Mediation regression

[acue]

Hi all,

I have a question regarding an analysis I just ran. I ran a mediation analysis using a bootstrap approach (requiring me to look at the confidence intervals for significance). Looking at the CI, I found that the independent variable had an indirect effect on the dependent variable through the mediator. When I looked at the individual paths, however, I noticed that the total effect (path c) was not significant. Baron and Kenny would have stopped here (this would have been step 1 for them) and not continued any further. I, however, am not using their approach. Am I safe to say that there is an indirect effect, even if the total effect was not significant because the other paths were significant and the CI did give me evidence that there was an indirect effect?

[Dragan]

You should provide a little more detail. More specifically, what is the p-value for the individual direct path (c). And, when you subsequently include the mediating variable what is the p-value associated with the variable for the direct variable?....it should increase.

[acue]

You should provide a little more detail. More specifically, what is the p-value for the individual direct path (c). And, when you subsequently include the mediating variable what is the p-value associated with the variable for the direct variable?....it should increase.

Below is an example of what I mean:

IV to Mediators (a paths)
Coeff: .1285
se: .0176
t: 7.2811
p: .0000

Direct Effects of Mediators on DV (b paths)
Coeff: .0439
se: .0182
Z: 2.4108
p: .0159
Wald: 5.8117

Total Effect of IV on DV (c path)
Coeff: .0108
se: .0132
Z: .8195
p: .4125
Wald: .6715

Direct Effect of IV on DV (c' path)
Coeff: .0046
se: .0136
Z: .3344
p: .7380
Wald: .1119

Bias Corrected and Accelerated Confidence Intervals
Lower: .0012
Upper: .0109

[richieboy]

Interesting post, could anybody please explain in basic terms how I would test mediation of one variable between two others using regression? Is it hierarchical regression?