Teoria de Resposta ao ItemRepresentação e utilidade do modelo logístico de traço latente na psicometria actual

Abstract

Framework Rasch model is the simplest Item Response Theory (ITR) model and constitutes a potentially useful approach at both the construction and refinement psychometric levels of psychological assessment instruments. It is an ITR one parameter logistic model in which the quantity of the latent trait existing in the person and the quantity of the same latent trait reflected in the various items of the instrument can be independently estimated and compared, because both items and subjects are measured using the same common metric, the logit scale. This scale, defined in logarithmic probabilities units, is an interval scale, in which the interval units between the response categories have a consistent value or significance. The goal of this dissertation is focused on the presentation of the Rasch model potentiality as compared with the classical test theory (CTT) approach when applied to the specific case of psychological assessment instruments that recur to polytomous items to which the subjects respond using a Likert type rating scale - the Rating Scale Model (RSM). Methodology In order to achieve this goal, related to the application of the Rasch RSM, four psychometric studies were carried out with independent samples of students attending secondary education. The first study, with a sample of 12th grade students (N = 265), was carried out using the Career Decision Self-Efficacy Scale — Short Form (CDSE-SF). In the second and third studies, comprising students attending respectively the 10th, 11th and 12th grades, we ran an analysis of the Positive and Negative Affect Schedule (PANAS), using both exploratory/confirmatory factor analysis (EFA/ CFA), and Rasch modeling (N = 519), respectively. Finally, in the fourth and last study, also comprising 10th, 11th and 12th grade students (N = 508), a Rasch analysis of the Rosenberg Self-Esteem Scale (RSES) was made. In each of these four studies, in addition to the items’ measurements, data needed for the sociodemographic characterization of the samples were collected. Results From the results obtained we highlight those that are more relevant to the goals previously stated. Within the three studies carried out with the RSM the results obtained gathered psychometric evidence that confirms, concerning the CDSESF, the PANAS and the RSES, their respective: 1) content validity (i.e. the good adjustment of the data to the model allowed both the parameterization of the subjects and the calibration of the items to be made with high precision); 2) structural validity (i.e. unidimensionality and lack of DIF) and; 3) substantive validity (i.e. the categories of the response scale function adequately). The results of the only study which was based on CTT, and in which the PANAS was used, revealed some adjustment difficulties of the model to the data, which only partially replicated the original factorial structure of the instrument. Conclusions The results suggest that the RSM provides a reference framework useful for the psychometric refinement of psychological assessment instruments. In fact, from its application it can be concluded that, in the specific case of the Portuguese versions of the CDSE-SF, the PANAS and the RSES, their respective items are representative and relevant for the domains of the constructs under assessment (e.g. content validity), have correspondence with the constructs defined in each of the instruments (e.g. structural validity), and are assessed using response scales whose diagnostic to the empirical functioning of their respective categories revealed their adequacy (e.g. substantive validity).