Can We Do Better than Using 'Mean GCSE Grade' to Predict Future Outcomes? An Evaluation of Generalised Boosting Models

A key predictor of young people's future outcomes is their level of academic achievement whilst at school. In England this is most commonly measured by achievement in GCSEs. However, not all pupils will have taken the same set of GCSE examinations as, for example, they may make different subject choices. For this reason, GCSE performance is often aggregated into a simple measure such as "mean GCSE grade" before being used in statistical models. This paper investigates the merits of using an alternative method, based upon the relatively new technique of Generalised Boosting Models, which does not require for GCSE results in different subjects to be aggregated together. The importance of this research is that by evaluating the predictive performance of such a method we can ascertain how much useful information is lost in the process of GCSE aggregation. The results show that traditional predictions based upon simple aggregated measures of GCSE attainment are fairly similar to those based upon the more complex approach. This provides some confidence that, for the majority of outcomes, only a small amount of predictive information will be lost through the use of aggregated measures of GCSE performance.