Hello Yufei,

for the full mediation hypothesis, the absence of a direct effect is the key issue. In my point of view, this is best tested in a real SEM (not PLS) where you can fix the direct effect to zero and test whether this model fits. Sometimes, people contrast that model with a less restrictive one, in which the direct effect is estimated (but significant). In case of a full mediation, the total effect equals the indirect effect.

Much more important is

a) the significance of the indirect effect

b) whether the causal assumptions hold (the latter unfortunately does not get much attention in the behavioral sciences).

I wrote a bit about the latter here:

https://www.researchgate.net/post/I_want_to_know_the_assumptions_of_multiple_mediation_analysis_with_AMOS_output_Can_someone_give_any_guidelines

and here

https://www.researchgate.net/post/Structural_equation_modelling_measurement_model_and_structural_model_yield_the_same_output?isAnswerFieldFocused=true

HTH

--Holger

@Yufei Huang

In case of full mediation, when indirect effect (ab) is significant and direct effect (c') is insignificant, total effect (ab+c') still should be significant. As far as the concept of mediation is concerned, it is meant for enhancing an existing or establishing a significant relationship between X and Y via mediating effect of M.

Imran Anwar Holger Steinmetz Thank you so much for your contributions !