| ||||

SMABS 2004 Home Organization About Jena Sponsors Links Imprint / Contact SMABS Home |
## Contributions: Abstract## SEM analyses contradict intuitive attributes operationalized by balanced self-ratings: the case of state and trait anxiety
Structural equation models (SEMs) allows one to formulate substantive intuitions of psychological attributes operationalized by self-ratings. Here, I focus on attributes defined as hypothetical dimensions, along with which individual differences can be measured using balanced questionnaires. In reality, testing SEMs questions the objectivity of such concepts. First, an attribute suggests congenericity, i.e., models which are rather difficult to fit closely. Second, absence of congenericity may be accounted for by bi-stable models (Vautier et al., 2004). As a special case of Eid's (2000) CT-C(M-1) model, bi-stability stresses substantive indeterminacy of the factors. Third, even congenericity is compatible with multidimensionality of the common factor: A minimal latent state-trait model entails that a common factor can be decomposed as the sum of state residuals, and one trait factor. My goal was to assess the suitability of bi-stable latent state-trait models which integrate data from state and trait questionnaires simultaneously. Two-wave STAI data from Vautier and Jmel (2003) were analyzed using two alternative, bi-stable, comprehensive latent state-trait models, method factors being treated as state factors. Both models provided acceptable goodness of fit, and supported the empirical need for hypothesizing two trait factors, and four state-residual variables by model. Reliability analyses allowed estimating the relative contribution of the various latent variables. I found no statistical argument for rejecting one of both alternatives. Bi-stability of the data is not compatible with their unique interpretation at the ontological level. Uncertainty about the existence of measurable attributes, which was introduced at the beginning of the modeling process, was still present at the end of the process. Successful modeling does not solve the problem of whether self-ratings do measure objective attributes. Consequently, interpretation of an observed composite score seems more a matter of projection, in the psychodynamic sense of the term, than a matter of measurement. |