Multitrait Multimethod Matrix and Construct Validity

Selecting Traits and Methods for an MTMM

The selection of traits and methods for an MTMM analysis should follow a set of guidelines. First, each method should be well suited to evaluating all of the constructs of interest. Second, measurement methods should be as independent of each other as possible. Finally, constructs should be included that are expected to vary in their degree of association. In other words, the matrix should include some constructs postulated to be highly related and some constructs thought to be unrelated. This range of association is recommended in order to provide a basis of comparison for discriminant validity.

Organization of an MTMM

Once MTMM data are collected, correlations among the observed scores are computed and arranged with a specific organization. This can be illustrated with a hypothetical example (see Table 1). This example has three traits (feelings, beliefs, and intentions to act), each of which has been assessed using three measurement methods: a Thurston equal appearing interval scale, a Likert summated ratings scale, and a Semantic Differential scale. Thus, there are nine observations for each person, thereby producing a 9 × 9 correlation matrix.

The MTMM is organized by method, with each of the three constructs embedded in each method block. The interpretation of the MTMM involves identifying the four different types of construct method pairings within the matrix, which take the form of either a diagonal or a triangle. These pairings are evaluated using a set of guidelines.

The first type of construct method pairing is the monotrait heteromethod diagonals, which are[Page 667] correlations between different measures of the same construct. These correlations are in bold and include, for example, the correlation between feelings evaluated by both the Likert Scale and the Thurstone scale (r = .75). These correlations are also known as the validity diagonals.

Table 1 Example of a Multitrait-Multimethod Matrix
Measure		Thurstone			Likert		Semantic	Differential
	F	B	I	F	B	I	F	B	I
Thurstone
Feeling	(.98)
Belief	.62	(.95)
Intentions	.19	.17	(.93)
Likert
Feeling	.75	.60	.18	(.95)
Belief	.59	.86	.17	.60	(.94)
Intentions	.19	.18	.74	.21	.20	(.95)
Semantic Differential
Feeling	.78	.59	.20	.76	.60	.17	(.99)
Belief	.60	.88	.19	.59	.84	.19	.64	(.95)
Intentions	.19	.20	.77	.18	.18	.77	.20	.20	(.92)
Note: Monotrait-monomethod values are in parentheses; heterotrait-monomethod values are underlined; monotrait-heteromethod values are in bold; heterotrait-heteromethod values are unmarked.

The second type of construct method pairing is the heterotrait heteromethod triangles, which are the correlations between different measures of different traits (i.e., correlations between pairings that have neither trait nor measure in common). These correlations are unmarked in the matrix and include, for example, the correlations between feelings evaluated by the Likert scale and beliefs evaluated by the Thurstone scale (r = .60).

...