The Controlled Oral Word Association (COWA) test is used to assess phonemic fluency and executive function. Formal validation of test scores is important for accurate cognitive evaluation. However, there is a dearth of psychometric validation among American Indian adults. Given high burden of dementia risk and key contextual factors associated with cognitive assessments, this represents a critical oversight. In a large, longitudinal population-based cohort study of adult American Indians, we examined several validity inferences for COWA, including scoring, generalization, and extrapolation inferences, by investigation of factor structure, internal consistency, test-retest reliability, and differential test functioning. We found adequate unidimensional model fit, with high factor loadings. Internal consistency reliability and test-retest reliability were 0.88 and 0.77, respectively, for the full group. COWA scores were lowest among the oldest, lowest education, bilingual speakers; group effects for sex and bilingual status were small; age effect was medium; and education effect was largest. However, Wide Range Achievement Test (WRAT) score effect was stronger than education effect, suggesting better contextualization may be needed. These results support interpretation of total COWA score, including across sex, age, or language use strata.
Keywords: American Indian; differential item functioning; reliability; validity; verbal fluency.