Acceso usuarios
Gabriela Susana Lozzia, María Silvia Galibert, María Ester Aguerri, Horacio Félix Attorresi
El objetivo de este trabajo es presentar el desarrollo de un banco de ítem de razonamiento verbal a partir de la Teoría de Respuesta al Item (TRI). Se presenta la TRI y su aplicación en la elaboración de bancos de ítem que posibilitan el diseño de tests adaptativos. Los ítem son de elección múltiple y miden la habilidad para reconocer y discriminar relaciones entre palabras. Un banco de ítem es un conjunto de ítem que miden una misma variable y cuyos parámetros están calibrados (estimados) en una misma escala. La construcción de un banco es un proceso de creación-calibración de ítem que se realiza en sucesivas etapas. Como los sujetos de las muestras son diferentes en cada etapa, los ítem a calibrar deben ser administrados junto con un pequeño grupo de ítem calibrados en etapas anteriores, los cuales sirven de enlace para que todas las estimaciones resulten en la misma escala. La estimación de los parámetros se lleva a cabo por el método de máxima verosimilitud marginal ajustando el modelo logístico de tres parámetros con el programa XCALIBRETM. Los análisis del funcionamiento diferencial (Differential Item Functioning - DIF) se basan en el test normal para la diferencia de los parámetros de dificultad, dicha diferencia con sus errores estándar para cada ítem es proporcionada por BILOG-MGTM. Se eliminan aquellos que no ajustan al modelo y los que presentan DIF. El banco cuenta hasta el momento con 93 ítem.
A possibility of integration cognitive theory and modern psychometric. One of the advantages of the Item Response Theory (IRT) compared to the Classical Test Theory is the possibility of building measurement instruments with properties independent to the subjects to be measured with them. Within IRT it is verified that the difficulty and discrimination item parameters remain invariant no matter which subject population is used. An item bank is a set of items which measure the same variable and whose parameters are calibrated, that is estimated, in a common scale. Whenever an item bank is available, administrations of adaptive tests become possible; this means the bank offers the possibility of choosing those items which assess each subject more accurately. Also, it allows designs of tests with pre-established characteristics according to the measurement objective. The application of the IRT is still incipient in Argentina and Differential Item Functioning (DIF) studies have not been known so far. The aim of this work is to show the development of a Bank of Verbal Reasoning Items according to the IRT. This theory, as well as its application to the elaboration of item banks, are here in presented. These items measure the ability to identify and discriminate relationships among words. Such ability is related to the ideative factor of the verbal comprehension, which is common to all tasks of deductive, serial and probabilistic reasoning, classification and problem solution. Therefore, it is one of the main factors to get the intellectual aptitude profile of students. Each item consists of a pair of basic words with some relationship existing between them and four alternative pairs of words. The instruction commands to choose the pair with the most similar relationship to the one existing between the words in the basic pair. The construction of this bank has involved an iterative process of invention-calibration of the items. In the invention stage, the items have been selected and improved through successive pilot trials. In the selection processes, some classical statistics procedures were considered such as the question evaluation index. The calibration stage has been developed in two main phases: the initial phase and the current one. In the initial phase the calibration was done with samples from different student populations. Although some of them had responded different item sets, the calibration was done in a simultaneous running of the program BILOG-MGTM in order to obtain the estimates in a common scale. The assumption of unidimensionality was checked through a scree plot implemented by MicroFACTTM. The current phase consists of adding new items to the bank until it has a substantial number of them. Before calibration, DIF analysis have been done in both phases. Taking into account the different subjects of the corresponding samples, it is necessary to manage the items together with a small group of already calibrated ones. These items serve as a link so that the parameter estimates of the new items result in the same scale as the one of the rest of the bank. Link items have been chosen with difficulty parameters representing different levels of ability and high discrimination parameters. DIF analysis have been based on the normal test of the difference between the difficulty parameters, which has been obtained from BILOG-MGTM. This software as well as XCALIBRETM have been used to calibrate the items by fitting the three-parameter logistic model. Stout's test of essential unidimensionality analysis has been done by using DIMTEST. After removing those items which present DIF or where the model does not fit, 93 items have been remained for the bank.
