In the paper, Hill investigated the effect of the share of opposite gender friends within students' friendship networks (rather than the share of girls only). One problem with this sort of analysis is that students choose their friends, and parents choose the schools that their children attend. So, there is likely to be some selection bias involved - if you simply compared students with more opposite gender friends with students with fewer opposite gender friends, you can't tell if any differences you observe are because of the share of opposite gender friends, or because of the selection bias. Hill overcomes this selection problem using instrumental variables analysis (which I have earlier discussed here): he essentially finds some variable that is expected to be related to the proportion of opposite gender friends, but shouldn’t plausibly have a direct effect on academic outcomes, and uses that variable in the analysis. In this case, he uses as an instrument the proportion of opposite gender schoolmates in their close neighbourhoods. As he says:
Students with more opposite gender schoolmates in their close neighborhoods have more opposite gender school friends, and, given that the gender composition of schoolmates in a student’s neighborhood is essentially random, provides plausibly exogenous variation in the share of opposite gender friends from which a causal effect on the outcomes of interest can be estimated.The hypothesis being tested is that having more opposite gender friends decreases academic performance, through two potential mechanisms:
First... individuals in class may distract or be distracted by opposite gender friends more than same gender friends, reducing the quality of classroom inputs for individuals with a greater share of opposite gender friends...
Second... higher shares of opposite gender friends may increase the returns to leisure and therefore increase the time spent socializing at the expense of studying.So, opposite gender friends are hypothesised to be a distraction to both in-class learning and out-of-class studying time. Hill finds that:
...an increase in the share of opposite gender school friends reduces academic achievement... A standard deviation increase in the share of opposite gender friends causes a half standard deviation reduction in GPA scores.That effect is relatively large - the standard deviation of GPA in his sample is 0.8 (on a four-point GPA scale). However, when disaggregating by gender he finds that the effect is only statistically significant for girls. In other words, having a higher share of opposite gender friends is bad for girls, but possibly not so for boys. However, there is one part of his results that I do not fully agree with:
Results also indicate that opposite gender friends increase the probability of the student being in a romantic relationship, which may have adverse effects on achievement.He uses the same instrumental variable approach to investigate the effects of the share of opposite gender friends on the probability of having been in a relationship in the past 18 months. Unfortunately, the instrument is not as valid in this case, as there is a plausible direct relationship between the share of opposite gender classmates living in the close neighbourhood and the probability of being in a relationship. Increased opportunity for interaction with opposite gender classmates (as would occur if there are more of them in the neighbourhood) increases the chances of starting a relationship (for at least some students!). So, while the results are plausible they are not necessarily causal.
Anyway, that doesn't take away from the overall results of the paper, which is that it is better for girls to have a higher share of female friends (and a smaller share of male friends), in terms of academic performance at high school. Which adds to the argument for the advantage of single gender schools (at least for girls), or for introducing single gender classes within coed high schools.