Inequality in Academic Performance and Juvenile Convictions: An Area-Based Analysis

Abstract

This paper focuses on the links between inequality in academic performance and juvenile conviction rates for violent crime, stealing from another person, burglary in a dwelling and racially motivated offences. We use area-based aggregate data to model this relationship. Our results show that, above and beyond impacts of absolute access to resources, young people who grow up in school cohorts marked by higher levels of disparity in educational achievement may be more prone to commit violent crime and racially motivated offences than those with less disparity. This association is however not found for property-related offences. Our results further show that higher between-schools inequality and higher within-school inequality are both associated with higher conviction rates for violent crime and racially motivated offences.

Keywords:

• education inequality
• crime
• England

Acknowledgements

We would like to thank research staff at the Centre for Research on the Wider Benefits of Learning for their useful comments on this paper and Rachel Barker of the Department for Children, Schools and Families (DCSF) for helpful suggestions. Dr Bob Baulch from the Institute of Development Studies, University of Sussex, and Professor Michael Carter from the University of Wisconsin-Madison provided helpful insights into the use of inequality indices.

We would particularly like to thank Vicenzo Lombardo (University of Napoli) for his initial review of the literature for this paper.

Finally, we would also like to thank the DCSF for their financial support of this project and for making the data available.

Notes

¹ We use the term ‘local education authority’ here as LEAs were the official entities which existed at the time when the data were compiled. LEAs have since been aligned with local authorities and are not now separately identified.

² Classification of LEAs changed between 1997 and 1998: there were 109 LEAs in England in 1996, and there were 150 in 2006. Large LEAs were divided into smaller authorities, in some cases up to four or five smaller units. We standardised the classification of LEAs based on areas being contiguous or magistrates sharing its legislation across two areas and were left with 133 LEAs. All statistics for LEAs that changed during this period were recalculated.

³ The Theil index satisfies the Pigou-Dalton Transfers Principle, which indicates that if a student from a lower part of the test scores distribution gets lower grades and a student in the higher end of the distribution gets higher grades and all other grades remain unchanged, then the numerical value of the inequality index should increase. The other properties satisfied are symmetry and homogeneity. Symmetry indicates that changing the distribution from higher achievers to lower achievers and vice versa will not change the measurement of inequality, whereas homogeneity indicates that if all test scores increase by the same amount, the measurement of inequality should remain unchanged. In addition, it is possible to decompose the Theil index into the between (or across) and the within components of the distribution (the term ‘between’ and ‘within’ refer to schools in our case).

⁴ A different approach could have been to place random effects at the intercept and slope in order to capture within-cohort, between-LEA variability in the average conviction rates and within-cohort, between-LEA variability in trends, respectively. However, the preference for fixed effects model is based on the fact that fixed effects fully account for differences due to cohort membership, which is not necessarily captured by random effects (Pinheiro and Bates, 2000).

⁵ Failure to incorporate the longitudinal and multilevel aspects of the data, i.e. repeated observations for cohorts within LEAs, would lead to an inflated estimate of the residuals. In other words, we would be incorporating known information as if it were unobserved and random. Furthermore, given the large size of LEAs, which supports the assumption that juvenile crime is not committed across LEAs boundaries, the error term is assumed uncorrelated across LEAs, i.e. no spatial correlation.