Elsevier

Economics of Education Review

Volume 66, October 2018, Pages 154-166
Economics of Education Review

Achievement effects from new peers: Who matters to whom?

https://doi.org/10.1016/j.econedurev.2018.07.008Get rights and content

Highlights

  • Educational achievement peer effects are estimated for Australian school students

  • The transition from primary to secondary school is exploited for identification

  • Peer effects estimates are based only on peers from different primary schools

  • Small but positive and statistically significant average peer effects are found

  • There is also a shining light effect which is strongest for middle achieving students.

Abstract

This paper presents estimates of achievement-related peer effects on school students’ literacy using data from national test scores, across multiple literacy measures and student cohorts, for the population of public secondary school students in Years 7 and 9 (aged 12/13 and 14/15 years) in the Australian state of Victoria. Identification is achieved via individual fixed effects and by distinguishing between secondary school peers who attended the same primary school as the individual and those who did not. Estimates of peer effects are based on the new peers, whose primary school achievement could not have been affected by the individual. The results provide strong evidence for the existence of peer effects, with small but positive and statistically significant effects from having higher-achieving peers on average and from having a higher proportion of very high-achieving peers. Further, it is individuals in the middle of the ability distribution who benefit most from having high achieving peers.

Introduction

Peer effects refer to externalities in which the actions or characteristics of a reference group affect an individual's behaviour or outcomes. They have been studied in numerous contexts. This paper examines a specific form of peer effects, related to the effect on a student's achievement of the achievement of his or her peers, which if the effects are of sufficient magnitude, have critical implications for students, parents, schools and policy makers. Effects from peer averages imply that parents can improve their child's expected educational achievement by selecting a school with a higher-ability intake, and reforms introducing greater school choice could widen educational inequalities. Non-linear peer effects have additional efficiency implications, such that schools may be able to improve the average achievement of their students, and policy makers the efficiency of the schooling system as a whole, by manipulating the allocation of students across classes or schools. As a result, peer effects in school achievement have attracted an enormous amount of attention in the literature. Establishing the existence and magnitude of peer effects, however, is beset by practical difficulties (Angrist, 2014, Manski, 1993, Manski, 2000, Moffitt, 2001), a fact that has probably also contributed to the longevity and ubiquity of the literature. Despite the vast number of studies, disagreement as to the nature, magnitude and even the existence of peer effects in school achievement remains.

The two key difficulties in estimating achievement peer effects arise from endogenous sorting into or within schools and the fact that the individual is a peer of their peers and may therefore influence their peers – the reflection problem. One branch of the school achievement peer effects literature has attempted to address the second challenge by utilising measures of peer quality that pre-date any potential social interactions between the individual and their peers. A promising new strand of this literature exploits the transition between primary and secondary school (so far only in England), and the fact that most of the secondary school peers of any individual attended a different primary school (Gibbons and Telhaj, 2016, Lavy et al., 2012, Mendolia et al., 2018, Zhang, 2016). Specifically, these papers use the prior performance of peers when they were in different schools – a measure that is immune from reflection problems – to measure the impact of peers’ earlier achievement on current student performance. Lavy et al. (2012) also exploits multiple measures of individual student achievement in various learning domains to remove any fixed student-specific effect on achievement, which mitigates confounding effects from sorting into schools, making for a very strong identification strategy. They find no evidence of an average peer effect on individual achievement. Gibbons and Telhaj, 2016, Zhang, 2016 and Mendolia et al. (2018), however, all find some evidence of small, positive average peer effects. All four studies use data drawn from the same underlying data set; the English National Pupil Database.

The research design adopted in this paper also exploits this transition, when students encounter new classmates at the start of secondary school who did not attend the same primary school. The secondary school achievement of individuals is regressed on the primary school achievement of their secondary school new peers, for whom there is no reflection problem. The approach also makes use of multiple measures of individual achievement, specifically a set of four language or literacy-related measures of achievement. This enables use of an individual fixed effects approach to account for common individual effects on achievement across the four language-related subjects of reading, writing, grammar and spelling. Because the data cover multiple cohorts of students, school-subject fixed effects can also be included to wash out time-invariant correlated effects that might differ across subjects. In other words, peer effects are estimated by exploiting within-individual variation in the performance of new peers across the literacy domain when they were in primary school.

In contrast to the other papers in this strand of the literature, however, this paper uses data from Australian schools rather than English schools and focuses on outcomes at age 12/13 rather than outcomes at age 13/14 or later. In doing so the paper makes a standalone contribution to the knowledge base supporting education policy makers by shedding light on how well the conclusions of these earlier English studies generalise to a similar but not identical educational context. Its single most important contribution, however, is to present evidence of peer effects, including small but non-trivial average peer effects, using a strong identification approach exploiting within-individual variation in test scores. A new placebo test is also added.

Specifically, this paper estimates peer effects using administrative data on the test scores of students attending public schools in the Australian state of Victoria. The data come from the National Assessment Program – Literacy and Numeracy (NAPLAN) conducted across Australia, which provides test scores for five subjects – numeracy, reading, spelling, grammar, and writing – for all students in grades 3, 5, 7, and 9. The national testing system which generates these data was introduced across Australia in 2008, with tests taking place each year in the specific grades, and the paper exploits data from 2008 to 2013. This provides four cohorts of Year 5 (primary school) students subsequently observed in Year 7 (in secondary school), and two cohorts with students also observed in Year 9.

In addition to estimating average peer effects, this paper also assesses the impact of having a high proportion of peers who were in either the top or bottom 10% of the achievement distribution, while also allowing these effects to differ between genders and across the distribution of individual ability (that is, allowing for non-linear and heterogeneous peer effects). This allows testing of empirical support in Australian secondary schools for a wide range of models including the ‘bad apple’ (disruptive students harm everyone), the ‘shining light’ (excellent students provide a great example for all), the ‘invidious comparison’ (outcomes are harmed by the presence of better achieving peers), and ultimately to assess the possible benefits of tracking, at least at the school grade level (see Sacerdote, 2011).

The remainder of this paper is set out as follows. The next section briefly reviews those studies in the school achievement peer effects literature to which this paper most closely speaks. The data are described in more detail in Section 3, along with further discussion of the approach to estimation. The main results are discussed in Section 4, with extensions in Section 5 and concluding discussion in Section 6.

Section snippets

Related literature

Recent studies on peer effects in school achievement have looked for both average peer effects and peer effects from and to different points in the ability distribution, using a variety of strategies to overcome the identification problems associated with reflection and endogenous sorting. One strand of this literature exploits random or quasi-random assignment to new peer groups stemming from programs like Metco in the US (Angrist & Lang, 2004) or the Extra Teacher Program in Kenya (

Data and identification

The Australian testing system, NAPLAN, was introduced in 2008, with tests taking place each year in grades 3, 5, 7 and 9. Test score data from 2008 to 2013 are available for the population of students in the relevant grades in public schools in the state of Victoria. This provides four cohorts of Year 5 (primary school) students who are subsequently observed in Year 7 (in secondary school) from 2010 to 2013. In addition, there are two cohorts with students observed in Year 9 in 2012 or 2013.

Homogenous peer effects in year 7

Table 4 contains parameter estimates on the new peers variables included in Eq. (1) for Year 7 students (OLS equivalents – without individual student fixed effects – are given in Appendix Table 1). The first three columns show the parameters on the average, bottom 10% and top 10% new peer effects when each of the peer variables are entered singly in Eq. (1), and without controlling for old peers or any individual prior achievement scores. Each subsequent column in Table 4 corresponds to a

Heterogeneity in the year 7 peer effects

Studies have examined evidence for heterogeneous peer effects by characteristics like gender, race and ability in addition to homogenous peer effects. Table 7 contains parameter estimates on the new peers variables for males and females separately and across the distribution of prior school achievement for males and females jointly and separately (using specification 6 of Table 4). Table 7 panel A shows that the average new peer effect varies little across the distribution of prior achievement,

Discussion

This paper provides evidence about the existence of achievement-related peer effects in the first year of secondary schooling. It exploits measurements of peer quality prior to any social interaction and uses individual and school-subject fixed-effects for identification. This approach controls for common individual factors that influence achievement across subjects and domains, such as ability, motivation and background factors, along with school by subject effects common across individuals.

Acknowledgements

This research was commissioned by the Victorian Department of Education and Training (DET). This paper uses unit record data from the NAPLAN administrative data collection, provided by DET. Thanks also to seminar participants at University ErlangenNuremberg, Melbourne Institute, University of Tilburg, University of Sydney, ESAM, and University of Zurich for comments and suggestions. The findings and views reported in this paper are those of the authors and should not be attributed to DET or any

References (27)

  • T.S. Dee

    A teacher like me: Does race, ethnicity, or gender matter?

    American Economic Review

    (2005)
  • E. Duflo et al.

    Peer effects, teacher incentives, and the impact of tracking: Evidence from a randomized evaluation in Kenya

    American Economic Review

    (2011)
  • H. Forgasz

    Streaming for Mathematics in Years 7-10 in Victoria: An Issue of Equity?

    Mathematics Education Research Journal

    (2010)
  • View full text