It is important to understand the distinctions among methods before choosing an approach to any research question under consideration. To this end, I will discuss five general distinctions among methodological approaches: single versus multiple trait methods, single-indicator versus multiple-indicator methods, interchangeable versus structurally different methods, temporal versus nontemporal methods, and metrical versus categorical data methods.
An important first distinction concerns the question of whether methods are appropriate for analyzing one construct or several constructs. To be in line with the terminology of multitrait-multimethod analysis, all constructs will be called "traits" in this chapter. Data-analytic methods for a single trait can be applied to separate a common trait-specific source of variance from systematic method-specific influences and unsystematic measurement error. Consequently, convergent validity as well as method specificity can be estimated. Data-analytic methods for multiple traits additionally allow the analysis of the generalizability of method effects across traits. If one considers different raters to be different methods in one's research, one could analyze, for example, whether a rater bias generalizes across the traits (i.e., whether a rater over- or underestimates all traits of an individual in the same way) or whether there is an interaction between the rater and the trait indicating that the method effect depends on a trait. For example, a rater might overestimate an individual's neuroticism but not his or her extraversión. Only multitrait analyses can detect the generalizability versus trait specificity of method influences. If the method effect generalizes perfectly across traits, measurement error can be separated from systematic method-specific effects. In this case, the different methods serve ?s different indicators for one trait, and the different traits measured by the same method serve as different indicators for the method effect. However, if a method effect does not generalize perfectly across traits and there is only one indicator for a trait-method unit, measurement error and method-specific effects cannot be sufficiently separated (Eid, Lischetzke, Nussbeck, & Trierweiler, 2003). However, this separation is possible if there are multiple indicators for each trait-method unit.
Was this article helpful?