signs of path coefficients

[MaKre]

I found a typical mediator model with my data: A -> B is better explained by A->C->B. The beta between C and B is negative which was expected. The path between A and B is positive, which was also expected. However, the beta between A and C is positive which was not expected. Moreover, the correlations between A-variables and C-variables are negative (p < .05). How can a negative correlation between two types of variables turn into a positive beta between two latent variables when using SmartPLS (Version 3)? A (formative) and B (reflective) are single item constructs and C is a reflective measurement model with 5 indicators.

Thank you for having a look into this problem.

Kind regards

[Hengkov]

Hi,

There are only two problems that might cause this to happen. First, you act as a moderator variable suppressor and Second, is the collinearity problem that may occur in the formative indicators.

Greetings,