This difficulty is called multicollinearity. Google this term for ideas. Best, David Booth

Moderators are nearly always correlated. Dont sweat it. You can see about centering, but generally not a problem. You couldnused median-split moderation test in SEM with equalitty conatraint imposed across 2 grouos low/high.

Hope this helps

Matt

@Matthew I have to say that I don't agree. In any ols regression, which is what SEM is, probability statements are corrupted with correlated IVs. See any linear Stat models book. Best wishes, D. Booth

When you say, "variables a, b, and c are all significantly correlated with one another" do you mean "statistical significance"? Statistical significance just means that we think the correlation is not zero. In your context, you're more interested in whether the correlation is high as well as significant. High enough to make your estimates unreliable which, as David Eugene Booth suggested, is the result of multicollinearity. There isn't a fixed value for "high enough": IV's and Moderators with low multicollinearity will still create problems if the overall model coefficients are also low.

One approach is to see what happens to coefficients when you remove one of your moderators. If they change a lot then you have a multicollinearity problem. There are statistical tests to check if an apparent change in estimates is "statistically significant". Refer to a linear stat models book.

Thank you for the responses David Eugene Booth Hume F. Winzar Matthew Kerry

They are greatly appreciated!

Hello Edward Noon , just to add that you may also need to theoretically look at the face validity of the moderating variables. Do they theoretically seem likely to correlate? If they do, then there is already a problem and combining them in a single model may not have been the best. But if they theoretically don't seem likely to correlate then, come on, it's a doctoral thesis, you still have an argument. Your bargaining power will be key. But I am also thinking, I might be wrong, that if the multicollinearity is high as suggested by David Eugene Booth and Hume F. Winzar , then you have something else to work on besides your thesis, to argue for the possible similarity between the measures of your moderators (i.e., they may be measuring the same concept). I don't know if I make sense.

Don't remove just report it based on your statistical analysis which higher, moderate or low... Correct me if i'am wrong.. Tq