Building Optimal Predictive Models with Large Scale Assessment Data

Applying Bayesian statistical interference Professor Kaplan investigates the purpose and predictive validity of the PISA study. He finds that to develop PISA as an optimally predictive model, PISA would have to be considered as an indicator system; the important outcomes of education (cognitive or non-cognitive) would have to be defined; country-sensitive indicators would need to be developed; predictive models would need to be sensitive to changes in policy-relevant indicators; and prediction should be optimised by recognising and accounting for uncertainty every step of the way. “