I am currently assessing my measurement model which includes - besides other constructs - a 2nd order reflective construct about learning (consists of four 1st order constructs that are also reflective and measure different learning dimensions).
I noticed that some items of the 1st order constructs partly load on the "wrong" factors when conducting an EFA. Thus, establishing discriminant validity of the 1st order constructs is hardly possible. When conducting a CFA, however, every item loads quite well on its assigned construct (1st order and 2nd order). Cronbachs Alpha of 1st order constructs and the 2nd order construct is also good.
I was wondering whether I have to establish the discriminant validity of my 1st order constructs anyway? The reflective 1st order constructs of my reflective 2nd order construct are expected to correlate anyway to some extent, right? In the end, the whole concept is reflective.
Thus, would it be correct to just establish discriminant validity between my 2nd order reflective construct and other constructs in my measurement model (i.e., constructs not related to the 2nd order learning construct)?
A source explaining/mentioning this would also be really helpful.