Construct-signature validity (CSV) for intelligence testing

Construct validation in intelligence testing is often driven by factor-analytic models. Yet global fit can coexist with weak evidence that item demands represent the target attribute. This article introduces Construct-Signature Validity (CSV), a design-based framework. Within the CSV framework, one source of evidence for construct validity is the degree to which pre-specified ability-by-complexity patterns are observed in test data. CSV treats intelligence as observable in person-by-task-demand patterns rather than only as a trait inferred from undifferentiated covariation. CSV outlines three nested signatures: a within-subtest signature (SCI), indexed by ordered correlations among low-, medium-, and high-complexity band scores across ability levels; a within-domain signature (DCI), indexed by a shift of the peak cross-subtest coherence to the complexity level matching group ability; and an item-level signature (ICI), indexed by ordered success probabilities and drop-contrast conditions across complexity bands. CSV is demonstrated with a secondary analysis of norming data from the Anadolu-Sak Intelligence Scale (ASIS). Items from four nonverbal subtests were classified a priori into three complexity levels, and children were divided into three IQ groups. SCI matched the predicted order at both low- and high-ability bands. However, the evidence was partial because the adjacent correlations were close and the key differences were small. The medium band showed adjacency without the predicted transition. DCI supported the expected peak shift in the high-ability group, but only partially in the low-ability group, and was absent in the medium-ability group. At the item level, percent-correct profiles and a binomial GEE model showed a reliable ability-by-complexity interaction. CSV complements factor models by providing testable evidence about where construct representation and calibration hold or fail across the ability distribution.