I am trying to evaluate the psychometric properties of a questionnaire instrument using bifactor SEM. A bifactor model with five group factors has the best fit, and this model has “good” RMSEA and SRMR. However, CFI and TLI are low (around .85). The psychometric coefficients derived from this model (e.g., omega hierarchical) paint a pretty dismal picture of how the instrument is performing. So, how do I interpret this? I can see three possible conclusions and would appreciate your help sorting them out:
- I trust the model despite its lower CFI/TLI and conclude that the instrument is psychometrically problematic
- I distrust the model due to its lower CFI/TLI and conclude that I don't know what the instrument's psychometric performance is
- I assume that a good instrument “should” fit a bifactor model and take its lower CFI/TLI as evidence of poor psychometric performance
Thanks in advance!