An Evaluation of Some Models for Culture‐fair Selection

Predictive Bias in Graduate Admissions Tests

“…He further argues that consensus utility models (Darlington, 1971;Gross & Su, 1975;and Petersen & Novick, 1976), 'must fail, though not for technical reasons. Social consensus is subject to test by the courts, and the courts have already cast a shadow over the use of multiple cutting scores" (Weitzman, 1978, p. 7).…”

“…Gross and Su (1975), Petersen (1976), Petersen and Novick (1976), Novick and Ellis (1977), and Novick and Lindley (1978) have given examples of procedures for explicitly assigning different utilities to the selection of disadvantaged vs. advantaged students with the same criterion scores. These researchers have examined the impact of such a quantification of value preferences on differential selection rules.…”

Contributions to the Quantitative Assessment of Item, Test, and Score Fairness

“…G is a variable indicating subgroup membership. Petersen and Novick (1976) compared several models for assessing fair selection, including the regression model (Cleary 1968), the constant ratio model (Thorndike 1971), the conditional probability model (Cole 1973), and the constant probability model (Linn 1973) in the lead article in a special issue of the Journal of Educational Measurement dedicated to the topic of fair selection. They demonstrated that the regression, or Cleary, model, which is a differential prediction model, was a preferred model from a logical perspective in that it was consistent with its converse (i.e., fair selection of applicants was consistent with fair rejection of applicants).…”

“…In essence, the Cleary model examines whether the regression of the criterion onto the predictor space is invariant across subpopulations. Linn (1976) in his discussion of the Petersen and Novick (1976) analyses noted that the quest to achieve fair prediction is hampered by the fact that the criterion in many studies may itself be unfairly measured. Even when the correct equation is correctly specified and the criterion is measured well in the full population, invariance may not hold in subpopulations because of selection effects.…”

Validity

“…Indeed, some JOOdels proved to be mutually incompatible or even contradictory, thereby offering true Hegelian contrasts (Darlington, 1971;Petersen Or Novick, 1976). SUCh no:1els assume, either eJq)licitly or tacitly, an unbiased criterion measure that is fairly and uniformly applied across groups.…”