Evaluation of the Factor Structure and IRT analysis of an Icelandic Preschool Language Development Test

Abstract

Málfærni ungra barna (MUB) is a newly published Icelandic language development test intended to assess the language development of two to four-year-old children, along with confirming or rejecting deviation in language development. The test consists of two subtests, one intended to estimate receptive language skills and the other expressive language skills. The purpose of this study was to evaluate the factor structure and IRT parameters of the MUB test on a cluster sample of 643 Icelandic children. The three models tested were a unidimensional model, correlated two-factor model, and a bifactor model. Previous research on the factor structure of the test indicated that a correlated two-factor model best describes the factor structure of the test. However, only a limited number of factor models were tested in earlier research. One of the models that has not been tested on the MUB test is the bifactor model, which was of central focus in this study. The results showed that the bifactor model best fitted the data of the models tested. The direction and magnitude of loadings on the specific factors in the bifactor model indicated that the general factor was very dominant, and the specific factors explained only a small variance each after accounting for the general dimension. These results suggest that the specific factors do not provide sufficient information to justify subtest scores for the test, and indicate that the MUB test should be scored as a unidimensional measure. Unidimensional IRT analysis of the test showed that the test provides the most accurate information about children with average language skills, and that the test lacks difficult items to provide information for the whole ability range of language skills in preschool children. Floor effects were seen in the youngest age group and ceiling effects in the oldest age groups, which indicates that the age group that the test is intended for is too wide.