Can a mediator that is correlated with another mediator ARTIFICIALLY enhance the effect of x on y? Everything I'm reading about suppression suggests that suppressor variables clarify the relationship between x and y. Is it possible that a suppressor variable muddies the water, that the significant relationship that emerges between x and y with the inclusion of M2 is actually just due to its correlation with M1? (Note M1 entered into the model with x and y alone reduced the effect of x on y to the point of insignificance, and M2 is associated with x, y and M1.)
Alternatively, is there another statistical concept that depicts this phenomenon? Can multicollinearity have this effect?