Effects of single-sex secondary schools on student achievement and attitudes.

Restricting to Catholic schools reduces background variation, and the standardized achievement tests from a national survey are a real strength. But for girls' schools — where nearly every headline finding originates — single-sex status and religious-order governance are almost perfectly confounded. Religious orders bring distinctive academic cultures, hiring practices, and institutional missions that could independently produce every observed result. The paper ran a governance check for boys' schools and found no change in effects, but this check was impossible for girls' schools because the two variables overlap almost completely.

Single-sex and coeducational schools also differ on school size, student-to-faculty ratios, faculty gender composition, per-pupil expenditure, and curricular diversity. Only four school-level variables entered the regression, and the 75-school sample precluded adding more. No pre-high-school achievement measure is available, so the most important predictor of later achievement is missing from every model. The selection equations that predict school-type attendance explain only 3% of variance for girls and 7% for boys, which means the model captures almost nothing about the selection process — not that selection bias is small.

The paper uses ordinary least squares regression with a fixed design-effect correction of 1.5 to approximate students nested within schools. A multilevel model would have been more appropriate and would likely produce larger standard errors for school-level variables, potentially changing which results cross thresholds. All 74-odd tests are one-tailed with no correction for multiple comparisons. The overall count of results crossing the threshold exceeds what chance would produce, so the pattern is real in some form, but individual marginal results cannot be distinguished from noise.

The claim that effects are "stronger for girls than boys" was never formally tested — no pooled model with a gender interaction was run. Comparing whether one group's coefficient crosses a threshold and the other's does not is not evidence that they differ. The asymmetric exclusion of vocational schools (five girls' schools, three coeducational, zero boys' schools) selectively trims lower-performing girls' schools, biasing the comparison in favor of the single-sex group. Attrition rates by school type are unreported.

The single-sex dummy variable in the regression absorbs every unmeasured school characteristic that differs between school types — governance, size, faculty composition, curricular diversity, per-pupil expenditure, and school policies. The evidence supports "something about these particular single-sex schools is associated with better outcomes." It cannot support "single-sex organization produces better outcomes."

The proposed mechanism — that separating academic and social environments removes distractions — has no measurement in the study. No variable captures social pressure, peer distraction, or academic-social separation. The data are equally consistent with every alternative explanation the paper lists (and several it doesn't), yet the mechanism is presented in the abstract and conclusion as though it were an empirical finding.

The paper acknowledges its most important limitations — the religious-order confound, the missing achievement baseline, the imperfect selection controls — but its conclusions do not honor them. It states that governance and single-sex status are inseparable for girls, then attributes girls' school effects to gender composition in the abstract. It frames each analysis format's weakness as favoring the hypothesis: cross-sectional results may overestimate, but change analyses underestimate, so positive results from either count. When boys' schools show no achievement gains from sophomore to senior year, this is attributed to gains having already occurred by sophomore year rather than treated as evidence against a boys' school effect. The reasoning consistently resolves ambiguity in favor of the preferred conclusion.

The paper is forthcoming about several important limitations. The religious-order confound is stated plainly. The missing achievement baseline is acknowledged. The selection equations' low explanatory power is reported. Data come from a publicly available national survey. Funding is clearly disclosed. Where transparency falls short is between what the limitations section concedes and what the abstract and conclusions claim. The paper's caveats, if taken seriously, would substantially weaken its headline conclusions — but those conclusions are stated without qualification.

The literature review is shaped to favor the paper's conclusions. Dale's three-volume study of British schools — the largest body of prior secondary-level research, which found a modest advantage for boys in coeducational schools — is dismissed by questioning the researcher's objectivity ("zealous advocates of coeducation") rather than engaging with the data. Coleman's 1961 study of teenage social dynamics serves as the theoretical anchor without acknowledging that the social context had changed substantially by 1986. A contested psychological construct (Horner's "fear of success") is cited without noting the criticism it had attracted. The decision to use one-tailed tests is justified by prior literature the paper describes as inadequately controlled.

The primary funder was the Radcliffe Research Support Program through Radcliffe College's Murray Research Center. Radcliffe was one of the Seven Sisters women's colleges whose graduates the paper's literature review cites as evidence for single-sex education's benefits. The paper's strongest conclusion — that single-sex schooling particularly benefits girls — aligns with the institutional interest of its funder during a period when women's colleges faced conversion pressure. Several discretionary design choices (one-tailed tests, no multiple-comparison correction, asymmetric school exclusion) consistently favor more and larger positive findings for girls' schools. The underlying data source is independent and publicly available, and the conflicts are modest compared to pharmaceutical or industry-funded research, but the alignment is notable.