Dear researchers
“When factors are correlated, sums of squared loadings cannot be added to obtain a total variance”. I extracted two factors based on Maximum likelihood EFA and promax rotation. There is not cross loading. The factor loading of all items was greater than 0.5 ranging from 0.529 to 0.833. According to “Rotation Sums of Squared Loadings”, eigenvalues for the first and second factor were 11.13 and 9.99 respectively.
How can “% of Variance” of factors be calculated when factors are correlated? (“% of Variance” for each factor and next “Cumulative %”)
Best wishes
Hamid
Hi Hamid,
I just did an EFA on my data using maximum likelihood and promax rotation. I am using SPSS 25. I believe that the program adjusts for the small correlation between the factors in the variance calculations. If someone else has thoughts on this, I would like to hear them as well. Best wishes,
Nick
If loadings are correlated this means that some information is included more than once in your loadings, leading to over estimation of how much variance is actually explained. To calculate the total variance explained by factors 1-x you need to calculate the residual after applying those factors then calculate the sum of the squared intensities. This will give you the unexplained variance remaining so the explained variance will be total variance minus this.
Total variance is the sum of squares for your original data, so simply go back to the original data to calculate the sum of the squared data - that will be your total variance. The rest should be straightforward.
use varimax rotation under principal component analysis in SPSS. the factor that emerge are not correlated and variance can be procured from rotated variance table easily
@ Jeevan Jyoti
Thank you Jeevan
The aim of research was exploratory not PCA! Really the factors were oblique.
If you want to refine factors further, I would suggest to use bifactor model. The correlated factor model can't parsimoniously account for effect of a general factor. Also note that by default, SAS and SPSS use Pearson's correlation for factor analysis. If you are using non-continuous items, your factors may appear multidimensional when in fact they are not. Based on your eigenvalues, I do not think that'd be the case, but something for you to consider. If you are using MPlus or R, this can be easily corrected. Use 'psych' package in R, the output shows amount of variance of each factor.
- Badges
- Science topic
- Similar topics
- Phytochemistry
- Extracts