Shaky Student Growth? A Comparison of Robust Bayesian Learning Progress Estimation Methods.

Area of Science:

• Educational Measurement and Statistics
• Psychometrics
• Educational Psychology

Background:

• Learning progress assessments are crucial for teachers' data-based decision-making.
• Estimating student learning progress traditionally uses latent growth modeling or individual student progress estimation.
• Recent research explores robust estimation for outliers and Bayesian approaches for student growth estimation.

Purpose of the Study:

• To combine robust estimation and Bayesian approaches within a linear latent growth model framework.
• To extend existing research on student growth estimation by integrating robust Bayesian methods.
• To compare the performance of different Bayesian linear latent growth models for assessing learning progress.

Main Methods:

• Comparison of three Bayesian linear latent growth models: Gaussian, Student's t-distribution (robust), and asymmetric Laplace (robust).
• Utilized a dataset of N = 4970 second-grade students with reading comprehension data from eight measurement points.
• Employed leave-one-out cross-validation and posterior predictive model checking for model comparison.

Main Results:

• Both robust models (Student's t and asymmetric Laplace) outperformed the Gaussian model in estimating learning progress.
• The Student's t-distribution model showed a slight statistical advantage, while the asymmetric Laplace model offered more realistic posterior predictive samples and higher measurement precision.
• Robust models performed comparably well, indicating their effectiveness in handling potential outliers in learning progress data.

Conclusions:

• Robust Bayesian estimation methods are effective for learning progress assessment.
• The choice between Student's t and asymmetric Laplace models may depend on specific priorities like statistical fit versus predictive accuracy and precision.
• Findings support the use of advanced statistical techniques to enhance the reliability and utility of learning progress assessments for instructional decision-making.