Peer Effects in the Classroom

Summary of working paper 7867
Featured in print Digest

Students who are exposed to unusually low achieving cohorts tend to score lower themselves.

How can advanced economies get the biggest increase in human capital for their education dollar? That is, how productive are their investments in education? In answering these questions, one tricky problem is "peer effects": students are "good" peers if they produce positive learning spillovers, so that students exposed to them gain more for each dollar spent on their education, or "bad" peers if they have the reverse effect.

It is hard to know whether such peer effects exist, but if they do, they are crucial to current debates on which policies maximize the productivity of a country's education spending. The United States is debating school choice; European countries are discussing whether to eliminate ability tracks from their education systems; Latin American countries are debating whether to devolve control and funding of education to localities. Many arguments against school choice, decentralized funding, and ability tracking rest on the belief that peer effects are important and have a particular asymmetry: that is, bad peers gain more by being exposed to good peers than good peers lose by being exposed to bad peers. If this asymmetry is strong, then investments in human capital are maximized when students are forced to attend schools with a broad array of abilities and backgrounds. Such coercion is obviously impossible with ability tracking and can be hard to achieve with choice or local funding.

In Peer Effects in the Classroom: Learning From Gender and Race Variation (NBER Working Paper No. 7867), NBER Research Associate Caroline Hoxby tries to determine whether peer effects exist and, if they do, what form they take (for instance, are they asymmetric?) She begins by noting that true peer effects are hard to measure. Parents who provide home environments that are good for learning tend to select the same schools. Even within a school, interested parents lobby to have their children assigned to particular teachers. Thus, if high achievers tend to be clumped in some classrooms and low achievers in other classrooms, we should not assume that the achievement differences are caused by peer effects. Most of the achievement differences probably are due to parents, who would influence their children a lot even if they could not get them in classrooms with particular groups of peers.

It is not just parents' activities that make peer effects hard to measure, though; it is also schools' activities. Students with similar abilities may be assigned to the same classroom in order to make it easier to teach. Teachers with a knack for handling the unruly students may have classes full of them. Thus, classroom achievement could differ because the initial student composition differs, not because peers influence one another.

To identify true peer effects, Hoxby compares groups within a given school that differ randomly in peer composition. To illustrate: suppose that a family shows up for kindergarten with their older son and finds that, simply because of random variation in local births, that son's cohort is 80 percent female. The next year, they show up with their younger son and find that, also because of random variation, that son's cohort is 30 percent female. Their two sons now will go through elementary school consistently experiencing classrooms that have different peer composition on average. Their older son will be exposed to more female students (who tend to be higher achievers and less disruptive in elementary school). Their younger son will be exposed to more male students. Because the two boys have the same parents and the same school, the main difference in their experience will be peers. If it turns out that male students systemically do better (or worse) when exposed to more female students, then that systematic difference must be attributable to peer effects.

Hoxby also compares school cohorts that differ in racial composition or initial achievement, rather than in gender composition. She uses data from the entire population of elementary students in Texas from 1990 to 1999 (the Texas Schools Microdata Sample). Her measure of achievement is a student's score on the Texas Assessment of Academic Skills, which is administered in all Texas public schools.

Hoxby finds that peer effects do exist. For instance, her results suggest that having a more female peer group raises both male and female scores in reading and math. She points out that only some of the "good" peer effect of females can be direct learning spillovers because females do not know math better than males on average, although they are better readers. The fact that females raise math scores, therefore, must be due to phenomena more general than direct learning spillovers -- for instance, females' lower tendency to disrupt.

In Texas, black and Hispanic students tend to enter school with lower initial achievement. Does this matter? Hoxby finds that it does. Students who are exposed to unusually low achieving cohorts tend to score lower themselves. Interestingly enough, black students appear to be particularly affected by the achievement of other black students. Hispanic students appear to be particularly affected by the achievement of other Hispanic students. In fact, Hispanic students do better when in majority Hispanic cohorts, even though the additional Hispanic students tend to have lower initial achievement. It may be that in classes with more Hispanics, a student who is learning English is more likely to find a bilingual student who helps him out.

Hoxby finds little evidence of a general asymmetry, though, such as low achievers gaining more by being with high achievers and that high achievers lose by being with low achievers. After taking steps to eliminate changes in achievement that could be caused by general time trends or unusual events -- such as the appearance of an especially good teacher in one school -- Hoxby concludes that, on average, a student's own test score rises by 0.10 to 0.55 points when he or she is surrounded by peers who score one point higher.

-- Linda Gorman