Dear All!
I recently ran an EFA on my data (PCA with Varimax rotation) and obtained a solution with two or three cross-loadings (greater than .32). I again ran an EFA on the same data (this time PAF with oblimin) and obtained a solution with no cross-loadings. Moreover, the factor correlation matrix had three values above .3.
Please, suggest if I should report the values obtained from orthogonal rotation or if it's better to report the results of oblique rotation?
Dear Oshin Gulsia
Selecting the rotation method is a key element of the EFA, two approaches can be applied, which are the following.
Theory Approach:
I would like to recommend you almost always recommend correlated factors (i.e., oblique rotation) for social science data as that depicts the reality more. If it is not the case, then you are recommended to use the uncorrelated factors (i.e, varimax rotation).
Data Approach:
You can select direct obilimin and then check the component correlation matrix if you have all correlations absolute values are greater than (0.32 or some use 0.5) then it means that the factors are strongly correlated and you should stick to with the oblique rotation, but if it is not the case then you should assume that they are orthogonal related.
Best Regards,
Belal Edries
Belal Edries Thank You Sir for clarifying my doubts. The component correlation matrix obtained after running an EFA with direct oblimin consists of some values greater than .32. However, some values are quite low as given in the picture attached herewith. I shall be humbled if you can advise me on the same!
Dear Oshin Gulsia
Based on your component correlation matrix, it indicates that there is a strong correlation between all the factors expect Factor 4. Therefore, since not all the values are greater than 0.32, thus, the results suggest using the uncorrelated factors (i.e, varimax rotation).
Best Regards,
Belal Edries
Belal Edries I would be glad if you can suggest me if the following output results are reliable i.e. the cross loadings are minimum enough to be ignored
Dear Oshin Gulsia
As I can see from your components loadings it seems totally fine, you just may consider increasing the cut off values to 0.4 so you will get cleaner loading per factor.
Best Regards,
Belal Edries
If the results from an orthogonal rotation are problematic (e.g., cross loadings, low communalities, low significance loadings), the researcher can opt for an oblique rotation. Which you did. Oblique rotations are less restricting and they have a feature for correlated factors. This is why your findings were better (you made a correct move). You might want to report both of the findings for the robustness of the analysis. If your sample was small or the ratio of cases per variable was low, you would need to validate the findings from the oblique rotation.
Hair et al. give a very good guidance on the choice of rotation and reporting.
Good luck!
- Badges
- Science method
- Similar topics
- Methodology
- Research Methodology
- Qualitative Research