I assume that you did an a priori power analysis so that the non-significance is not due to lack of power/type-II error. Then you do neither have a indirect effect, nor a mediation in classical terms (Baron & Kenny approach). As you pointed out, you have not enough evidence to assume that there is a relationship between X and M, which may also cause the non-significance of the indirect effect and hence not enough evidence to support an indirect effect. Therefore, M might be a predictor for Y but not a mediator between X and Y.

Hello Özge,

beyond what Rainer mentioned: Mediation (from a causal perspective) is a difficult issue: Could you please

a) provide a short description of the involved variables (including controls)

b) present a path diagram in which you locate all effects that you theoretically assume (again: including the controls and their effects)

c) provide a table of the estimated effects (a, b, c, c')?

If you like, you can send me (e.g., openly or via email) a covariance matrix and I run a model.

Sometimes, weird and/or significant/non-significant effects have a statistical reason, --often, however, they result from biases due to errors in the causal model

Let's find out :)

Best,

Holger

Holger's idea is good, I also would appreciate to see the data. Implying causation is a whole other topic. A significant (statistical) mediation model is not sufficient to claim causality.

Hi Rainer, yes, absolutely, and a (saturated) regression-type model does not inform you about potential errors. Especially her sentence that the x->m effect ("a") is significant with control but not without them "triggered me" as that either stems from patterns of relationships with the x variable (involving some negative ones) or some specification errors (e.g., collider bias). This is important for the question whether controls should be adjusted (-> correct sign. effect) or not (avoiding incorrect sign. effect).

Best,

Holger

Hi, I have a similar question to the one above: https://www.researchgate.net/post/What_methodical_reasons_could_explain_an_insignificant_indirect_effect_within_a_simple_mediation_model

Would you be open to sharing your reflections on my question as well?

Thank you in advance!