Geoffrey Woodhouse

School Effectiveness Research and Educational Policy

This paper discusses a series of recent critiques of school effectiveness (SE) research from within the academic community and the responses to them by SE researchers. It uses these as a basis to explore the nature of current SE research and its relationship with government policy. It is argued that much SE research has been too closely concerned with specific government policies as well as having weak theoretical and empirical support. The general response of the SE community to these criticisms is judged to be inadequate and recommendations for future directions are made.

Educational Performance Indicators and LEA League Tables

Abstract A common procedure for using examination results as performance indicators is based upon residuals from regression analysis. These analyses are typically applied to aggregated data: this paper demonstrates that such procedures give unstable results. It is suggested that aggregate‐level analyses are uninformative and that useful comparisons cannot be obtained without employing multilevel analyses using student‐level data.

Adjusting for measurement error in multilevel analysis

Measurements in educational research are often subject to error. Where it is desired to base conclusions on underlying characteristics rather than on the raw measurements of them, it is necessary to adjust for measurement error in the modelling process. In this thesis it is shown how the classical model for measurement error may be extended to model the more complex structures of error variance and covariance that typically occur in multilevel models, particularly multivariate multilevel models, with continuous response. For these models parameter estimators are derived, with adjustment based on prior values of the measurement error variances and covariances among the response and explanatory variables. A straightforward method of specifring these prior values is presented. In simulations using data with known characteristics the new procedure is shown to be effective in reducing the biases in parameter estimates that result from unadjusted estimation. Improved estimates of the standard errors also are demonstrated. In particular, random coefficients of variables with error are successfully estimated. The estimation procedure is then used in a two-level analysis of an educational data set. It is shown how estimates and conclusions can vary, depending on the degree of measurement error that is assumed to exist in explanatory variables at level 1 and level 2. The importance of obtaining satisfactory prior estimates of measurement error variances and covariances, and of correctly adjusting for them during analysis, is demonstrated.

Progress from GCSE to A and AS Level: Institutional and gender differences, and trends over time

The authors study the relationship between results obtained in examinations for the General Certificate of Education at Advanced and Advanced Supplementary (A/AS) level and those obtained by the same students two years earlier in examinations for the General Certificate of Secondary Education (GCSE). They used comprehensive data on four cohorts examined between 1994 and 1997 to build a multilevel, longitudinal model of student progress. It was found that progress differs between males and females, and between students of different ages, and that the average GCSE performance of the students in an establishment is a significant predictor of individual progress. Once establishments are matched on this measure, and students are matched on their own GCSE performance, the effects of most establishment types are substantially reduced: in particular, the average progress of students in maintained grammar schools does not differ significantly from that of students in maintained comprehensive schools. Less stabilit...

Multilevel Models in the Study of Dynamic Household Structures

A modelling procedure is proposed for complex, dynamic household data structures where households change composition over time. Multilevel multiple membership models are presented for such data and their application is discussed with an example.