Assessing the Assumption of Measurement Invariance in College Admissions Testing

• In Event: 65.079 - Roundtable Session 23
  In Roundtable Session: 65.079-17 - Measurement Invariance: Practical Applications

Abstract

Traditional predictive bias analysis relies on an examination of differences in observed intercepts and slopes across ethnicity- and gender-based subgroups. We argue that this prevailing approach only offers an omnibus predictive bias test whose results are not necessarily informative regarding the nature and underlying reasons for any bias found. We offer an expanded conceptualization and assessment of predictive bias that relies on psychometric theory and measurement invariance. We illustrate the use of the expanded framework using college admissions data and results allow for a re-interpretation of previous findings and novel insights. The expanded conceptualization can be used to guide future theory-based research on the nature of predictive bias as well as its underlying reasons.

Authors

• Steven Andrew Culpepper, University of Illinois at Urbana-Champaign

• Herman Aguinis, The George Washington University

• Roger Millsap, Arizona State University