Academic Performance and Single-Sex Schooling: Evidence from a Natural Experiment in Switzerland

The within-school design eliminates the self-selection problem that weakens most single-sex education research. But the randomization is asserted by school administrators, not documented through any protocol. The structural pattern across cohorts is unexplained: seven of eight cohorts had exactly one single-sex class out of four or five, then Cohort 7 suddenly had three, and Cohort 8 had none. Only nine of 39 classes are single-sex — too few treated clusters for reliable inference. The most consequential problem is that teacher-assigned grades with no standardized alternative cannot separate learning gains from grading behavior, and the paper's discussion of teacher-class "symbiotic relationships" describes exactly how grading could shift without teachers intending it. The survey adds only weak evidence: the more comprehensive psychological measure (math self-concept) did not reach standard thresholds, and the one that did (self-assessment) was marginal.

The regression framework — student-level random effects, cohort indicators, standard errors clustered at the class level — is appropriate in structure. But with only nine treated clusters, cluster-robust standard errors produce intervals that are too narrow, so every reported significance level is likely overstated. No power analysis is reported, which matters most for the German null result: German grades have a standard deviation of about 0.4 versus 0.7 for math, reducing the study's ability to detect effects, yet the paper treats "no effect in German" as a positive finding about subject specificity. The most striking red flag is the t-statistic of −12.375 for the female-teacher effect within single-sex classes (Model 7). An effect that precisely estimated from 898 observations across at most nine classes taught by eight teachers is not identifying "teacher gender" — it is capturing one or two specific individuals. The psychological-channel test also runs counter to the paper's claims: adding the self-concept and self-assessment measures as controls leaves the single-sex coefficient intact, which is evidence against mediation, not for it.

The paper claims single-sex education improves math proficiency but measures teacher-assigned grades at a school with no standardized exams. A teacher who develops a different dynamic with an all-girls class could shift grades without consciously changing standards, and no external benchmark exists to check. The grading-on-a-curve robustness test rules out one specific alternative (boys capturing top grades under a forced distribution) but not the broader problem. The teacher-gender finding is labeled a gender effect, but with three female math teachers total, the coefficient is a weighted average of those three individuals' grading patterns — teacher identity and teacher gender are nearly the same variable. The proposed psychological channel is contradicted by the paper's mediating-variable test: the single-sex grade effect survives adding the psychological measures as controls, meaning those measures are not explaining the grade difference. The abstract's claim that single-sex schooling "strengthens female students' self-confidence" overstates the survey data, where the self-concept measure did not reach standard thresholds.

Three reasoning problems recur throughout. First, the paper frames its finding as improved math proficiency, but all it can show is higher teacher-assigned grades — and it never seriously considers that grading behavior could differ across class types, even while discussing teacher-class dynamics as a mechanism for the teacher-gender interaction. Second, the teacher-gender finding is presented as evidence about male versus female teachers when the data cannot distinguish gender from individual teacher quality. Third, the psychological-channel claim is contradicted by the paper's test — if self-confidence were driving the grade effect, controlling for it should reduce the single-sex coefficient, but it doesn't — yet the paper still treats this channel as "an important driver." The German null result is used as confirmatory evidence of subject specificity without assessing whether the study could have detected an effect in German if one existed.

The paper discloses its funding source and reproduces the survey items. No data, code, pre-registration, or ethics approval is reported. For a 2011 economics working paper, some of these absences were standard practice. The more consequential omission is in the limitations discussion: the paper acknowledges it cannot cleanly identify the psychological-channel causation and that its competition-based interpretation is speculative, but does not acknowledge that teacher-assigned grades cannot distinguish learning from grading behavior.

Individual citations are accurately represented. The paper is candid about which interpretations are speculative and presents contrary evidence on competitor-gender effects (Gneezy and Rustichini 2004; Gneezy, Niederle, and Rustichini 2003). The literature review omits the substantial body of null and negative findings on single-sex education, presenting the field as understudied rather than actively debated. The contradiction between this paper's finding (male teachers help girls more) and Carrell, Page, and West 2010 (female professors help female students more) is noted in a footnote but not treated as the significant tension it is for the interpretive framework.

This is publicly funded university research with no commercial conflicts. Two authors are affiliated with an institute in the same town as the school, and the school administration provided data and facilitated the survey — a cooperative relationship that creates mild structural pressure but not a material conflict. The study includes design features that could have produced unfavorable results (testing German alongside math, checking balance on entry exams).