As someone who has taught in and led boys’ schools for almost 30 years, dealt with thousands of parents who prefer single sex schooling and felt satisfied that we were doing a good job for the young people educated in our system, I was naturally interested in a recent study claiming that single sex education makes little or no difference to student performance and attitudes when compared with co-education.
The study, published by the American Psychological Association, is a meta-analysis of data from 184 studies drawn from the testing of 1.6 million students from 21 nations. It looks at mathematics performance and attitudes, science performance, educational aspirations, self-concept, gender stereotyping and a number of other outcomes; it purports to cover the principal comparative benefits on offer. After trying to discredit the research which has shown an advantage for single sex schooling, the authors conclude that the advantage is illusory.
Study co-author Janet Shibley Hyde claims that the benefits of single-sex classrooms “just aren’t supported by science”. “The two best predictors of kids’ school success are the parents’ education and the family income,” she says, and when you take this into account any apparent advantage from segregated schools or classrooms dwindles to insignificance. Her claims appear to focus on US data. She holds that parents opting for single sex schools tend to be middle class and more affluent than those patronising co-ed schools – leading to what is known in the social sciences as selection effects.
However, only 57 of the 184 studies analysed had controlled for  – that is, collected data on – such parental characteristics, and this is a major gripe of the authors. The other is that there were few (12) studies in which students were randomly assigned to either single sex or co-ed schools or segregated classes within mixed schools. (As an aside, there may be parents who would agree to having their child randomly assigned to one or the other, but there must be philosophical and practical difficulties in such a process.) Nevertheless the researchers were able to conclude: “There is little if any evidence of an advantage of SS schooling for girls or boys for any of the outcomes.” In other words, in so far as they rely on research, advocates of single sex schooling are seduced by poor quality studies.
I have to say at this point that I rely almost entirely on the evidence of my eyes and ears, including the characters, behaviour and academic achievements of students to judge the success of the schools I have been involved with. And I believe the same applies to parents, as well as past pupils. But the authors of the study have their sights set not on individual satisfaction but on public policy – in particular, as Americans, on the fairly recent popularity of and government support for single sex schooling in the US.
From that point of view their meta-analysis could be a game changer. But while there is no question that parental education and status are proven predictors of educational outcomes, in the USA in particular, it is altogether another matter to dismiss the single sex educational advantage.  
Two aspects of the study give me pause. The first is the nature of meta-analyses, and the second is the nature of the authors’ assumptions. Some meta-analyses are better than others Meta-analyses are not without critics amongst academics themselves. They have been denounced by various authors as “an exercise in mega-silliness, “shmeta analysis” and “statistical alchemy for the 21st century”. In 1990 meta-reviewers of a medical study were even tagged as an obstetrical “Baader Meinhof gang”.
Study design appears to be a thorny problem. Mark Russo, an expert in meta-analysis writes “If it is well conducted, the strength of a meta-analysis lies in its ability to combine the results from various small studies that may have been underpowered to detect a statistically significant difference.” He insists that “the study design of a meta-analysis determines the validity of its results.”
Here the opposite is the case; from individual studies which frequently had detected an advantage for single sex education (for example in the middle years), the combined analysis fails to find a significant result (a fact which could even indicate that a meta study was poorly designed). On the strength of their conclusions the authors then offer reasons for why advocates of single sex education are wrong. I find it surprising that such an approach was not questioned when this study was itself peer-reviewed.
Furthermore the study appears to fall short of best practice standards for meta-analyses. Flaws include inadequate coding methods (71% coded by one author only); apparent failure to correct for inherent bias in some studies; the inclusion of a high proportion (more than 1/3) of unpublished and therefore not peer reviewed studies, inflation of the size of the total of participants by multiple analyses of the same subjects for different purposes, and the lumping together of very different countries.
Professor Shibley Hyde and colleagues themselves make much of the paucity of randomised and controlled studies on the subject, indicating that this seriously limits the effectiveness of their analysis. Therefore it is quite surprising, and inconsistent, that they come to unequivocal conclusions about the effects of single sex schooling, conclusions which unfortunately have been stridently echoed in commentaries on the article. Gender theories My second reservation concerns the theoretical aspect of this study.
Two of the authors are active in gender studies; indeed, Professor Shibley Hyde is widely recognised for her work in this field and has invested much of her academic career arguing that men and women are much more alike than is popularly supposed. An a priori tenet of gender studies is that gender differences are primarily a social construct.
In contrast, other academics in psychology and neuroscience have built just as illustrious careers harnessing evidence for essentially the opposite view, that gender differentiation is hardwired in human nature. This is a view underpinning much single  sex schooling.
So there must be grounds to question the rigour of a comparative meta-analysis conducted by these authors who chose the studies, selected the categories, coded the articles, further excluded some again, all based on their opinion, and then reported on the things they wished to report, but not on all the findings that were possible, nor even all of the influences (aspects of culture) that were contained in the studies. Surely it would detract from a meta-analysis of climate change by advocates of climate change if, to some extent, coding and selection were to be based on their own or predominantly single opinions.  So too this study appears problematic.
The authors of the study under discussion state at the outset that it is “theory driven”. They theorise that social factors which make gender salient will lead to greater gender stereotyping. Strangely enough, though, when they analysed the controlled studies they found the opposite effect among girls: they were more likely to be gender stereotyped in co-ed classrooms. (There was not enough data to analyse this effect among boys.)
On the other hand, the “girl power” theory espoused by some single sex educators (girls will do better at maths and science when not dominated by boys, as in co-ed settings) is not supported by the Shibley Hyde study, the authors report. The data show “only trivial differences from girls in co-ed schooling”; neither were the segregated girls’ educational aspirations any higher or their self concept more positive. (But they weren’t any worse.)
Similarly, the theory that biological differences are significant, so that both girls and boys will do better when taught by methods appropriate to their gender, is dismissed by the authors. They say the controlled studies showed “no substantial advantages”. In other words, students did seem to do better in segregated schools but you could explain the difference away by selection effects.
There was little evidence, however, for another of the authors’ pet theories. The “expectancy-value theory” holds that single sex schools make girls, for example, more aware of existing segregation in adult occupations involving maths and science, and imply a certain value in such arrangements, thus discouraging girls’ ambitions in those areas.
The authors seem convinced, though, about the dangers of gender stereotyping in single sex schooling and want future studies to gather information on schools’ underlying assumptions about gender – specifically, beliefs about “large biological differences” and “girl power”. “Only then,” they say, “can we determine whether the effects of SS programs depend on the messages that are conveyed to teachers, students and parents.”
Well, of course the effects of schooling depend in part on “messages” about things other than the academic content of lessons. And these messages concern more than gender. In the schools of my acquaintance, for example, parental involvement, the behaviour code and mentoring practices give both parents and students messages about freedom, authority, values and virtues. These things should also be measured in order to determine the value added by single sex education.
In conclusion, make no mistake about it, the authors of this study also have a message they want schools to propagate, the message that gender is a social construct with very little to do with biology. And that is a message that should make parents very choosy about schooling for their children. Dr Andrew Mullins is headmaster of Wollemi College in Sydney and author of Parenting for Education (Finch 2005). His doctorate investigated the neural bases of character. The author is grateful to Dom McLaughlin, UNSW lecturer in Psychology, for his statistical observations.