Reconsidering the Psychometrics of the GRS-S: Evidence for Parsimony in Measurement

Measurement-level considerations tested the extent to which treating the Likert-type scale rating as categorical or continuous produced different fit across unidimensional, correlated trait, and bifactor latent factor structures. Item- and scale-level analyses demonstrated that the GRS-S could be reduced from a 72-item assessment on a 9-point rating scale down to a 30-item assessment on a 3-point rating scale.

Reliability from the reduced assessment was high (ω > .95). Receiver operating characteristic (ROC) curve comparisons between the original and reduced versions of the GRS-S showed that diagnostic accuracy (i.e., area under the curve) of the scales was comparable when considering cut scores of 120, 125, and 130 on the WISC-IV Full Scale ( Wechsler Intelligence Scale for Child–Fourth Edition ) and verbal IQ and the WIAT-III (Wechsler Individual Achievement Test–Third Edition) composite score.