2
$\begingroup$

I want to Run CFA model (N=400), but I have a question about the correlation between items. I know I should have high correlation between items that make up one factor (cor>.3). I have two factors Anx ( first 4 items) and Eff ( the rest). Anx items looks good, but some of my efficacy items has correlations >0.8. should I remove those? I looked at VIF s scores :

##   Anx_1.post   Anx_2.post   Anx_3.post   Anx_4.post Eff.1_1.post Eff.1_2.post 
##         2.71         2.55         2.58         3.11         4.62         2.78 
## Eff.1_3.post Eff.1_4.post Eff.1_5.post Eff.2_1.post Eff.2_2.post Eff.2_3.post 
##         5.51         6.02         5.57         4.58         4.83         6.29 
## Eff.2_4.post 
##         1.83

enter image description here

$\endgroup$
4
  • $\begingroup$ Why do you think you should remove them ? What is your research question ? Please provide further details about your study and the data $\endgroup$ Jul 18, 2020 at 15:43
  • $\begingroup$ Hi Robert, thanks a lot for your reply. I am trying to come up with a some sort of score that represent test anxiety (4 survey questions) and self efficacy (9 questions) to use in the models to explain differences in grades. So I decided to do EFA and CFA to determine which questions to use . I have an issue with highly correlated items as I just wondering if items are highly correlated , let say .9, does it mean they measuring almost exactly the same thing, this dropping off one of them remove that redundancy $\endgroup$
    – yuliaUU
    Jul 18, 2020 at 22:07
  • $\begingroup$ I also try to see why my model have quite good TLI and CFI scores but poor rmsea $\endgroup$
    – yuliaUU
    Jul 18, 2020 at 22:07
  • $\begingroup$ Isn't the objective of doing factor analysis is to find "common factors" that can explain several measured events that are often times correlated? $\endgroup$ Jul 19, 2020 at 8:55

1 Answer 1

1
$\begingroup$

You should not remove items of them is correlated highly with another unless you have good reason to believe that they are measuring exactly the same thing and the only difference in the scores for those items is measurement error. If they are measuring the same thing different ways, then by removing one (or using an average is another common mistake) then you are losing information.

Be guided by the underlying theory of what is driving the data generating process along with expert knowledge.

$\endgroup$
2
  • $\begingroup$ thanks a lot. And how can i then explore the multicollinearity? according to this source : strong correlation is strong sign of multicollinearity (edupristine.com/blog/detecting-multicollinearity) cor .90 that we would need to begin worrying about multicollinearity at (according to Knekta). $\endgroup$
    – yuliaUU
    Jul 22, 2020 at 18:30
  • $\begingroup$ That link is talking about regression. You are doing factor analysis. $\endgroup$ Jul 22, 2020 at 19:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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