Скачать или смотреть Advancing Equity and Accuracy in High-Stakes Testing

Unveiling the potential of latent variable models in high-stakes testing for a fairer and more accurate evaluation system.

Here are 5 key takeaways from an innovative case study on assessing measurement and prediction invariance in high-stakes testing:
1️⃣ Latent Variables Offer Deeper Insights: Researchers can uncover underlying biases in test scores that traditional methods like observed score analysis often miss using latent variable models. This "X-ray vision" approach reveals how group-based disparities may stem from deeper, unobserved factors.
2️⃣ Measurement Invariance Matters: The study showed significant group differences in the SAT mathematics subtest (SAT-M) measurement intercepts, particularly for Black and Hispanic students. These differences highlight the importance of testing measurement invariance to ensure scores reflect true abilities.
3️⃣ Improved Accuracy with 2SLS Estimator: The novel two-stage least squares (2SLS) estimator proved more accurate and stable than traditional methods like moderated multiple regression (MMR). This approach accounted for measurement errors and provided clearer insights into group differences.
4️⃣ Practical Implications for Policy and Testing: The findings suggest that some observed group differences in high-stakes testing might be artifacts of the testing process. Addressing these biases through better test design and analysis could lead to fairer outcomes for underrepresented groups.
5️⃣ A Call for Future Research: This work paves the way for exploring deeper theoretical mechanisms behind group differences, such as stereotype threat or socioeconomic disparities. Future research should consider nonlinear models and multilevel analyses to expand on these findings.

Get article: Culpepper, S. A., Aguinis, H., Kern, J., & Millsap, R. 2019. High-stakes testing case study: A latent variable approach for assessing measurement and prediction invariance. Psychometrika, 84(1): 285-309. https://link.springer.com/article/10....