There is an increase in the use of bifactor modeling to test for general factors in multidimensional scales. However, its use seems to make sense only when there are high correlations between first-order factors. How large do you think these correlations should be in order to make bifactor modeling a reasonable approach?
Dear Valencia,
As far as I know, there is no consensus in the literature regarding minimum correlations among factors for using bi-factor modeling. However, most of the research using bi-factor modeling tries to explain a latent construct typically failed to provide a good fit in typical CFA or theoretical structures nested in nature such as intelligence, or testing method effects. However, at worst, moderate to strong correlations may be preferable ranges in .30 to .70
Bifactor models are appropriate to examine when a strong common trait is theoretically expected from a measure despite multidimensionality. This includes measures that have demonstrated an adequate SEM fit for correlated-factors and second-order models (Reise, 2012). Like Ertuğrul Şahin, I don't know of a specific cutoff point. Though I've been told to examine bifactor models of existing measures if there are moderate correlations (r = .30) or higher between the subfactors. In addition, a bifactor CFA with follow-up model-based reliability estimates is needed to know whether you can use the total score of a multidimensional instrument (Reise, Bonifay, & Haviland, 2013).
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling