I actually encountered this problem with one of my variables (perceived risks) which is hypothesized to have a negative effect on my other variables. When I ran correlation, it is not significantly correlated with the other variables. However, when I ran convergent and discriminant validity, perceived risks yielded acceptable values. I also tried running my model on SEM and the model fit is also okay despite the insignificant correlation of perceived risks with the other variables.
I'm also new to using SEM so I still need some enlightenment :) Greatly appreciate any advice (and if you could also share some valuable material about this). Thank you!
Hello Charmaine,
correlations are completely uninformative. Create a model that represents your theoretical (=causal) assumptions on the role of your variable.
In addition think about the causal identification strategy. That is: if you propose an effect of X on Y, think about variables that act as confounders of this effect (i.e., variables that have both effects on X and on Y. Next think about variables which have specific effects on X or on Y and can be used to add testable implications or serve as instrumental variables.
A final comment: Good fit and having significant effects are two completely different things. You can have decent fit with no relationship in the mode being signficant. That only shows that the model does not have implications which contradicts the data.
HTH
Holger
Thank you for your very comprehensive answer to my question, Holger Steinmetz . Truly appreciate it.
-Charm
If a variable has non-significant correlations with all other variables in the model, individually (including possible confounding variables), it is generally less likely that it will contribute to the final model. But that depends on the model. It is possible that interactions with other variables could be significant. Also, in theory, the combined effect of a number of other variables on that variable could be statistically significant, even though the individual effects are not significant.
- Badges
- Science method