Using Group Differences in True Score Relationships to Evaluate Measurement Bias.

Area of Science:

• Psychometrics
• Educational Measurement
• Statistical Modeling

Background:

• Measurement bias and predictive bias are critical in standardized testing.
• Existing models like the Cleary model primarily address predictive bias.
• Understanding these biases is essential for fair assessment in admissions and employment.

Purpose of the Study:

• To develop a novel linear model for assessing measurement bias using Errors-in-Variables (EIV) regression.
• To differentiate measurement bias from predictive bias, especially when true-score means differ between groups.
• To re-examine empirical findings on test bias in college admissions and employment contexts.

Main Methods:

• Development of a linear Errors-in-Variables (EIV) regression model for measurement bias.
• Comparative analysis of measurement bias and predictive bias under varying true-score means.
• Re-evaluation of existing empirical data on test performance prediction for minority groups.

Main Results:

• The proposed EIV model offers a new approach to measurement bias assessment.
• Ignoring the distinction between measurement and predictive bias can lead to misinterpretations due to regression toward the mean.
• Empirical findings on over-prediction for minorities are consistent with significant measurement bias against them.

Conclusions:

• The study provides a refined framework for understanding and quantifying measurement bias.
• It underscores the importance of distinguishing between measurement and predictive bias for accurate test evaluation.
• Results suggest that observed over-prediction in minority groups may stem from inherent measurement bias rather than solely predictive issues.