0

I am fitting a SEM model with 6 exogenous latent variables and 2 endogenous latent variables with paths from each exogenous variable leading to both endogenous variables. All the path coefficients are in line with predictions and correlational analysis. However, one of the path coefficients that I get is significant but with the opposite sign from what I have in the correlation between the two variables. It doesn't look like a suppressor variable to me because although it correlates with other variables, these correlations are not very strong. What can cause the sign in the path relationship to switch?

Svetlana
  • 1
  • It's difficult to answer without knowing more about your theoretical model. This 'sign-flipping' could simply be caused by multicollinearity. Alternatively, there may be substantive / theoretical reason for why the sign flips.

    I think this thread gives a pretty good potential explanation: https://stats.stackexchange.com/a/41645/290668

     – Rak May 17 '21 at 16:19
  • If you could add more information about your problem, one could give a better answer (e.g. model output with correlation and path coefficients). Otherwise, it could just be a normal consequence of conditioning (see Simpson's paradox visualization https://en.wikipedia.org/wiki/Simpson%27s_paradox). – Kuku Jun 05 '21 at 14:12

0 Answers0