within subjects that are tested, the topics in which scores can be increased most easily, the school grades that are tested, and on the pupils who may be pivotal in reaching a threshold. Figlio and Loeb (2011) summarize all this evidence, and Rouse et al. (2013) also review evidence on a range of responses by schools. Whether this focus on the things tested is a bad thing depends on the tests. This focus may in fact be what society wants and intends, and if the test is well-designed it may be wholly appropriate. Conversely, if the high-stakes tests are not well-designed, then the lack of broader knowledge and skills can be deleterious.

One way of gauging the degree of “teaching to the test” is to compare performance on high-stakes tests with that on low-stakes tests covering the same material. Jacob (2005) compared test score gains in math in a high-stakes test to those on a comparable, but low-stakes, test; he showed that the gains for eighth graders were confirmed in the low-stakes tests, but that those for fourth-grade pupils were not. Similarly, Figlio and Rouse (2006) find a smaller impact of accountability on low-stakes tests than on high-stakes tests.

Beyond focusing school resources on a subset of subjects, topics, and pupils, researchers have documented other practices, which while not illegal, are certainly not as intended. Figlio and Getzler (2006) show that students expected to be low-performing were disproportionately assigned into “special education” categories that were exempt from accountability measures. Figlio and Winicki (2005) show that schools change their lunch menus at the time of the tests, “substantially increasing calories in their menus on testing days”; Bokhari and Schneider (2011) show that pupils in schools under stronger accountability threat “are more likely to be diagnosed with Attention Deficit/Hyperactivity Disorder (ADHD) and consequently prescribed psychostimulant drugs”; and Anderson et al. (2017) show that pupils in such schools have a higher chance of being obese through less exercise in school.

Finally, there is straightforward cheating on the test by teachers or school administrators. Jacob and Levitt (2003) show that the frequency of cheating appears to respond strongly to relatively minor changes in incentives, such as those implied by school accountability measures. Bertoni et al. (2013) also implicitly detect cheating by noting that test scores in Italy are lower when external monitoring of tests takes place.

The existence of these inappropriate behaviors does not mean that accountability measures should be abandoned; the costs need to be weighed against the benefits.

School Choice and Market-Level Attainment

The previous section presented evidence to suggest that accountability—public information about schools’ performance—raises standards, although potentially with unintended consequences. The review now turns to the overall effect of school choice as a market mechanism on attainment, before considering the overall effect on segregation in the following section.

Whether school choice has the potential to raise market-level attainment is the subject of three, separate but related, strands of research. The first relates indices of competition to schools’ performance in systems of coordinated admissions, where schools compete for students without the introduction of voucher schemes or charter schools. The second assesses the impact of competition from independent schools through the introduction or expansion of school vouchers. The third instead estimates the impact of competition from alternative “outside options”: Catholic schools and Charter schools. In each strand of research, the identification problem is to isolate the impact of competition from factors correlated with competition, such as neighborhood composition, which might also affect productivity. In a meta-analysis, Jabbar et al. (2019) focus on how the context of school choice—the “rules of the game”—determines the competitive effects. Overall, the effects of school competition on student achievement were positive, but, the authors conclude that “the effects are too small to have a major impact on educational quality and inequality on their own.”

Coordinated Admissions

As described in “Allocation Mechanisms,” coordinated admissions are school choice systems where parents have a choice of schools within a school district or region. These areas may be more or less competitive, and may have higher and lower attainment as a result. The key issue is to identify a causal effect between competition and pupil attainment; many studies report associations between them (reviewed in Levin & Belfield, 2003; see, for example, Borland & Howsen, 1992) but plausibly exogenous differences in competition across areas are rare.

One leading study in this field is Hoxby (2000). She notes that current school district boundaries are related to historical school district boundaries, which in turn often follow natural landscape features, thus suggesting an instrumental variable and a spatial measure of competition. The findings suggest that metropolitan areas with more school districts and therefore higher competition, exogenously so because of landscape features, have higher attainment. These findings have been challenged by Rothstein (2007), however, arguing that they are not robust to simple changes in data coding or sample selection (including students in private schools); taking these into account, he finds no impact of competition. These suggestions, in turn, are strongly rejected by Hoxby (2007).

Dijkgraaf et al. (2013) study the impact of competition in the Netherlands, which has a long history of school choice characterized by the authors as “a full voucher program with 100% funding.” To construct an instrument for the concentration of schools in the local area, Dijkgraaf et al. (2013) use the number of school sites (rather than schools), argued to be a plausible instrument due to historical institutional factors that largely prohibit new schools. Increases in competition are associated with a small decrease in attainment in this context. De Haan et al. (2016) study school consolidation in the Netherlands as a result of reform, which reduced competition by creating fewer, larger, schools. Exploiting variation across municipalities (driven by non-linearities from the reform), they find no negative effect of the reduction in competition, and an overall positive effect resulting from economies of scale.

An alternative approach to studying the impact of competition is to exploit administrative boundaries. For the United Kingdom, Gibbons et al. (2010) use the distance of a primary school from its nearest local authority boundary to instrument the amount of competition it faces; they find no overall effect of choice or competition on school performance.

Finally, some research studies the implementation of school choice. Lavy (2010) studies the switch from interdistrict bussing to a school choice system in Tel-Aviv, Israel. As this is not experimental variation, Lavy uses alternative identification strategies (difference-in- differences) and comparison groups (untreated tangent neighborhoods and other cities) to show that the choice system increases school completion and raises cognitive achievement.

Voucher Schemes

An educational voucher entitles a child to attend a different school than their “default.” Details vary across schemes, but in essence it is seen as an “escape” from a low quality or poorly matched local school. (See Epple et al., 2017, for a summary of key features of voucher schemes worldwide.) The biggest voucher programs are in Chile, Sweden, and the Netherlands, but they are present in a wide range of other countries: Belize, Canada, Colombia, Denmark, India, Japan, New Zealand, and Poland (Bettinger, 2011; Epple et al., 2017), and of course in the United States, where there is normally specific entitlement (based on income, for example) rather than general entitlement.

This section focuses on the empirical evidence for whether competition induced by voucher schemes raises attainment.4 The identification problem is that voucher school enrolment (and therefore competitive pressure) is likely to be endogenous to the quality of local public schools. Residential location may also be endogenous to the quality of local schools for some households (discussed further in the subsection “Evidence From Residential Demand”).

There appear to be no definitive answers yet. In their substantial review, Epple et al. (2017) argue that the bulk of the findings suggest no significant effect, yet “multiple positive findings support continued exploration.” Similarly, Urquiola (2016) surveys the literature of the impact of competitive pressure from private schools following voucher provision on the performance of public (state) schools, concluding that the evidence is mixed. Egalite (2013) reviews the evidence up to November 2012, and concludes that the findings from studies with the most robust design are uniformly positive: Competition from private school vouchers increases student attainment. These peer-reviewed papers (Figlio & Rouse, 2006; Rouse et al., 2013; West & Peterson, 2006) study Florida, however, where vouchers are combined with public accountability, so are unlikely to isolate the impact of competition.5

Turning to research outside the United States, comparing areas with varying expansion of free schools, Böhlmark and Lindahl (2015) find small positive results from competition and choice in Sweden, 10 years on from the reform. The benefits are realized for longer-term outcomes such as university attendance, and are due to competition effects rather than higher productivity at the independent “free schools.” These findings are consistent with earlier research from Sweden (Sandström & Bergström, 2005). Using a difference-in-differences approach for Chile, Hsieh and Urquiola (2006) find that areas with higher growth of private schools have the same test score growth as areas with lower growth of private schools. Böhlmark and Lindahl (2015) suggest that the difference between findings for Chile and Sweden is due to the degree of selection by schools, subsequent sorting of students by socioeconomic background, and therefore household choices driven more by consideration of peer group rather than school productivity. In contrast, using instrumental variables for the growth of private schools across Chile, Gallego (2013) finds that “a one-standard-deviation increase in the ratio of voucher school to public schools in a market increases test scores by about 0.10 standard deviations.”

Muralidharan and Sundararaman (2015) provide the first experimental evidence on the provision of vouchers to private schools. In India (Andhra Pradesh), a “two-stage lottery-based allocation of vouchers” created student-level and market-level experiments. Muralidharan and Sundararaman (2015) find no evidence of competition-induced spillovers to public school students. Those in public schools in villages with competition from private school vouchers have the same attainment as those in public schools in control villages.

In earlier work, Hoxby (2003) finds that public schools respond to competition from the choice program by raising the achievement levels, although subsequent work challenges these findings (Bettinger, 2005; Bifulco & Ladd, 2006; Ladd, 2002, 2003).

Overall, there seems to no unanimous evidence (outside Florida, where the reforms also included accountability measures) that competition induced by voucher schemes raises performance of public schools. The only experimental evidence, from India, finds no effect, which is perhaps the overriding finding—students in public schools are not typically harmed or helped through competition induced by voucher schemes.

Other “Outside Options”: Competition From Catholic and Charter Schools

Competition for public/state schools can come from alternative schools, such as Catholic schools and, in the United States, Charter schools. The evidence for each source of competition is discussed in this section.

Card et al. (2010) evaluate whether competition between publicly funded secular and Catholic primary schools in Canada lead to more productive schools. Areas with more children from a Catholic background have greater demand-side pressure for secular schools, which should raise standards. The identifying assumption is that the proportion of Catholic families across areas is otherwise uncorrelated with school outcomes. The authors find that, conditional on the percentage of Protestant families in an area, there are small positive effects from competition.

In contrast, evidence for England finds little evidence that competition from Catholic schools raises area-wide pupil attainment. Allen and Vignoles (2016) use national pupil-level data for secondary schools, instrumenting the number of Catholic schools with the historical Catholic population. This research design is similar to West and Woessmann (2010), who find that historically induced competition from Catholic private schools increases attainment, across a sample of OECD countries that participated in PISA 2003.

Competition from Charter schools is not exogenous. Identification strategies to overcome this problem are to exploit the timing of Charter school entry (Cordes, 2018) or instrumental variables (Bettinger, 2005; Gao & Semykina, 2020; Imberman, 2011), most often combined with student fixed effects. Cordes (2018) finds small positive effects of Charter school competition on public school performance in New York City, which increase with proximity to the Charter school (highest in schools that are “co-located”—in the same building). Cordes (2018) notes that previous research has typically studied the impact of competition from Charter schools at the district or larger geographical area level. These studies find either small positive (Sass, 2006), neutral effects (Bettinger, 2005; Bifulco & Ladd, 2006; Zimmer & Buddin, 2009), or negative effects (Imberman, 2011; Ni, 2009). Gao and Semykina (2020) adopt an alternative measure of competition from Charter schools based on travel time, finding typically neutral competition effects.

Consistent with previous reviews (Epple et al., 2016; Gill, 2016) it is clear that there is no overwhelming support for strong competitive effects from Charter schools. Research is largely consistent (with the exceptions of Imberman, 2011, and Ni, 2009) that students in public schools are not harmed by the presence of local Charter schools, at least. Small positive or neutral competitive effects are typical, providing some evidence for “healthy competition hypothesis” (Gill, 2016).

School Choice and Market-Level Sorting

The section “School Choice and Market-Level Attainment” presented no overwhelming evidence that school choice and/or competition between schools induces a strong improvement in pupil attainment. This is surprising, as the section “Parents’ Preferences for Schools” concluded that parents typically value school quality, at least as measured by test scores. This inconsistency will be considered in the final section “Taking Stock.” The framework for studying school choice and accountability also highlighted the potential for school choice to influence segregation across schools, which will be the subject of this section.

Theoretical and empirical research has contributed to the complex interaction between school choice and sorting of students across schools. Beginning with the theoretical literature, Hoxby (2003) argues that there are no very general theoretical predictions about student sorting under school choice. For example, Nechyba (2003) shows that a pure state school system leads to more spatial segregation than a private system.

Similarly, Epple and Romano (2003) analyze three different student assignment regimes: neighborhood schooling (a strict residence requirement for admission), school choice with no choice costs, and choice over many school districts, and show that different public policy regimes have dramatic effects on the nature of sorting. Neighborhood schooling leads to strong income stratification across neighborhoods, whereas costless, frictionless choice equalizes peer groups across schools. Calsamiglia et al. (2015) assess the impact of school allocation rules on socioeconomic sorting into schools and neighborhoods. The main findings are that priority in oversubscribed schools to local applicants leads to segregation, whether or not a truth-revealing allocation mechanism is used.

Previous empirical research has used either event analysis or counterfactual simulation to study segregation under school choice. In the former strand of research, the consensus is that introducing school choice has not led to markedly higher segregation between social groups in England (Allen & Vignoles, 2007; Goldstein & Noden, 2003; Gorard & Fitz, 2000; Noden, 2000). Analyzing student-level data from England, Burgess et al. (2007) find that the degree of student sorting by ability and socioeconomic status varies considerably across the country, however.

Elsewhere, school choice is related to increases in segregation across schools. In Chile, vouchers led to increased sorting as the “best” students left for private schools (Hsieh & Urquiola, 2006). In New Zealand, where schools were permitted to charge noncompulsory fees in addition to per-pupil funding from government and interview parents for their “enrollment scheme,” sorting between European and Minority students increased (Ladd & Fiske, 2001). In Stockholm, segregation by ability increased in response to reforms that replaced residence-based admissions criteria with free school choice, where admission to oversubscribed schools is dependent on academic grades (Söderström & Uusitalo, 2010) and in Sweden more generally (Böhlmark et al., 2016). For the United States, segregation by race increases in North Carolina in response to Charter school expansion (Bifulco & Ladd, 2007). The only paper in this strand of literature from Asia finds consistent patterns. In Seoul, South Korea, schools became more segregated by student performance after a school choice policy was introduced, with segregation increasing from a low level (Oh & Sohn, 2019).

In the latter, counterfactual, strand of research, segregation is typically found to decrease under “neighborhood” allocation to schools (Allen, 2007; Bernelius & Vaattovaara, 2016; Bifulco et al., 2009; Östh et al., 2013; Taylor, 2009). Glazerman and Dotter (2017) find that a neighborhood schools policy would decrease segregation by race but increase segregation by income.

It is known that higher residential segregation is linked to higher segregation in schools (Böhlmark et al., 2016; Burgess et al., 2005; Lindbom, 2010; Taylor & Gorard, 2001). Lindbom (2010), drawing on analysis from Lindbom and Almgren (2007), concludes that residential segregation in Sweden is the “main villain of the piece,” being the predominant cause of school segregation. Aside from residential segregation, possible drivers include those on the supply side, for example “cream-skimming” of desirable pupils (West et al., 2004, 2006), and, on the demand side, differences in preferences and inequalities in information between groups (see “Parents’ Preferences for Schools”). The contribution of households’ preferences to segregation is particularly under-explored, with only one working paper decomposing segregation between multiple sources (Oosterbeek et al., 2019).

Overall, the evidence from a number of countries suggests that the combined process of choice by parents and proximity rationing by schools leads to greater sorting. Evidence from Madrid, Spain, shows that school choice policies must have careful design to achieve diversity objectives (Gortázar et al., 2020). How to design school choice systems to encourage diversity is an area of on-going research. Bjerre-Nielsen and Gandil (2020) demonstrate that redesigning school districts in Denmark is insufficient, as households respond by exiting to another public school, or, to some extent attending a private school or moving home.

Do Schools Matter?

This review has largely been concerned with the allocation of pupils to schools, and whether school choice and accountability improve pupils’ outcomes. Before taking stock in the final section, this penultimate section summarizes the evidence that schools do in fact matter—that they have a causal impact on pupil attainment. In each case, the identification problem is that “better” students might select “better” schools. Identification strategies to overcome this problem typically use regression discontinuity design, comparing those marginally accepted or rejected, or exploit lotteries for admissions that act as natural experiments in allocation to schools. Even in this case, however, non-random attrition may bias the results: Those that are marginally rejected are more likely to “opt out” of the state system (Bibler & Billings, 2020; Howell, 2004).

To summarize this literature, Beuermann and Jackson (2020) conduct a meta-analysis on all publicly available studies that use “quasi-random assignment to a preferred (non- Charter) public school (either through lottery or selective enrollment exam).” Beuermann and Jackson (2020) find that

The precision-weighted average test score impact across all studies is positive. However, it is small and cannot be distinguished from zero. Importantly, this is not the result of averaging large positive and large negative effects, but rather reflects the fact that most studies (13 out of 17) cannot reject null impacts for the overall population.

Epple et al. (2017) summarize the findings on the effect of attending a voucher school thus: “A perhaps surprisingly large proportion of the best-identified studies suggest that winning a voucher has an effect on achievement that is statistically indistinguishable from zero.” In the only experimental evidence, Muralidharan and Sundararaman (2015) find no differences in test scores two and four years after allocation between lottery winners and losers in India, with the exception of Hindi, which is not taught in public schools.

For marginal students attending an “elite” exam school, the estimated effect of admission on educational attainment is neutral (Abdulkadiroğlu et al., 2014; Clark, 2010; Dee & Lan, 2015; Dobbie & Fryer, 2014; Lucas & Mbiti, 2014; Zhang, 2016) or positive (Clark & Del Bono, 2016; Deming et al., 2014; Ding & Lehrer, 2007; Pop-Eleches & Urquiola, 2013). Longer-run outcomes are more consistently positive (Berkowitz & Hoekstra, 2011; Clark, 2010; Clark & Del Bono, 2016) with the exception of Dobbie and Fryer (2014). Dustan et al. (2017) highlight the trade-offs that households face in admission to elite schools, in the Mexican context. Marginal admission increases test scores in math, but also significantly increases the probability of high-school dropout, particularly those with longer commutes and weaker prior attainment.

A limitation of this strand of research is that the estimated effects are generalizable to marginal students only. Exceptions are where the country context permits multiple “cut-offs” for admission. Lucas and Mbiti (2014) are able to study heterogeneous effects of entry to the top schools in Kenya through the multiple thresholds within each school caused by district-specific quotas. Despite large changes in peer group (scoring around half a standard deviation higher on the baseline test compared with the alternative school), school resources, and teacher experience, there are no significant improvements in test scores for marginally admitted students. There are no significant differences in test score effects by baseline scores, quality of alternative options, gender, or socioeconomic status.

Finally, the effects of attending a “preferred school” through coordinated admissions are typically positive and reasonably sized (Abdulkadiroğlu et al., 2020; Deming et al., 2014; Jackson, 2010) or neutral (Cullen et al., 2006).

The discrepancy between high demand for these schools and lack of test score gains, overall, may be due to causal improvements in longer-run outcomes that are unrelated to shorter-run test score gains. To test this hypothesis, Beuermann and Jackson (2020) use survey data from Barbados, in addition to administrative educational data, to estimate the causal effect (from regression discontinuity in test scores) of attending a preferred school on “a broad set of social and economic outcomes measured in adulthood.” Consistent with the meta-analysis, they find no effect of preferred school attendance on test scores at the end of secondary school. Longer-term outcomes are improved, however: Marginal pupils are more likely to earn a further education credential at age 18, and between ages 25 and 40 there is considerable improvement in an index of adult well-being.

This body of research is expanding rapidly, moving to explore the heterogeneous effects of school effectiveness (Jackson et al., 2020). Research at the frontier estimates returns to school effectiveness, rather than measures of academic attainment or indicators for “elite” school status.

Taking Stock

Research on school choice and accountability is important to determine whether these two commonly used tools to improve standards of public education are effective. The evidence suggests that accountability—public information about schools’ performance—raises standards. In part, this works through changing the pattern of teacher turnover, with less effective teachers more likely to leave. Although accountability can lead to perverse incentives, the balance of evidence suggests that accountability should remain unless these costs outweigh the benefits.

Despite intuitive theoretical arguments, a substantial body of empirical evidence finds that competition induced by school choice has had limited (though positive) effects on the performance of public schools. One condition for this to occur would be if parents do not value standards of education. At face value, this is not the case—research from two strands of literature (parents’ revealed preferences from school attributes and the response of property prices to local school quality) is unanimous that parents do value “school quality.” Whether parents value school effectiveness, however, rather than academic standards (that conflate school effectiveness and the peer group) is more debated. MacLeod and Urquiola (2019) provide a theoretical model for why parents may rationally care about test scores above value added, with education partly viewed as an investment good, rather than only a consumption good. The essence is that school choices will be influenced by factors other than effectiveness if parents care about longer-run outcomes, for example employability or prospects in the marriage market, and schools differ in their access to the most productive matches.

Overall, there is mounting evidence that parents do value school effectiveness. Research that uses a rich set of school attributes and/or estimates school effectiveness on longer-run non-academic outcomes tends to find that parents do value this dimension. The consensus that the competitive effect of school choice is small is therefore surprising, but may be due to parents’ preferences for proximity and the peer group. Abdulkadiroğlu et al. (2020) note that estimated preferences for peer quality may be because parents use this observable attribute of schools to proxy for unobservable measures of quality, concluding that “Distinguishing between true tastes for peer quality and information frictions is another challenge for future work.”

A second conundrum is why parents have such strong preferences for “quality,” when there is inconsistent evidence that schools causally improve academic outcomes. Empirical research largely using boundary discontinuities finds limited support that attending a “better” school improves attainment, either through a voucher to attend a private school, or an exam to attend an “elite” school. There is slightly more evidence of benefits from assignment to a preferred school assigned through an algorithm. There is mounting evidence for an impact on longer-run and non-academic outcomes, however, particularly for “elite” schools, which might suggest that parents have some information about the potential wider benefits to these schools. The insight from MacLeod and Urquiola (2019) that education is partly viewed as an investment good may also contribute.

There is evidence for heterogeneity in preferences between more and less advantaged households, with less advantaged households typically placing less weight on school quality. Further research should explore whether this reflects constraints in access or information, however, as the provision of information can reduce such inequalities (Hastings & Weinstein, 2008). Further research into the effect of information provision and the “choice architecture” (Glazerman et al., 2020) in the digital age is required.

In response to competition, schools may choose alternative responses, covert or overt selection of students to improve observed “quality,” for example. Abdulkadiroğlu et al. (2020) provide suggestive evidence that schools in New York City have increased screening of students in the years following the introduction of centralized school choice. Outside of the economics literature, there is also evidence that schools respond to competition by increased expenditure on marketing (Oplatka & Hemsley-Brown, 2012) as a lower risk alternative to pedagogical reform.

There are some unambiguous benefits of school choice. Centralizing school choices is clearly welfare improving, but surprisingly not universally in place in developed and developing countries. The benefits of truth-revealing assignment mechanisms within a centralized system are evident, although with the trade-off that parents cannot express the strength of their preferences in mechanisms currently used in practice. Further research is needed to determine the optimal design of school choice to meet policy objectives, for example reducing segregation and enhancing social mobility through more equal access to effective schools.

References

Notes

  • 1. Throughout the review, “parent” and “household” will be used interchangeably.

  • 2. Ireland’s school admissions process is still largely decentralized, for example (matching-in-practice.eu).

  • 3. Abdulkadiroğlu et al. (2015) propose the Choice-Augmented Deferred Acceptance (CADA) algorithm to allow the strength of parents’ preferences to be taken into account (rather than a random assignment) when multiple pupils have the same priority at a school, while retaining the truth-revealing property of SPDA. This has not, to our knowledge, been used in practice.

  • 4. The section “Do Schools Matter?” summarizes the evidence that attending a voucher school improves attainment.

  • 5. See also following work (Figlio et al., 2020).