|
|
Construct discrimination validation |
|
Message/Author |
|
yao lu posted on Wednesday, February 22, 2012 - 12:03 pm
|
|
|
Hi Drs Muthen, I read that in CFA, when two factors are highly correlated, it indicates very low discrimination, i.e., combining two factors to one is suggested. In my study, F1 and F2 generates a high correlation of 0.89. However, I also hypothesize that F1 and F2 has a significant positive relationship. So my question is when two factors are hypothesized to be significantly correlated, would a high correlation between them wont be an issue in terms of discrimination validity? Thank you very much for your help. |
|
|
This question is more appropriate for a general discussion forum like SEMNET. |
|
|
A high correlation is no problem for discriminant validity as long as the difference between the variables can be measured reliably. Height and Weight are highly correlated, for example, but the difference between them is easy to measure. The issue is just that psychometric measurement is typically so imprecise that it will be very difficult to measure differences between variables with a latent correlation of .90, at least at the individual level. Perhaps focus on profile differences across groups. |
|
Back to top |
|
|