Hello all,
I am running into a problem I have not encountered before with my mediation analyses. I am running a simple mediation X > M > Y in R.
Generally, I concur that the total effect does not have to be significant for there to be a mediation effect, and in the case I am describing, this would be a logical occurence, since the effects of path a and b are both significant and respectively are -.142 and .140, thus resulting in a 'null-effect' for the total effect.
However, my 'c path X > Y is not 'non-significant' as I would expect, rather, the regression does not fit (see below) :
(Residual standard error: 0.281 on 196 degrees of freedom Multiple R-squared: 0.005521, Adjusted R-squared: 0.0004468 F-statistic: 1.088 on 1 and 196 DF, p-value: 0.2982).
Usually I would say you cannot interpret models that do not fit, and since this path is part of my model, I hesitate to interpret the mediation at all. However, the other paths do fit and are significant. Could the non-fitting also be a result of the paths cancelling one another?
Note: I am running bootstrapped results for the indirect effects, but the code does utilize the 'total effect' path, which does not fit on its own, therefore I am concerned.
Note 2: I am working with a clinical sample, therefore the samplesize is not as great as I'd like group 1: 119; group2: 79 (N = 198).
Please let me know if additional information is needed and thank you in advance!
I'm not sure I understand the problem that you are seeing. If your c' path is not significant, but your a and b paths (and the indirect effect a*b) are, then this simply means that you have "full mediation", i.e., the effect of X on Y is completely mediated through M. This is not at all problematic. In fact, it is often what people like to see.
I am not sure, if this is a typo, but you wrote " 'c" did you mean c' the direct effect?
Your indirect effect is -.142 * .140 = -0.01998, but that does not automatically implie that the total effect needs to be zero. The total effect is the indirect effect + the direct effect c' (that's why I am asking, since it is not clear to me). So, magnitude and sign of the total effect depends on the direct effect. Maybe you could share more information?
Additionally, you should keep in mind that mediation, suppression and redundancy are closely related and you should check, what is happening with your data. See the papers by MacKinnon.
MacKinnon, D. P., Krull, J. L., & Lockwood, C. M. (2000). Equivalence of the mediation, confounding and suppression effect. Prevention science, 1, 173-181.
MacKinnon, D. P., & Lamp, S. J. (2021). A unification of mediator, confounder, and collider effects. Prevention Science, 22(8), 1185-1193.
Christian Geiser I wish that were the case! It is not that my c' path is not significant, it is that the regression model for this path does not fit (the model is not significant therefore i cannot know whether the relationship X > Y is significant or not, since a badly fitting model's parameters should not be interpreted; even if I did interpret this model, the predictor X is not significant, though)
Rainer Duesing That was indeed a typo ! Sorry!
I've also found that my R-code does not need the direct effect to calculate the mediation after all (logical since the mediation contains indirect paths a and b). Nonetheless, can I interpret my mediation analysis when I am not 'allowed' to interpret my direct effect, as it does not fit the data? Is it possible for a regression of X > Y not to fit the data but a regression of X > M > Y to fit the data (with very small effects I might add)? If there was mediation, would we not expect X > Y model to fit the data but for X's effect on Y to simply not be significant?
I'll look into the papers Rainer suggested and see if that shines some light !
Somehow it is not clear to my, what you mean with "does not fit"? Could you please provide the output of the whole analysis? I think this would be helpful.
- Badges
- Science method