ABSTRACT
Typically, in multitrait-multimethod (MTMM) analyses, the focus is on the traits: Is there convergent validity (i.e., there is trait variance) and is there discriminant validity (i.e., the trait correlations are not too large)? The focus in this presentation is on method effects and adopts two strategies: First, the meaning of methods is assumed to depend on what sort of method is used: times, informants, or instruments. Second, the traits are multiple indicators of a single trait. For time as method, the case is made for a multiplicative MTMM as the method of analysis, where method-method correlations reflect the stability of the trait. For informant as method, the social relations model is used as a framework. Method variance can be partitioned into two parts: how the informant generally views targets, and how the informant uniquely views the particular target. When informants are indistinguishable, method effects must be assumed independent across informants, but when distinguishable they may be correlated. Instrument effects are a highly heterogeneous category, and some preliminary statements are made. In summary, method effects should be viewed as being tied to a particular trait, and not as a factor that operates in the same way across multiple traits. Moreover, both the meaning and analysis of method effects depends very much on the type of method.
