Prediction of Performance in Standardised Assessments from Computer-Based Formative Assessment Data

Overview

This study investigates the predictive validity of computer-based formative assessment data for standardised summative assessment outcomes. The research addresses the complementary roles of formative and summative assessments in educational contexts, examining whether objective, low-disruption formative assessment data collected throughout compulsory schooling can reliably forecast high-stakes standardised achievement measures. The investigation leverages large-scale longitudinal data spanning multiple assessment timepoints to establish empirical relationships between these assessment modalities.

Methods and approach

A large sample of students was evaluated at multiple time points across compulsory schooling using both computer-based formative assessment systems and standardised summative assessments. Regression models were systematically trained to predict summative assessment abilities using various feature subsets derived from formative assessment data and auxiliary variables. The analysis compared model performance across different feature configurations and competence domains, with particular attention to domain-specific alignment between predictive features and target outcomes. Model diagnostics included variance accounted for and systematic bias analysis relevant to decision-making applications.

Results

A model incorporating mean abilities across different competence domains demonstrated optimal predictive performance, accounting for 30-48% of variance in summative assessment outcomes. However, this predictive capacity remained substantially below the variance explained by prior summative assessment measures. Predictive formative assessment features exhibited domain-specific alignment, with features from the same or similar competence domains as the target summative outcome demonstrating the strongest associations. Systematic model biases were identified across conditions, indicating that direct application of these models for high-stakes decision-making would require consideration of these systematic deviations.

Implications

The findings establish empirically that formative assessment data contains meaningful predictive information regarding future standardised achievement, though with substantial unexplained variance. This relationship enables earlier detection of learning progress trajectories, providing teachers with quantifiable evidence to adapt instructional strategies before summative assessment points. The domain-specific predictive patterns suggest that formative assessment systems should maintain competence-specific tracking rather than relying on general performance indicators.

From a policy perspective, these results provide a quantitative basis for reconsidering assessment architectures that rely heavily on high-stakes summative testing. The capacity to predict summative outcomes from formative data, combined with the reduced student stress and ecological validity advantages of computer-based formative assessment, suggests potential for reducing reliance on traditional summative assessments in some contexts. However, the systematic biases identified necessitate careful calibration if such predictive models are used to inform consequential educational decisions.