Could "Or" Give Us More? Equity Implications of Combination Rules When Correlations Vary Across Groups

• In Event: Gifted Identification in Students From Traditionally Underrepresented Populations Symposia

Abstract

Using multiple measures to identify students for gifted programs is considered best practice in the field of gifted education. Identification systems generally include conjunctive (“And”), disjunctive (“Or”) and compensatory (“Mean”) rules for combining multiple measures. As the correlations among assessments decrease, the conjunctive and compensatory systems identify fewer students (unless the cut-off for the mean score is adjusted for shrinkage) and disjunctive rules identify more students (McBee at al., 2014).
However, both published research and current district identification procedures assume that correlations among assessments are the same across groups. In other words, in both research and practice, gifted educators assume that the correlations among multiple identification measures are the same for students from different backgrounds. Correlations among measures could vary by group for a variety of reasons. Some of the most obvious reasons are: 1) the variability of the measures could vary across groups; 2) the reliability of one or more of the measures could be higher in one group than another; and/or 3) the correlation among the true scores of measures (i.e., the correlations of the latent constructs themselves) could vary across groups. If the correlations among measures are lower for one group than another, the group with the lower correlations would be disadvantaged by conjunctive and compensatory rules. However, they would be advantaged by disjunctive rules. This nuance has a substantial effect on equity and effectiveness and yet has been underappreciated to date.
In this paper, we explore these issues empirically. First, using data from three large districts that provided ability, achievement, and teacher rating scale data for all students within a grade level, we examine whether there are racial/ethnic differences in the pattern of correlations. Our results suggest that in general, across the three districts, the correlations are noticeably higher for White and Asian students than they are for Black and Latinx students. Such findings would suggest that Black and Latinx students would benefit disproportionately more than their White or Asian peers from disjunctive rules than from conjunctive or compensatory rules. Next, we examined the performance of the three combination rules, using a simple two measure system in which teacher rating scales and ability scores are combined to identify students as gifted. Our preliminary results suggest that with one notable exception (Latinx students in district 1), the “Or” rule did identify proportionally more traditionally underserved students, which narrowed the identification gap between underserved and non-underserved students. This makes sense in light of the observed lower correlations for Black and Latinx students. Our full paper will evaluate additional combinations of assessments. The implications of this work are clear. If the correlations among the assessment measures used for gifted identification are lower for traditionally underserved populations (such as Black and Latinx students), then using disjunctive rules to combine multiple measures might provide an easy method to increase the equity of districts’ identification systems.

Authors

• D. Betsy Mccoach, University of Connecticut
  Presenting Author

• Scott Joseph Peters, NWEA
  Presenting Author

• Daniel A. Long, University of Connecticut
  Presenting Author

• Anthony J. Gambino, University of Connecticut
  Presenting Author

• Pam Peters, Michigan State University
  Presenting Author

• Del Siegle, University of Connecticut
  Presenting Author