Implications of the Implicit Association Test D-Transformation for Psychological Assessment

Abstract

Psychometricians strive to eliminate random error from their psychological inventories. When random error affecting tests is diminished, tests more accurately characterize people on the psychological dimension of interest. We document an unusual property of the scoring algorithm for a measure used to assess a wide range of psychological states. The "D-score" algorithm for coding the Implicit Association Test (IAT) requires the presence of random noise in order to obtain variability. Without consequential degrees of random noise, all individuals receive extreme scores. We present results from an algebraic proof, a computer simulation, and an online survey of implicit racial attitudes to show how trial error can bias IAT assessments. We argue as a result that the D-score algorithm should not be used for formal assessment purposes, and we offer an alternative to this approach based on multiple regression. Our critique focuses primarily on the IAT designed to measure unconscious racial attitudes, but it applies to any IAT developed to provide psychological assessments within clinical, organizational, and developmental branches of psychology-and in any other field where the IAT might be used.

Using the IAT to predict ethnic and racial discrimination: Small effect sizes of unknown societal significance.

“…Relying on zero-order correlations between the IAT and all criteria in general is too crude and begs the fundamental question of the behavioral implications of different IAT scores (Blanton, Jaccard, & Burrows, 2015;Blanton, Jaccard, Strauts, Mitchell, & Tetlock, 2015). Greenwald et al (2015) offer two examples that they contend illustrate how the average effect size estimates for race IATcriterion correlations might have real-world consequences.…”

Implicit and explicit attitudes towards sport among young elite athletes with high versus low burnout symptoms

“…As dependent variable, D-score was calculated by dividing the ST-IAT raw scores (difference in mean reaction time between the two blocks) by the within--individual standard deviation of response latencies calculated across the compatible and incompatible trials (Greenwald, Nosek, & Banaji, 2003). D-scores can take up values from -2 to +2; and 0.15, 0.35 and 0.64 have been interpreted as slight, moderate and strong preference (or aversion), respectively (Blanton, Jaccard, & Burrows, 2014).…”

Implicit Measures of Actual Versus Ideal Body Image: Relations with Self-Reported Body Dissatisfaction and Dieting Behaviors

“… 1 Because there is still debate about the best way to handle reaction time data, we repeated the analyses without dividing the difference in mean response latency between the two blocks by the pooled SD (raw-score; Blanton et al 2015 ). When outcomes differ markedly from the original D-scores, we report this in a footnote.…”