Multitrait–Multimethod Matrix

Brief History and Introduction

The multitrait–multimethod matrix was established in 1959 by Donald T. Campbell from the Graduate School of Northwestern University and Donald W. Fiske from the Department of Psychology at the University of Chicago. A multitrait–multimethod matrix is used for validation, the process involving accumulating relevant evidence in order to justify measures, to provide a sound scientific basis for interpretations and uses of the results (e.g., test scores), and to establish evidence for construct validity. In this entry, traits and constructs are used interchangeably.

Specifically, a multitrait–multimethod matrix aims to provide convergent and discriminant evidence that demonstrates the relationships of the measure of interest to other measures. Convergent evidence is the degree to which the measure of interest and other measures are intended to assess the same or similar constructs. Discriminant evidence is the degree to which the measures of different constructs are not related in reality. Convergent and discriminant evidence are both required to demonstrate that the measure of interest supports the construct underlying the proposed interpretations and/or uses of the measure. They can both be assessed using a multitrait–multimethod matrix.

Matrix Elements and Validity Requirements

A multitrait–multimethod matrix consists of the correlations when each of several traits is measured by each of several methods. Table 1 presents a synthetic example of the multitrait–multimethod matrix for three traits (i.e., A, B, and C) measured by two methods (i.e., Method 1 and Method 2). In the table, A1 denotes Trait A measured by Method 1, B1 denotes Trait B measured by Method 1, and so on. The matrix has four blocks. The monomethod blocks are at the top left and bottom right, each of which consists of the reliability diagonal with values enclosed by parentheses and the adjacent heterotrait–monomethod triangle enclosed by solid lines. The bottom left is the heteromethod block, which has the validity diagonal with values in bold face, and two heterotrait–heteromethod triangles bordered with the dashed [Page 1118]lines. Note that the two heterotrait–heteromethod triangles are not identical.

A few observations need to be addressed based on this table. First, reliability is the agreement between two measures for the same trait through similar or same methods. The reliabilities could also be viewed as the monotrait–monomethod values. In this synthetic scenario, for example, the reliabilities for the three traits are higher for Method 1 and lower for Method 2.

...