Abstract
The article explores changes in the examination performance of a random sample of 500 English secondary schools between 1992 and 2001. Using econometric methods, it concludes that: there is an overall deterministic trend in school performance but it is not stable, making prediction accuracy poor; the aggregate trend does not explain improvement over time at school level, where there is very considerable variation in improvement paths; there is a degree of persistence with respect to changes in performance at school level but it is short-lived; whilst there is evidence of a general upward trend across schools, there is a large amount of year-to-year variation and little evidence of sustained improvement at school level; and the model applied has little ability to forecast the direction of change for particular schools in the following year(s).
Notes
For reasons of simplicity we did not adopt proportional sampling of schools which would probably have enabled us to come even closer to national figures regarding the percentages of pupils achieving particular performance levels.
Because, to the best of our knowledge, this model has not previously been used in educational research, we have written a heuristic account of the Arellano and Bond GMM estimator, and of its associated diagnostic tests: this technical Appendix is available on request.
The tests of autocorrelation led to the rejection of no autocorrelation for order 1 and the acceptance for order 2, which suggests that the model is statistically well specified. The Sargan test of overidentifying restrictions led to the acceptance of the null hypothesis once lags of the independent variables used as instruments were restricted to two, which again suggests that the estimated model is statistically well specified. With three lags of the dependent variable, the Sargan test indicated that the model was mis-specified, even using only one instrument. The overall fit, given the Wald statistic, is significant at the 1%-level, which suggests that the independent variables are jointly significant with respect to variations in school performance.
Value-added data at national level first became available in 2001. Depending on the minimum number of time-points thought necessary to explore trends, the most likely dates for satisfactory data to emerge will be from around 2008 onwards.