I have a longitudinal latent model with 1 mediator X-->M--->Y and 2 moderators Z and W

The two moderators Z and W are correlated (.66, they are two dimensions of a construct).

The hypotheses is that the moderator Z buffers the X--->M path and

the moderator W buffers the M--->Y path.

Because of the correlation between the 2 moderators and because they are two dimensions of a broader construct when I run the SEM latent moderated mediational model I thought it would have been more appropriate to include also the not hypothesised moderators so that both Z and W would buffer both X-->M and M-->Y. The results confirmed my hypotheses that Z and W buffered only the expected path.

I just got the R&R from a journal and the reviewer suggests to not include the not-hypothesised moderators and to run two alternative models: one with the moderators only in the expected place and the other with the moderators only in the not expected. The model they suggests works anyway but I think it is not correct to test the moderation without controlling for the other.

What is your view? Do you have any reference/textbook exploring the issue of correlated moderators in a double moderated mediational model?

Thanks in advance

Roberta