Hi everyone
I applied both PCA and EFA to 17 variables and it shows that the 17 can be grouped into 3 variables. What is the right way to group the scores? The scores are on a Likert-type scale (1-5) should I add them up or is there a different way to group the score together? I want to apply an ANOVA to the 3 group variables and see the differences in means.
@Abdulrahman You can sum the scores of observed items loading under each factor and form three composite variables. Or you can also form composite variables by averaging the sum of scores for each latent variable. Both methods are appropriate and results won't be affected at all and using any of them, you can run ANOVA. Wishing you best.
Hello Abdulrahman,
The three most common algorithms for computing factor scores (in EFA programs in statistical packages) are: (a) Bartlett; (b) Anderson-Rubin; and (c) Regression. The adding up of scores (whether simple summation, as mentioned by Imran Anwar, or via some loading threshold for inclusion) is common as well. However, this last method won't guarantee orthogonality of scores across independent factors.
Here's a quick reference that walks through a lot of the options for creating factor scores: https://stats.stackexchange.com/questions/126885/methods-to-compute-factor-scores-and-what-is-the-score-coefficient-matrix-in
Good luck with your work.
PCA and EFA are both based on normality presumptions. You can run these methods for a sum of some of these Likert-based scales.