Dear community,
my model has 4 formative and 2 reflective LVs with direct impact on the preceived benefit of the goods in question. The questionnaire was answered by approx. 360 participants of an online panel.
Now I have a problem with the formative LVs. They consist of 6-10 indicators.
After running the Smart PLS Algorithm, each of the formative LVs show some indicators with negative weights. Multicollinearity is below threshold value (VIF < 3,3) but maybe there is a small multicollinearity issue in the measurement models.
As recommendet elswhere in this forum and in COHEN et. al (2003), p. 428 f., I extracted 3-4 principal componens per LV (factor extraction with VARIMAX rotation), saved their values and used them as indicators for the formative LVs with zero collinearity. Unfortunately the extracted factors still show negative regression weights. I am at a loss with the possible reasons for the negative weights.
Does anybody has an idea what to do?
Many thanks in advance and best regards
Wilhelm
negative weights after principal components regression
-
- PLS Junior User
- Posts: 2
- Joined: Mon Mar 07, 2011 10:17 am
- Real name and title:
Hello Wilhelm,
if you still have negative regression weights after eleminating multicollinearity maybe some indicators are "negatively" formulated (e.g. "is expensive" when the latent variable measures cheapness).
You can also check this by looking at the factor loadings (Rotated Component Matrix) of your factor analysis.
Regards,
Karsten
if you still have negative regression weights after eleminating multicollinearity maybe some indicators are "negatively" formulated (e.g. "is expensive" when the latent variable measures cheapness).
You can also check this by looking at the factor loadings (Rotated Component Matrix) of your factor analysis.
Regards,
Karsten