Page 1 of 1

Path significance

Posted: Thu Jan 17, 2019 6:28 am
by skr
Hello,

While testing a particular structural model to obtain the direct effect of 3 independent variables on the dependent variable, all the path relationships get statistically significant at 5% significance level. Now, when two new constructs are added as the antecedents of that particular dependent variable into the existing model, the existing 3 path relationships fail to become significant. However, the new two path relationships are statistically significant. Could you explain the reason?

Re: Path significance

Posted: Fri Jan 18, 2019 8:40 am
by jmbecker
The new and old variables are collinear, i.e., they covary and thereby share variance. The explained variance of Y of the old variables is now partly or completely explained by the new collinear variables. That makes the old variables become insignificant. To a certain degree and if the new variables are necessary and important it is a good thing (your control for the influence of the new which renders the old insignificant). However, if collinearity levels are severe it could also be a problem because of the variance inflation in the estimates and therefore lower statistical power. The phenomenon is also linked to the phenomenon of suppression.

Re: Path significance

Posted: Fri Jan 18, 2019 8:58 am
by skr
Furthermore, when I went for testing a different structural model by considering those new constructs as the mediator, all the indirect effects become significant (while direct paths are still not significant). Is this a case of full mediation?

Re: Path significance

Posted: Fri Jan 18, 2019 9:10 am
by jmbecker
Yes, that could be. It also shows that there is a dependency between the new and old --> collinearity as I mentioned earlier. If it makes theoretically sense to build a mediation model that could be a good solution.

Re: Path significance

Posted: Fri Jan 18, 2019 9:26 am
by skr
Thank you Dr. Becker.