On the Notion of “Trivial” Misspecifications in Measurement Models: UFA (Unrestricted Factor Analysis) vs. CFA (Confirmatory Factor Analysis)

• In Event: Sunday Roundtable Session 11:25 am
  In Roundtable Session: Measurement Issues in Structural Equation Modeling (Table 13)

Abstract

Compared to traditional confirmatory factor analysis (CFA), unrestricted factor analysis (UFA) results in less structural parameter bias and generally better data-model fit. However, when model fit is reasonable for CFA (over UFA), CFA should be preferred on the basis of parsimony. Using simulations, the current study examined the sensitivity of CFI, RMSEA, and RMSEAD in adjudicating between UFA and CFA. Results showed that 1) incorrectly specified CFAs demonstrated good stand-alone fit but were rejected often in nested model comparisons with UFA, and 2) CFAs with negligible factor correlation bias <|.10| often failed the model equivalence test while those with non-ignorable bias >|.30| passed. This disconnect is shown to be associated with cross loading condition effects.

Authors

• Timothy R. Konold, University of Virginia

• Elizabeth A. Sanders, University of Washington