Proposal for New Methods of Factor Analysis Related to Item Response Time Data

Abstract

This paper proposes two new approaches to factor analysis. The first is a factor analysis model for item response time data which assume that the relationships between item response time data and latent traits are non-linear in nature. The second involves pre-specification of certain parameters while, at the same time, estimating others, and then considering the cognitive process involved in the solution of problems. The materials for the item response time data study were obtained using computerized tests. The authors then compared the results of using this two methods with traditional factor analysis. Findings included : (1) traditional methods of factor analysis often fail to obtain satisfactorily interpretable estimates while the first of the author's new methods are always successful. (2) In terms of Schwarz's BIC, the method of factor analysis using pre-specified parameters as developed by the authors was most efficient. In particular, it accurately predicts item response time distributions based upon an a priori designation of the cognitive processes involved. This capability should prove particularly useful in test development and instructional design.