Precedent for dealing with small sample or multicollinearity

[chrisblocker]

I've concluded the reason 1-2 of the path weights in my results are positive (when they should be negative) is due to multicollinearity.

i.e., when I isolate the variables in separate models the paths are as expected.

My question is this: Is there a precedent in the literature for dealing with a complex model in separate blocks? Or research that discusses how certain variables were isolated in separate analyses to examine effects due to high correlations with other LV's?

In my case, both constructs are both theoretically important and meet the AVE > shared variance criterion for discriminant validity -- and I'd rather not:

a. remove one
b. create a 2nd order LV with both predictors

I also have sample size issues (n=170) when I try to conduct group analyses and analyzing within blocks would make things easier.


Any suggestions would be great! Thank you