Confirmatory factor analysis with preterm NBAS data: A comparison of four data reduction models

Abstract

A widely used data reduction (cluster) system for the Neonatal Behavioral Assessment Scale (NBAS) is a seven-cluster model developed by Lester (1983). Minimal information, however, is currently available concerning the psychometric properties of these clusters. The purpose of this study was to analyze the psychometric properties of Lester's factors and compare it to other data reduction models. Preterm infants who received NBAS examinations prior to discharge from the Special Care Nursery were subjects in the study. Three alternative models to Lester's were developed based upon previous factor studies and clinical experience. Confirmatory factor analyses were then performed on the four models using the LISREL VI maximum likelihood procedure. The results indicated that none of the four models tested could be confirmed. The two-factor model (orientation and arousal) appeared to be the best fitting of the four models. The results further indicated that multiple-factor models such as Lester's have significant correlations among the factors. In terms of item analysis, low to moderate reliability coefficients were found for the NBAS items. Although preliminary, these findings indicate a need to reconsider the nature of NBAS data. Development of scoring systems which account for the relationships among the factors might more accurately reflect the nature of behavior as measured by the NBAS.