The White British–Black Caribbean achievement gap: tests, tiers and teacher expectations

Abstract

A recent analysis of the Longitudinal Study of Young People in England (LSYPE) indicates a White British–Black Caribbean achievement gap at age 14 which cannot be accounted for by socio‐economic variables or a wide range of contextual factors. This article uses the LSYPE to analyse patterns of entry to the different tiers of national mathematics and science tests at age 14. Each tier gives access to a limited range of outcomes with the highest test outcomes achievable only if students are entered by their teachers to the higher tiers. The results indicate that Black Caribbean students are systematically under‐represented in entry to the higher tiers relative to their White British peers. This gap persists after controls for prior attainment, socio‐economic variables and a wide range of pupil, family, school and neighbourhood factors. Differential entry to test tiers provides a window on teacher expectation effects which may contribute to the achievement gap.

Acknowledgements

This research was funded by the Department for Children, Schools and Families (DCSF) (Project Reference EOR/SBU/2006/031).

Notes

1. In the USA many high schools place students in different ‘tracks’ that offer academic classes to higher achieving students and general or vocational tracks to lower achieving students. Setting or ability grouping in the UK is similar, except that in theory a student might be placed in a top set in one subject but a lower set in another subject, although in practice students tend to be placed in similar sets across different subjects (e.g. Hallam, 2002).

2. For a discussion of the benefits of tiering see Burghes et al. (2001).

3. Some authors consider any evidence of disproportionate representation in ability groups or tiers as prima facie evidence of bias, which Ferguson (1998) has defined as unconditional bias. In contrast, this study adopts a definition of bias as disproportionate representation in ability groups or tiers after control for legitimate objective measures of performance such as past attainment or grades, which Ferguson has defined as conditional bias. In common with Ferguson (1998, p. 280), I believe the latter is more appropriate if seeking evidence of bias in allocating students to tiers.

4. An initial analysis used age 11 science test marks as the measure of prior attainment. However, a significantly better correlation was achieved with average age 11 test mark (r = 0.63) than with age 11 science mark alone (r = 0.56). The total test marks obtained by each student across all age 11 tests were summed (range 0–280) and subject to a normal score transformation so the mean age 11 test score is represented by zero with a standard deviation of 1.

5. All variables described in Appendix 1 were included in the analysis but variables that were not significantly associated with entry tier (non‐significant WALD test) were removed through a process of backwards elimination. This was important in order to produce a parsimonious model because including a large number of redundant variables led to complete or quasi‐complete separation in the data.

6. This single odds ratio makes the assumption that the regression parameters are equal for all cumulative responses (tiers). For each model the WALD statistic was computed to test this null hypothesis against a model with variable parameters. The assumption of constant parameters held for the full contextual model but was not met for some of the simpler models. In these cases multinomial regressions which do not assume constant parameters were run and showed consistent direction of effects associated with ethnic group, so for simplicity of presentation the cumulative OR is reported here.

7. Age 11 mathematics test marks and age 11 average test marks were equally strong predictors of age 14 mathematics tier of entry (both r = .81). Given this functional equivalence in terms of prediction, age 11 mathematics test marks were used as the control because the test content was more closely related to the age 14 mathematics tests. Age 11 mathematics test marks (range 0–100) were normal transformed to have a mean of 0 and SD of 1.

8. This point has been misinterpreted in Gillborn (2010), primarily because he adopts an unconditional rather than conditional definition of bias (see footnote 3).

9. Using statistical methods to control for contextual variables within a tier is problematic because of the substantial drop in sample size.

10. While the order for test papers in November/December does not commit schools to entering any individual student for a particular tier, they are told that ‘schools’ orders should be as accurate as possible, as there is very limited time for processing and fulfilling late orders and correcting any shortfalls in ordering’ (Qualificationa and Curriculum Authority, 2004, p. 32).