The applications show that the single-indicator models can provide interesting insights into the MTMM structure. The major limitation of single-indicator models, however, is that measurement error can be separated from systematic trait-specific method effects only in models with method factors and only if method effects generalize across traits in a unidimensional way. This assumption, however, is very restrictive, and trait-specific method effects could be expected in several applications. For example, one peer rater might not consistently over- or underestimate different personality traits of a target person. Trait-specific method effects might be especially likely when the traits differ in their proneness to response sets (e.g., social desirability and leniency effects). If trait-specific method effects exist, reliability will be underestimated in single-indicator models because the effects that are due to trait-specific method effects cannot be separated from the error variable (Eid, 2000; Marsh & Hoce-var, 1988). These problems can only be dealt with appropriately in multiple-indicator models that are described in the next section. Hence, single-indicator models seem to be most appropriate when it is not possible to have multiple indicators for a trait-method unit.
Was this article helpful?