Before applying a normal-distribution stabilization model, agencies should thoroughly explore indicator data distributions with visual and numeric checks¶

Strategy · All strategies
Evidence · no claims cited

Description¶

This strategy directs education agencies to check whether indicator distributions align with a stabilization model's assumptions before applying it. The article recommends that "agencies should thoroughly explore data distributions, supplementing visual checks with numeric ones, to understand how well indicator distributions align with model assumptions of normality." Recommended checks include histograms at the subgroup–indicator level, Q–Q plots, scatterplots, and supplementary normality tests such as D'Agostino's K-squared test, with visual inspection treated as most useful because statistical tests are sensitive to sample size.

Design Implications¶

Context¶

Requirements¶

• Access to subgroup–indicator-level score distributions for each performance indicator under consideration for stabilization

Constraints¶

• Statistical tests of normality are sensitive to sample size and point estimates provide little information about distribution shape, so visual inspection may be most useful

Target Learners¶

• State education agencies
• Student subgroups including small demographic groups within schools

Target Learning Goals¶

• Reliable measurement of school and student-group performance for accountability and diagnostic decisionmaking

Related Strategies¶

• Use a decision template weighing assumption fit, DIF uniformity, simultaneity, significance testing, and ease of use when choosing a DIF method for polytomous performance assessments

Examples¶

-

Key Sources¶

• Rosendahl, M., Gill, B., & Starling, J. E. (2024). Stabilizing school performance indicators in New Jersey to reduce the effect of random error (REL 2025–009). U.S. Department of Education, Institute of Education Sciences, National Center for Education Evaluation and Regional Assistance. https://ies.ed.gov/use-work/resource-library/report/descriptive-study/stabilizing-school-performance-indicators-new-jersey-reduce-effect-random-error