Item Response Theory and Linear Mixed Models for Normative Data Generation: Evidence from the TAMV-I

Abstract Objective To evaluate item parameters of the TAMV-I (Verbal Learning and Memory Test) using Item Response Theory (IRT) and to generate normative data based on participant ability scores adjusted for sociodemographic characteristics. Method The sample included 1640 children and adolescents (ages 6–17) from Spain, Honduras, Ecuador, and Colombia. IRT analyses were conducted using the two-parameter logistic (2PL) model to estimate item discrimination and difficulty. Linear Mixed Models (LMMs) assessed the effects of age, age2, sex, country, and mean parental education (MPE) on ability (theta) scores. Results The 2PL model consistently outperformed the Rasch model across trials (p < 0.001), providing a superior fit and more precise item estimates. Discrimination improved across trials. Items such as Blouse and Shoe (Delayed Recall and Trial 4), along with Eye, Ear, Mouth, and Scarf (Recognition), provided the highest information in estimating participant ability. In Trial 6, low-ability participants were able to identify high-frequency words. LMMs revealed significant interactions: ability increased with MPE, with Ecuadorian participants outperforming others (ln[MPE] × country); performance rose until age 13 and then declined (age ×trial); girls outperformed boys in all countries except Honduras (sex×country); and Spanish participants scored highest across trials, except in Trial 1, where Ecuadorians led (trial × country). Low residual standard deviations (0.56–0.57) indicated high measurement precision. Conclusion IRT and LMM approaches support the development of culturally sensitive norms for the TAMV-I. Demographic and contextual variables significantly influence verbal memory performance and should be considered when interpreting results.