Causal Effects of Single-Sex Schools on College Entrance Exams and College Attendance: Random Assignment in Seoul High Schools

The lottery-based assignment is the strongest design available for this question — far better than the observational studies that dominate the field. The outcome measures are also strong: a uniform national exam administered and graded by the government, and official college enrollment records.

Three problems limit what the design can establish. First, the randomization balance check used students from across South Korea, not Seoul specifically — other cities had loosened the lottery to allow partial school choice, so the check validates a mix of full and partial randomization. Second, the noncompliance estimate (4.5% of students moved districts) comes from 2000 census data, nine years before the study cohort. Third, single-sex and coeducational schools differ sharply in governance: the heavy overlap between single-sex status and private status means the study compares a predominantly private treatment group against a predominantly public control group, and a single binary control variable cannot absorb the institutional differences between those systems.

The methods — hierarchical linear models for exam scores, district fixed-effects regressions for college attendance — are appropriate and correctly specified.

The problems are in interpretation and testing. The paper runs at least eight primary comparisons (two genders × two exam subjects, plus two genders × two college types) with no correction for multiple testing. Several results survive only at a relaxed 10% threshold. The paper leads with the larger uncontrolled estimates: the headline figure for boys' college attendance is 5.6 percentage points and "about 0.8 standard deviations," but after controls it falls to 2.4 percentage points — a 57% drop — while the private-school coefficient in the same model is 7.2 percentage points, nearly three times larger and the dominant predictor. The paper describes the smaller controlled estimates as "substantial" without flagging that several are borderline. The randomization check, conducted on the wrong population with modest sample sizes, could not detect moderate imbalances.

The paper frames its treatment as "attending a single-sex school." What students actually received was a bundle: gender-segregated classrooms, private institutional governance, different teacher hiring and retention practices, and different school culture. Because 80% of single-sex schools are private and 70% of coeducational schools are public, these components cannot be separated statistically.

The attenuation pattern reveals the problem. When controls are added, effects on boys' Korean scores drop 40%, girls' college attendance drops 50%, and boys' college attendance drops 57%. The paper also mentions — in a single sentence, without showing data — that controlling for the share of male teachers "substantially reduced" the all-boys school effect. Teacher gender composition is a direct consequence of private schools' hiring autonomy. If the effect is largely explained by a variable the paper identified as important but chose not to include in its published models, "single-sex schooling" does not accurately describe what is being measured.

The external validity claim — that Seoul's findings suggest single-sex school effects elsewhere are not just selection artifacts — does not follow. Seoul's institutional structure (uniform curriculum, uniform tuition, high-stakes national exam, predominantly private single-sex schools) has no counterpart in the countries where single-sex schooling is debated.

The underlying logic is sound: exploiting a lottery for causal identification is the right approach, and the paper handles several methodological details carefully (hierarchical modeling for nesting, robust standard errors, separate models by gender).

The reasoning falters in three places. First, the paper characterizes effects that shrink by half and survive only at the weakest conventional threshold as "substantial." Second, the most informative preliminary analysis — showing that teacher gender composition substantially reduces the boys' school effect — is mentioned in one paragraph but never shown, despite bearing directly on the central claim. Third, the paper concludes that its findings suggest single-sex schools "may produce positive outcomes not attributable to differences in characteristics of students and their families," which is stronger than the data support, given that the effects cannot be separated from private-school characteristics.

Funding is fully disclosed from standard academic and government sources. Data sources are described clearly. Methods are explained in sufficient detail for replication.

The gaps: no conflict-of-interest statement, no ethics review mentioned, no open data or code, and no confidence intervals reported. The limitations discussion acknowledges that generalization is uncertain and that teacher distribution is not random, but does not mention that the balance check used the wrong population, that compliance was estimated from a different decade, or that several core results weaken to borderline significance after controls. The decision not to show the teacher-gender analysis — while mentioning it exists — is the most consequential transparency problem.

The paper cites both supporters and skeptics of single-sex schooling. But the two most relevant comparison studies are handled selectively. Jackson (2012), which used rule-based student assignments in Trinidad and Tobago to study the same question, is cited only for a methodological point — its actual results are never discussed. Sullivan, Joshi, and Leonard (2010) is cited for a minor point about classroom discipline while its headline findings — that single-sex schooling was positive for girls at age 16 but neutral for boys and faded over time — go unmentioned, despite being directly relevant. The theoretical section relies heavily on studies from the 1960s through 1990s without updating for more recent work.

Funding comes from the Spencer Foundation, the National Academy of Education, and the NIH — standard academic sources with no stake in whether single-sex schools are found effective. The study design was constrained by a pre-existing government policy, limiting the researchers' ability to choose a design favoring a particular result. No structural conflicts of interest are identifiable.