Note : Construct validity of the Pearson Test of English Academic : A multitrait - multimethod approach 1

Linguistic abilities are multifaceted and cannot easily be measured directly. Consequently, multiple diagnostic methods are employed to assess multiple traits of language skills. This indirect means of assessment raises issues of adequacy, appropriateness, and utility of the measures in testing an individual’s ability or skill. With the contribution of Campbell and Fiske (1959) to the field of language testing, a multitrait-multimethod (MTMM) approach has been used by many researchers not only to validate inferences made based on test scores, but also to build theories. The MTMM approach is largely used to examine how different traits (multitraits) of language abilities and methods (multimethods) of testing materials influence the student’s performance. Campbell and Fiske’s (1959) original conceptualization was characterized by hypotheses that correlations among scores measuring a unitary ability (monotrait correlations) would be higher than those among scores using a single test method (monomethod correlations), and monotrait-monomethod correlations would be higher than those among measures of different traits using different methods (heterotraitheteromethod correlations). This stipulation has received criticism. The main criticism points to the ambiguity of the magnitude of adequate correlation coefficients and the reliance on correlations that do not allow for quantification of the amount of the specific variance in the data to make inferences about underlying dimensions, such as trait and method factors (Bachman, 2004; Marsh, 1989; Widaman, 1985). Other abilities or test methods may affect the testtaker’s performance. Another criticism is related to failure to separate method variance from random error in the MTMM correlation matrix (Brown, 2006; Schmitt & Stults, 1986). Although a comparison of the magnitude or strength of the correlations across traits and methods provides valuable information, a challenge lies in the absence of criteria to determine the magnitude of correlations necessary to claim significant differences (Bachman, 2004).