Differences in confirmatory factor analysis model close-fit index estimates obtained from AMOS 4.0 and AMOS 5.0 via full information maximum likelihood – no imputation: Corrections and extension to Palmer et al. (2003)

Abstract

A previous publication (Palmer et al., 2003) provided confirmatory factor analysis (AMOS 4.0) evidence in favour of supporting a three-factor model for the Trait Meta-Mood Scale (TMMS), corresponding to Attention, Clarity, and Repair. The analyses in Palmer et al. (2003) were based on the missing values estimation option in AMOS 4.0. When compared with AMOS 5.0, the incremental close-fit index values were vastly different to those obtained by AMOS 4.0, such that the three-factor model could no longer be accepted. Further CFA modelling identified a nested factor model with a first-order general factor, three first-order substantive factors, corresponding to Attention, Clarity, and Repair, in conjunction with two other first-order factors, which were interpreted as method factors: (a) a negatively keyed method factor; and (b) a method factor that corresponded to three similarly worded items within the Repair subscale. There was evidence to suggest that items 14 and 24 should probably not be used, given their lack of factorial validity. The differences in AMOS 4.0 and 5.0 fit index values were found to be constrained to incremental close-fit indexes (e.g., comparative fit index and Tucker-Lewis index), because of the differences in null model conceptualisation from AMOS 4.0 to 5.0. The results are discussed in light of the possibility that a large number of published studies that have used AMOS 4.0 may have come to grossly inaccurate conclusions.