High covariance between two factors. Is that a problem in a regression?

#1
Hi,

Is it problematic if there is a very high covariance between two factors (independent latent variables) (0.95 standardized).

I thought the implication of two independent variables that covary to a high degree was that they are likely to have almost the same effect on a specific dependent variable. Although when I test it, one factor is positively related to the dependent variable, while the other one is negatively related. Is that possible?

Thank you and best regards,
Nils
 

ledzep

Point Mass at Zero
#2
one factor is positively related to the dependent variable, while the other one is negatively related. Is that possible?
I think it is possible. For example, take cancer, smoking and income.

cancer (y) is positively correlated with smoking [no need to explain this..].
Smoking (x1) is positively correlated with income (x2) [People with constant income source can afford it constantly].
However, cancer may be negatively correlated with income (x2) [as people with high income can afford better care].

So, x1 and x2 are correlated but the effect on y may be different?

May be not so convincing..but sort of..
 

spunky

Doesn't actually exist
#3
why don't you create a model where only one factor loads on your variables instead of two and see what the fit of it is using something like the chi-square difference test or checking out your fit indices?
 
#4
Thanks Ledzep, that actually makes a lot of sense! As for your comment, spunky, the fit actually decreases. I'll stick to my two factors :)