Hi, I have a question about how an adjustment in one end of a model, can lead to large effects in an other end of the model.
I have a model with circa 15 LV:s, 200-300 MV:s and 15-20 paths, on a basis of 150 respondents. Because of the large number of variables, I have used Exploratory factor analysis in SPSS (Oblimin) to group MV:s into factors and use those factors as MV:s
I have a before and after experiment based research structure
The problem I have is that when I do small changes (theoretically allowable) in the set-up of the MV:s for one LV, in order to better the path coefficient, that may lead to a better path coefficient, between the two LV:s I wanted to better the path in between....but...other paths between other LV:s have all of a sudden gone haywire.
The ones I am thinking about are "nodes" when one LV has four-five incoming paths, from other explaining LV:s, and two to three outgoing. I seem to be stuck with a situation where I can make three of the four incoming paths good, but not all four at the same time..
An addition to this, it seems that adding one path into an LV:s kind of "knocks out" the path value on other ingoing paths, is it so that variance is a finite resource in some way?
Why so? What can be done about it?
Very Best Regards