0
$\begingroup$

I have gathered 41 variables that are supposed to explain dependent variable Y in a dataset. Is the following reasonable?

First, I will conduct EFA, reduce the dimension, conduct CFA to confirm/reduce. and then multiple linear regression be logical on the outcomes of the CFA? with the goal of finding a base model in the end

In all cases the same Y will be used across the process: is this correct too?

Can someone please advise me as I am not well versed in statistics?

$\endgroup$

2 Answers 2

3
$\begingroup$

It does not make sense to conduct CFA after EFA on the same data set. You cannot "confirm" (or better: test) a factor structure with the same data set that was used to explore/find that factor structure.

Also, it would be more logical for the outcome analysis to stay within the SEM framework with latent variables/factors (rather than use multiple regression with observed variables). That is, if you are already using latent factors to represent your constructs of interest and address measurement error, then it would be difficult to explain why you later moved back to an analysis that uses observed variables.

$\endgroup$
1
  • $\begingroup$ Thank you to give you some context this is a financial modeling problem (y=price), I already have an idea of the ‘Factors’ from literature so I will just do CFA, the thing is in trying to explain an obserable dependent variable with mostly observables values and 2 latent ones, and actually confirm/comment on the accuracy $\endgroup$
    – Kasere
    Commented Nov 14, 2023 at 14:00
0
$\begingroup$

I agree with Christian's answer (+1) but would add that, 1) EFA is not, strictly speaking, a data reduction method but a way to find latent factors. PCA is a data reduction method. 2) for regression purposes, PCA might not be ideal, because it does not consider the dependent variable at all. You could look into partial least squares regression.

$\endgroup$
2
  • $\begingroup$ Thank you, my goal is to confirm the accuracy of the latent factors from literature. Hence my reasoning for deriving factors with new variables and then doing a regression to observe the results, but im not sure that’s suitable $\endgroup$
    – Kasere
    Commented Nov 14, 2023 at 14:02
  • $\begingroup$ If you want to confirm the latent factors that others found, then you should do CFA and not EFA. $\endgroup$
    – Peter Flom
    Commented Nov 14, 2023 at 14:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.