A comparison of the use of artificial neural network and traditional psychometric methodologies in the discriminant validation of a multidimensional self-concept measure

Purpose. The major purpose of this study was to investigate the effectiveness of using an artificial neural network as an alternative approach to traditional psychometric methods in the demonstration of the discriminant validity of the six 15-item factor scales of the Dimensions of Self-Concept (DOSC), Form W measure prepared by Crowder and Michael in 1991. From use of their data base, attention was directed toward showing an analogy between application of artificial neural network methodology and use of conventional psychometric methodology with respect to obtaining indicators or indexes reflecting (a) internal consistency of scores in each scale, (b) degree of association or extent of interrelationships among the six factor scales that were anticipated to be relatively low in light of the initial work carried out by Crowder and Michael, and (c) power of item discrimination of contrasting sets of five items judged by the investigator and a psychometrician on the basis of data obtained from the initial work of Crowder and Michael to be relatively high or relatively low within each of the six scales. If comparability could be found in the results furnished by neural network methodology to those realized from application of traditional psychometric procedures, use of neural networks might be a plausible alternative in test development, particularly in the instance of achieving discriminant validity in multidimensional scales.; Conclusions. (1) Neural network methodology appears to be a viable alternative to the use of traditional psychometric procedures in the demonstration of the discriminant validity of a multidimensional psychological test in the affective domain of self-concept. (2) Neural network methodology can be expected to provide promising results in the identification of items that are most discriminating or least discriminating in relatively homogeneous scales in psychological measurement.