School Competition (Choice, Charter, Open Enrollment) and Test-Score Inequality

What the evidence says

Charter-school achievement estimates vary sharply by identification strategy and location, and the studies do not agree on a single direction. Lottery-based causal designs report the largest gains: eighth-graders in New York City charter schools score 20 to 30 points higher on state math and reading assessments than public-school peers, closing about two-thirds of the urban-suburban achievement gap in that metro area. In Boston, charter middle-school attendance increases math scores by 0.4 SD per year and English Language Arts by 0.2 SD per year, and charter high-school attendance increases ELA/math by about 0.2 SD per year and writing by 0.17–0.28 SD per year. In Florida, charter achievement starts below traditional public schools in year one (2.0–2.1 points lower in math, 13% of an average year's gain; 1.5 points lower in reading, 12%) and reaches parity in math while exceeding traditional public schools in reading by 1.3 points (10% of a year's gain) by year five. Associational comparisons elsewhere find no such pattern: charter achievement in California and Philadelphia is statistically indistinguishable from traditional public schools, and Minnesota charter schools score lower on average than comparable traditional public schools after nearly two decades of operation.

Where private schooling widens test-score inequality, the evidence points to a within-school channel rather than a between-school one. Private school entry increases the variance of private-sector test scores by about 0.08 SD, and this growth traces mainly to inequality growing within schools, not across them. The private schools studied set an instructional level that favors richer students at the expense of poorer ones, even within the population that attends private school, and the entry of an additional private school increases schools' incentive to move further toward that wealthier-favoring level — making wealthier students relatively more marginal to retain. The resulting rise in within-school inequality comes mainly from reduced learning gains among poor students, while private entry shows no measured effect on mean test scores in either the public or private sector. The same study estimates that a school could increase poor students' annual test-score gains by up to 0.36 SD — nearly a year of schooling — by shifting from the instructional level optimal for wealthy students to the one optimal for poor students.

A randomized voucher lottery in Andhra Pradesh, India isolates the causal effect of private-school access from who selects into it. Lottery winners and losers show no difference in Telugu and math scores after two and four years, even though a raw cross-sectional gap of 0.65 SD separates public and private students — indicating that gap mostly reflects selection rather than instruction. Winners do score higher when all subjects are averaged (+0.13 SD), and the instrumented effect of attending a private school on test scores overall is +0.23 SD, with a large gain in Hindi (+0.55 SD, p<.001) and a smaller one in English (+0.12 SD, p=.098). Private schools achieve Telugu and math outcomes equal to public schools while spending 40% less instructional time on Telugu and 32% less on math, redirecting that time to English, science, and Hindi — the mechanism the study ties the non-core-subject gains to. The program shows no measured spillover on public-school non-applicants or on students already in private schools, and program participation does not correlate with household socioeconomic characteristics. Mean per-student cost is over three times higher in the public system than in the private schools sampled.

Effects of school competition on traditional-public-school achievement are small and inconsistent. A meta-analysis of the broader literature finds a partial correlation between competition and achievement of 0.06 at the school level (borderline, p<.10, confidence interval crosses zero), 0.001 at the student level (statistically significant but very small), and −0.002 at the district level (not significant). Site-specific studies split on direction: no measured effect on neighboring traditional public schools in a multi-state RAND study, a decrease in student performance in Ohio and Michigan, and a modest increase in math with unchanged reading in Florida. The same meta-analysis states these competitive effects are too small to produce a major change in achievement or inequality on their own, and finds no overall negative effect on traditional-public-school students associated with competition.

Racial segregation is the one lever where the direction of harm is consistent, though the underlying mechanism is largely unstated. Segregated learning environments are associated with lower test scores generally, and schools with 90% or greater minority enrollment combined with concentrated poverty have lower average test scores, less qualified and experienced teachers, and fewer advanced courses. In North Carolina, Black students who made a racially segregating transfer into a charter school show larger negative achievement effects than Black students who stayed in traditional public schools or made a non-segregating transfer, and North Carolina charter schools overall produced larger negative effects on Black students' scores than on White students' scores. Separately, a meta-analysis finds the competition-achievement association grows larger as a school's minority student share rises (β=0.04 per unit at the student level), though whether that differential narrows or widens the achievement gap for those students is not established in the sources. No claim here quantifies how large the racial effects are in point terms — only their direction and relative size against comparison groups.

Where the evidence splits

Selection bias ("cream-skimming") is argued two different ways with no reconciliation offered in the sources. One paper argues, on theoretical grounds, that several kinds of selection bias make achievement comparisons between charter/private and traditional public schools "virtually impossible," and that excluding lower-achieving student groups from enrollment mechanically raises a school's measured average. A separate multi-site empirical study, using student-level longitudinal comparisons of prior test scores, finds no systematic evidence that charter schools skim high-achieving students — transferring students' prior scores were at or below local averages in every site studied. The sources do not resolve this: one is a theoretical account of a general mechanism, the other a direct empirical test of one specific channel (prior achievement) across seven locations, and neither states whether other selection channels the theoretical argument raises (for example, parental effort or informational barriers) were tested.

Charter/competition effects on neighboring traditional public schools split by state and by unit of analysis, and the sources only partly explain why. A multi-state RAND study and a Zimmer et al. analysis find no effect on neighboring public schools' scores; Ohio and Michigan studies find a slight decrease; a Florida panel study finds a modest increase in math and no change in reading; and a meta-analysis finds the association near zero at the district level but small and positive at the school level. The sources tie part of this to geography (different states) and to unit of analysis (school vs. student vs. district level in the meta-analysis), but do not explain why Ohio and Michigan diverge from Florida beyond location.

The overall direction of charter achievement effects splits by study design as much as by geography. Positive results in New York City and Boston come from lottery-based quasi-experiments; null or negative results in Minnesota, California, and Philadelphia come from associational comparisons. Florida's panel study, also not lottery-based, finds convergence to parity in math and a positive reading effect by charter schools' fifth year — a result the sources do not reconcile with the null Philadelphia and California findings, which use similarly non-experimental designs.

Thresholds and dose-response

Schools with 90 percent or greater minority enrollment combined with concentrated poverty have lower average test scores, less qualified and experienced teachers, and fewer advanced courses than other schools.

Charter school achievement follows a dose-response pattern over time in the Florida panel study: it starts below traditional public schools in year one (2.0–2.1 points lower in math, 13% of an average year's gain; 1.5 points lower in reading, 12%), then rises to match traditional public schools in math and exceed them in reading (+1.3 points, 10% of an average year's gain) by the fifth year of operation.

Shifting a school's instructional level from the level optimal for wealthy students to the level optimal for poor students is estimated to raise poor students' annual test-score gains by as much as 0.36 SD.

Open empirical questions

• Whether charter and voucher achievement effects persist beyond the two-to-five-year windows most of these studies cover — none of the sources here track outcomes into later schooling or adulthood.
• Whether the North Carolina finding of larger negative effects on Black students from segregating charter transfers replicates in other states; the sources note this explicitly as unresolved.
• Whether selection-bias-driven inflation of achievement comparisons (argued theoretically in Frankenberg 2003) can be measured directly beyond the single prior-achievement test used in the RAND 2009 cream-skimming analysis.
• Whether the stronger competition-achievement association found for schools with higher minority shares (Jabbar 2019) reflects a narrowing or a widening of the achievement gap for minority students specifically — the sources report the moderator but not its distributional consequence.