The Counterintuitive Result of Racial Bias in Nominations: Insights From Numerical Analysis

• In Event: Leveraging Nonempirical Quantitative Analysis for Theory Development and Exploration: A New Paradigm for Giftedness Research

Abstract

Racial disparities in gifted program enrollment have been a dominant concern for our field. In spite of decades of effort, including in some cases the creation of alternative identification procedures and cutoffs for Black, Hispanic, or low-income populations, enrollment gaps continue to persist.

Initial efforts to explain this phenomenon focused on bias in the ability and achievement test instruments used in the identification process. The assumption was that racial or cultural bias had been inadvertently designed into psychometric instruments. This bias would cause, for example, an IQ test to produce scores that were systematically lower for a Black student than a White student conditional on their having exactly the same level of underlying ability. Thus, the observed (roughly) one standard deviation gap between Black and White mean IQ was assumed to be an artifact of faulty test design rather than reflective of real differences in developed ability.

Decades of investigation into this hypothesis have revealed it to be utterly incorrect. Briefly: test score gaps do not shrink when tests are translated into nonstandard English (Quay, 1971) or when potentially biased items identified during content review are removed (Bianchini, 1976). Pyryt (1996) pointed out that Black and White students tend to miss the same items, but Black students miss them more frequently. And sophisticated differential item functioning (DIF) analyses have generally not discovered bias at the item level.

Relatively recent work in gifted education has focused on the nomination stage as a more likely point for racial bias to play a role in identification. Given that nomination often involves subjective judgment, and that individuals are susceptible to unconscious racial bias (ref), it would be unsurprising to learn that nominations are racially biased. McBee (2006) provided numbers from a state-level dataset indicating that while 14.65% of White students received a nomination, only 4.58% of Black students received one.

This paper describes results of a numerical analysis of a two-stage identification process for White and Black students. True ability, nomination, and testing scores are conceived as arising from a multivariate normal distribution. Covariance terms are adjusted to simulate imperfect validity at the nomination and testing stages. Bias at the nomination stage is modeled in two different ways. The first way is through lower validity for one group. For example, teachers may be less familiar with means by which students from nondominant cultures express their ability, thereby reducing the accuracy of teacher appraisals of the abilities of students from that group. The second, more insidious way bias can occur is subtractive. This means that a certain number of “points” is subtracted from the student’s perceived ability.

Results from the simulation show that low validity bias at the nomination stage does not reduce the number of students nominated but rather reduces the pass rate for nominated students at the testing phase. Subtractive bias reduces the number of students that are nominated but increases the pass rate at the testing stage. Results of this analysis constrain theories regarding causes of racial disparity in gifted programs.

Author

• Matthew McBee, East Tennessee State University