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

Area of Science:

• Educational Psychology
• Assessment Science

Background:

• Summative assessments (SAs) measure knowledge post-instruction, often in high-stakes settings.
• Formative assessments (FAs) guide instruction and feedback during learning.
• Computer-based FAs (CBFAs) offer objective, low-disruption data collection mirroring real-world behavior.

Purpose of the Study:

• To investigate the predictive power of formative assessment (FA) outcomes on summative assessment (SA) outcomes.
• To explore how effectively FA data can forecast student performance in high-stakes evaluations.
• To inform educational practices by understanding the link between ongoing learning and final achievement.

Main Methods:

• Utilized a large sample of children assessed across multiple time points during compulsory schooling.
• Developed and compared regression models predicting SA abilities using various FA-derived features and auxiliary variables.
• Estimated student abilities using data from computer-based formative assessments.

Main Results:

• A model incorporating mean abilities across competence domains achieved the best prediction of SA outcomes, explaining 30-48% of the variance.
• Predictive FA features typically belonged to the same or a related competence domain as the SA being predicted.
• Systematic model biases were identified, requiring careful consideration for practical decision-making.

Conclusions:

• Formative assessment data holds significant potential for predicting future summative achievement.
• Findings suggest that ongoing learning progress, as measured by FAs, is a strong indicator of later academic success.
• This research can support adaptive instruction and inform policies aimed at mitigating the impact of high-stakes summative testing.