Few questions in education produce more confident claims from more directions than this one, and few have a research base so thoroughly compromised by a single methodological problem.
The problem is selection. Students do not arrive in single-sex schools at random.
Why the raw comparison is uninformative
In most countries, single-sex schools are disproportionately independent, selective, religious, or historically established with strong alumni networks. Their intakes differ systematically on prior attainment, family income, parental education and parental engagement.
Comparing outcomes without adjusting for those differences measures the intake, not the schooling. Any school that admits students who were already going to do well will produce students who do well.
This is not a marginal confound. In several national datasets, controlling for prior attainment and socioeconomic background eliminates most or all of the raw advantage.
What the better studies find
The most credible evidence comes from three sources.
Value-added analyses that track the same students' progress and control for prior attainment generally find small or null effects on academic outcomes. Where effects appear, they are usually modest and inconsistent in direction across subjects.
Natural experiments — systems where assignment to single-sex or coeducational schooling is effectively random, such as certain lottery-allocated systems in South Korea — provide the strongest identification available. Results from these have been mixed, with some finding meaningful effects on university entrance outcomes and others finding little.
Systematic reviews and meta-analyses pooling across studies have generally concluded that effects on academic achievement are small once methodological quality is accounted for, and that the highest-quality studies report the smallest effects. That pattern — effect size shrinking as study quality rises — is the classic signature of a confounded literature.
Where an effect does seem to survive
Subject choice is the outlier, and it is the finding most worth taking seriously.
Girls in single-sex schools take advanced physics, and to a lesser extent advanced mathematics and computing, at higher rates than girls in coeducational schools. This gap survives more of the statistical controls than the achievement gap does, and it has been replicated in several systems.
The plausible mechanism is not academic. It is that a subject's gender coding is far less visible when there is nobody in the room to code it against. A physics class that is one-quarter girls tells a girl something about physics every single lesson. A physics class that is entirely girls does not.
There is a supporting finding: coeducational schools that stream some subjects by gender report similar shifts in take-up, which suggests the mechanism is about the immediate classroom rather than the whole institution.
The non-academic outcomes
Claims about confidence, leadership and later career outcomes are common in advocacy material and thinly supported in the research.
Studies measuring self-concept find inconsistent results, with some reporting higher academic self-concept in single-sex settings and others finding the opposite — a phenomenon sometimes attributed to reference-group effects, where being surrounded by high performers lowers self-rating regardless of absolute performance.
Longer-term outcomes are very difficult to study credibly, because the confounds compound over time and the alumni networks of established single-sex schools are themselves a substantial causal factor that has nothing to do with the pedagogy.
The claim that does not survive at all
The argument that boys and girls learn differently in ways requiring separate instruction — the "different brains" version — has very weak empirical support. Reviews of the neuroscience literature find that population-level differences in brain structure are small, heavily overlapping, and have not been shown to translate into differing instructional needs.
Where single-sex education works, it does not appear to work for this reason, and programmes designed around it have not outperformed those that were not.
What to make of it as a parent or student
The honest summary is that the schooling structure is a weaker variable than the school. Teacher quality, curriculum breadth, subject availability and peer group all have better-evidenced effects than whether the classroom is mixed.
Two questions are more informative than the single-sex question. Does this school actually offer and staff the subjects that keep options open — particularly advanced physics and computing? And what proportion of its girls take them?
Those numbers are usually available on request, they vary enormously between schools of both types, and they predict more about a student's eventual options than any of the arguments that dominate the debate.