Hello Raghu,

If you mean: PCA = principal component analysis; PFA = principal factor (one method in the common factor analysis family) analysis; and CFA = confirmatory factor analysis, then:

1. There's almost nothing to be gained by using PCA then PFA. If the number of variables involved is small to modest, PCA tends to bias estimated loadings (variable-component correlations) upward relative to PFA or just about any other common factor analysis method. That could well lead to different conclusions about plausible structures and salient variables within those structures. If the number of variables is quite large, then the resultant estimates of variable-component and variable-factor loadings tend to be much more similar.

2. PCA assumes that: (a) all observed variance is common variance, and (b) no unique or specific or error variance exists. For the data sets I typically encounter (human behavior, performance, and perception/sentiment/opinion), these are unrealistic assumptions.

Is it reasonable to run exploratory factor analysis first (via common factor analysis), then evaluate the stability of the model via confirmatory factor analysis on new or held-out data? Absolutely.

Good luck with your work.

Hi,

Yes, You do that.

Thanks

Yes both PCA and CFA can be done simultaneously.

Yes, Can be done simultaneously.