Test of Slope and Intercept Bias in College Admissions: A Response to Aguinis, Culpepper, and Pierce (2010)

• Published in: Journal of applied psychology 2013-01, Vol.98 (1), p.134-147
• Main Authors: Mattern, Krista D, Patterson, Brian F
• Format: Article
• Language: English
• Subjects: Academic achievement; Admissions; African American Students; Aptitude Tests - standards; Behavioral Research; Bias; Biological and medical sciences; Cognition; Cognitive bias; Cognitive Tests; College Admission; College admissions; College Entrance Examination Board Scholastic Aptitude Test; College Entrance Examinations; College Freshmen; Educational psychology; Estimation bias; Female; Forecasts; Fundamental and applied biological sciences. Psychology; Gender Differences; Grade Point Average; High schools; Hispanic American Students; Hispanic people; Human; Humans; Male; Minority groups; Orientation. Evaluation; Personnel Selection - standards; Predictability (Measurement); Prediction; Psychology. Psychoanalysis. Psychiatry; Psychology. Psychophysiology; Regression (Statistics); Regression analysis; Revival; SAT (College Admission Test); SAT assessment; Scores; Secondary schools; Statistical Bias; Student Admission Criteria; Students; Studies; Test Bias; White Students
• ISSN: 0021-9010; 1939-1854
• DOI: 10.1037/a0030610

Research on the predictive bias of cognitive tests has generally shown (a) no slope effects and (b) small intercept effects, typically favoring the minority group. Aguinis, Culpepper, and Pierce (2010) simulated data and demonstrated that statistical artifacts may have led to a lack of power to detect slope differences and an overestimate of the size of the intercept effect. In response to Aguinis et al.'s (2010) call for a revival of predictive bias research, we used data on over 475,000 students entering college between 2006 and 2008 to estimate slope and intercept differences in the college admissions context. Corrections for statistical artifacts were applied. Furthermore, plotting of regression lines supplemented traditional analyses of predictive bias to offer additional evidence of the form and extent to which predictive bias exists. Congruent with previous research on bias of cognitive tests, using SAT scores in conjunction with high school grade-point average to predict first-year grade-point average revealed minimal differential prediction ([delta]R[squared][subscript intercept] ranged from 0.004 to 0.032 and [delta]R[squared][subscript slope] ranged from 0.001 to 0.013 depending on the corrections applied and comparison groups examined). We found, on the basis of regression plots, that college grades were consistently overpredicted for Black and Hispanic students and underpredicted for female students. (Contains 6 tables, 2 figures and 6 footnotes.)