Girls, boys, and high achievers
How does the presence of high-achieving peers affect long-run outcomes of their schoolmates? Does gender play a role? The majority of the literature has been concerned with these questions separately. One strand of literature has focused on how the gender composition of a class/grade/group/team affect an outcome whereas another has explored the effects of peer ability (see Sacerdote 2014 for an extensive review of these literatures). Recent experimental studies suggest that women’s competitive performance – for example, the decision to enter a tournament – is affected by the gender composition of the team or of the competition. For example, women are more likely to compete when potential opponents are female (Niederle et al. 2013), and a woman’s belief about the probability that she answered a question in a field correctly is affected by the field’s stereotype (e.g. maths versus English), especially if her team partner is male (Bordalo et al. 2018).
In our recent paper (Cools et al. 2019), we ask whether there are long-term consequences to attending a high school with a larger or smaller number of female or male ‘high achievers’. We are particularly interested in exploring the long-run educational effects of this exposure. To do this, we use data from a recent cohort of high school students in the US from the National Longitudinal Survey of Adolescent Health (Add Health). Add Health was designed to enable researchers to study the impact of the social environment (factors such as friends, family, school, and neighbourhood) on adolescents’ behaviour. Data has been collected on students in grades 7-12 from a nationally representative sample of roughly 130 private and public schools in 1994-95, with follow-up surveys on a randomly chosen subgroup of students in each grade in subsequent years. The last wave (2008) contains information on educational attainment and other outcomes when the individuals are between 26 and 32 years old.
The assessment of a causal effect of exposure of students to high achievers requires a measure of achievement that is determined prior to interactions that occur at school (since, for example, a high-ability student may lead others to study more generating a correlation between own and peer performance). Because measures of a student’s achievement such as GPA are highly correlated with their own parents’ educational attainment, we use high levels of parental education as a proxy for high achievers. More specifically, we define a student as a ‘high achiever’ if either parent has a post-college education. A given student’s degree of exposure to ‘high-achieving’ males or ‘high-achieving’ females is then measured at the grade level as the fraction of males (respectively, females) in that student’s grade who have a post-college parent. In this calculation, the parents of the given student are left out of both the numerator and denominator.
Because the Add Health survey contains information on multiple cohorts (or grades) within each school, our identification strategy exploits the quasi-random variation across cohorts/grades of these shares of high achievers within the same school. That is, we are using the fact that within the same school there will be accidental variation across grades in how many high-achieving boys or girls are present.1
To study the effect of the exposure to high achievers, we regress the outcome in adulthood as a function of our main variables, MF (the fraction of male peers with post-college parents) and FF (the fraction of female peers with post-college parents), controlling for grade and school fixed effects. We include a large set of individual and peers’ characteristics, such as the individual’s age, race/ethnicity, an indicator for foreign born, the student’s score on the Peabody Picture Vocabulary Test (PVT), log household income, parental education, and whether the mother/father is in the household.2 The peer characteristics included are the percent female, the percent of each race/ethnicity, and percent who are foreign born in the student’s grade. Lastly, we also include a school linear time trend to allow for neighbourhood or school-choice changes over time.
Figure 1 shows the coefficients and the 95% confidence interval from the regression described above in which the outcome is whether a student went on to complete a bachelor’s degree in a four-year college, with the coefficient for girls in red and those for boys in blue. The regression is run separately for girls and boys. The coefficients on the key peer variables – MF and FF (the student’s fraction of ‘high-achieving’ male peers and female peers, respectively) – are insignificantly different than zero except for the case of MF for girls. That negative coefficient indicates that being exposed to a higher fraction of ‘high-achieving’ boys lowers the likelihood that a girl went on and completed a bachelor’s degree some 14 years later. The size of the effect is large: an increase of one standard deviation (net of fixed effects and time trend, this is 2.0 percentage points) in the percent of high-achieving boys decreases the probability of obtaining a bachelor’s degree by about 2.2 percentage points. Perhaps surprisingly, there is little effect of ‘high-achieving’ girls on either boys or girls.3
Figure 1 Coefficient estimates on high achievers for college
Note: Figure shows 95% confidence interval for coefficient estimates.
We investigate heterogeneous effects and find that the strongest negative impact of ‘high-achieving’ boys is on girls in the lower half of the ability distribution (as measured by their PVT score) and for those with at least one parent with some type of college degree. Furthermore, greater exposure to ‘high-achieving’ girls has a positive effect on the probability of achieving a bachelor’s degree for girls in the lower half of the ability distribution and those without a college-educated parent.
What mechanisms could be responsible for the negative effects on girls’ long-run educational outcomes? We explore this issue by using a variety of questions on confidence and aspirations and risky behaviour, asked in the first wave of the survey (i.e., while the students were in high school). Using factor analysis, we create an ‘index’ of self-confidence and aspirations and two indices of risky behaviour (with the first one reflecting mostly drinking behaviour and the second reflecting physical fights and arrests). We also create an indicator variable for whether the individual has a child before he or she turned 18. As shown in Figure 2, which plots the coefficients on MF and FF for girls and boys for each of these indicators, we find that girls exposed to a greater fraction of ‘high-achieving’ boys have significantly lower confidence/aspirations, are more likely to engage in some forms of risky behaviour (drinking), and are more likely to have a child before age 18. The impacts of ‘high-achieving’ girls on other girls are mixed, as they decrease some types of risky behaviour (drinking) but increase others (fighting). The results for boys are very different. There is no significant effect of high-achieving male peers on boys, but greater exposure to ‘high-achieving’ female peers decreases their risky behaviour (both indices) and the probability of teenage fatherhood.
Figure 2 Coefficient estimates on high achievers for self-confidence and risk
Note: Figure shows 95% confidence interval for coefficient estimates.
Our study shows that ‘high-achieving’ boys have a negative and persistent effect on girls’ longer-run education outcomes. Although we cannot say whether this arises because, for example, teachers pay more attention to these boys (and consequently less attention to girls) or because these boys have a direct effect on girls, the results suggest that the latter’s self-confidence/aspiration is negatively impacted and that they have a greater tendency to engage in risky behaviour including having a child before the age of 18.
Bordalo, P, KB Coffman, N Gennaioli and A Shleifer (2018), “Beliefs about Gender”, Working Paper.
Cools, A, R Fernández and E Patacchini (2019), “Girls, Boys, and High Achievers”, CEPR Discussion Paper No. 13754.
Feld, J and U Zölitz (2018), “Peers from Venus and Mars: Higher-Achieving Men Foster Gender Gaps in Major Choice and Labor Market Outcomes”, Working Paper.
Hoxby, C (2000), “Peer effects in the classroom: Learning from gender and race variation”, NBER Working Paper No. 7867.
Lavy, V and A Schlosser (2011), “Mechanisms and impacts of gender peer effects at School”, American Economic Journal: Applied Economics 3(2): 1-33.
Mouganie, P and Y Wang (2017), “High Performing Peers and Female STEM Choices in School”, Working Paper.
Niederle, M, C Segal and L Vesterlund (2013), “How costly is diversity? Affirmative action in light of gender differences in competitiveness”, Management Science 59(1): 1-16.
Olivetti, C, F Patacchini and Y Zenou (2018), “Mothers, Peers, and Gender-Role Identity”, Journal of the European Economic Association.
Sacerdote, B (2014), “Experimental and Quasi-Experimental Analysis of Peer Effects: Two Steps Forward?”, Annual Review of Economics 6(1): 253-272.
 Hoxby (2000), Lavy and Schlosser (2011), and Olivetti et al. (2018), among others, use the same identification strategy.
 The Peabody Picture Vocabulary Test is widely used to measure verbal ability and has been shown to be positively correlated with IQ.
 The negative effects of ‘high-achieving’ boys on girls are in line with those found by Mouganie and Wang (2017) in the context of high-school students in China and Feld and Zölitz (2018) in the context of a Dutch business school. Our data allows us to show that the negative effects start in high school and have important long-run consequences.