Media Briefings

MIXED ABILITY CLASSES RAISE AVERAGE ACHIEVEMENT AND REDUCE INEQUALITY: New evidence of ‘peer effects’ in early years education

  • Published Date: June 2013

School children aged 6 and 7 who are randomly assigned to classmates with higher levels of prior achievement learn substantially more by the end of the year than those who are randomly assigned to peers with lower levels of prior achievement. That is the central finding of research published in the June 2013 issue of the Economic Journal.

Analysing data from the Student/Teacher Achievement Ratio Project (Project STAR) in the US state of Tennessee, the study by Professor Aaron Sojourner is one of the strongest yet to demonstrate the importance of peers’ prior achievement.

As policy-makers struggle with a variety of often costly options for improving educational performance, Sojourner suggests that his results add a new element to the conversation:

‘Policies that promote greater achievement mixing in early years may raise average achievement and reduce inequality.

‘But my findings are not a panacea: policies must also create benefits for higher-achieving children to keep them in the system.’

How much do peers matter? Consider Sam, a child in first grade, the equivalent of the UK’s Year 2 (age 6/7). Sojourner estimates what would happen if he were moved from a class with peers who, on average, scored better than half the kids in the United States (and worse than the other half) to a class with peers who scored better than 60% of kids in the country (and worse than only 40%).

This 10-percentile increase in peer prior achievement would be expected to raise Sam’s achievement by 3.5 percentiles at the end of first grade. In other words, by having the somewhat higher-achieving peer group, Sam is expected to learn enough extra to move up the achievement rankings past 3.5% of kids in the country.

For context, consider that two children whose own achievement levels at kindergarten (age 5/6) differ by the same amount (10 percentiles) would be expected, holding all else fixed, to perform about 8 percentiles different. The peer effect is almost half of the own-achievement ‘effect’.

Sojourner’s research also looks separately at how peers affect children with different levels of incoming achievement. Those coming into first grade with low prior achievement levels are very sensitive to differences in peer quality. Children coming in with higher levels of prior achievement appear to be less sensitive, although this finding is less certain.

Sojourner says:

‘Peer effects are notoriously difficult to nail down. If we look at a school system and see that high-achieving kids tend to be in class with other high-achieving kids, is that because of peer effects or because of other factors: school principals deciding to group high-achieving kids together in the same classes, some kids getting better teachers or some parents being able to choose better schools for their kids?’

Project STAR, which originally examined the effects of class size on children’s achievement in the mid-1980s in Tennessee, provides a laboratory for understanding peer effects.

Unlike in most schools, first-grade children in each school were randomly assigned to classrooms for the year. Therefore, each child had randomly assigned peers, as in a scientific experiment. First-grade teachers within school were also randomly assigned to classes. This allows the effect of peers to be measured accurately while ruling out alternative explanations.

It is very rare to get both random assignment and a pre-assignment measure of achievement in the same data set. Missing achievement data for a portion of the sample had previously discouraged researchers from conducting this kind of analysis of peer effects in the STAR data.

To tackle this challenge, Sojourner’s research develops new methods for analysing peer effects in the presence of some missing data. This promises to be useful in settings beyond education, such as workplace teams, health habits, marketing, neighbourhood influences and crime.

The findings of this study pertain to changes within a range with substantial achievement mixing. But the data do not include classes with ‘tracking’ – separating children into groups based on ability.


Notes for editors: ‘Identification of Peer Effects with Missing Peer Data: Evidence from Project Star’ by Aaron Sojourner is published in the June 2013 issue of the Economic Journal.

Aaron Sojourner is an assistant professor at the University of Minnesota’s Carlson School of Management (

For further information: contact Aaron Sojourner on +1-612-624-9521 (email:; Twitter: @aaronsojourner); or Romesh Vaitilingam on +44-7768-661095 (email:; Twitter: @econromesh).