An empirical link of content and construct validity evidence

Abstract

Since the 1940s, measurement specialists have called for an empirical validation technique that combines content- and construct-related evidence. This study investigated the value of such a technique. A self-assessment instrument designed to cover four traditional foreign language skills was administered to 1,404 college-level foreign language students. Four subject-matter experts were asked to provide item dissimilarity judgments, using whatever criteria they thought appropriate. The data from the students and the experts were examined separately using multidimensional scaling followed by cluster and discriminant analyses. Results showed that the structure of the data underlying both the student and expert scaling solutions corresponded closely to that specified in the instrument blueprint. In addition, using canonical correlation, a comparison of the two scaling solutions revealed a high degree of similarity in the two solutions. Index terms: canonical correlation, construct validity, content validity, item dissimilarities data, multidimensional scaling.