Hi all,
I know this topic has been discussed for several times in this forum, so please apologize that I bring it up again: negative and positive indicator weights with formative constructs.
I read all the already existing entries and they really helped a lot to get an understanding of the issue, but I´m currently struggling how to proceed concretely ...
In the middle of my model is a second order construct (type II): 8 dimensions, each measured with 4 to 6 indicators.
What I did so far:
1) I did a PCA (Varimax) in SPSS and safed the factor values to proceed with them as formative indicators of the (now) first order construct in Smart PLS.
2) To check for any multicollinearity issues, I calculated the VIF (in SPSS) and also checked the correlations of the 8 indicators. Both values seem to be ok: VIFs are below 3, and correlations are below 0.9.
However, if I calculate the model in SmartPLS, I get negative and positive weights for the 8 indicators. So as multicollinearity is maybe not the big - resp. only - problem, there may be a suppressor effect (cp. the article from Cenfetelli/Bassilier (2009), p. 691ff).
So my questions are:
1) How should I go on now with my data? What to do/check next in SPSS or SmartPLS?
2) How can I concretely check and calculate if there is a suppressor effect indeed? I assume in SPSS, but how?
3) It was mentioned in this forum, that you can get rid of the negative weights when using ModePLS / SPAD PLS (viewtopic.php?t=117&highlight=negative+ ... +formative). Is this software for free? Where can I download it? Or where can I buy it? Would you recommend to switch to it in such a case?
4) Or should I "simply" get back to SPSS and do a regression for hypotheses testing instead of SEM?
I also already modeled the second order construct with only 4 resp. 2 indicators/dimensions (while keeping all the items) and did the same procedure: PCA in SPSS, checking the VIF and the correlations ... Values are again ok, but still negative and positive indicator weights.
Any recommendation and help on this issue would be very much appreciated!
Thanks already in advance,
Ben