Propuesta metodológica para el aumento de datos en muestras pequeñas para la estimación de parámetros de ítems

Abstract

In the item response theory (IRT), numerous factors have been studied that in uence the accuracy of the item parameter estimation, including the number of items that make up the test, the number of individuals that are evaluated, the estimation method used, the distribution of the adjusted scores of the individuals, among many others. It is known that the number of individuals evaluated is one of the factors that most in uence the accuracy of the estimation of these parameters depending on the model used (Sahin & Anil 2016). The parameter estimation methodologies used in the IRT require a minimum number of individuals evaluated to obtain accurate estimates of the items parameters (Hambleton 1989). However, these amounts can not always be reached for various reasons, such as for example that the population evaluated is not large enough to reach the required sample or that due to budget issues it is very expensive to access a minimum sample. The objective of this work consists of to propose and implement a multiple imputation with plausible values in a technique called DuPER for its acronym (Duplicate, Erase, Replace) which is a technique of data augmentation whose purpose is to expand a sample of individuals to obtain su cient information to make precise estimates of items parameters.