If you are doing a standard regression analysis, enter the variables into the equation in three "blocks." The first group would be the background variables, then the second would be the variables with direct and indirect effects, and the third block would be the mediating variables.

If you are correct that there are "direct" effects, you will see stronger coefficients when the second block of variables are added, which will be reduced when the final block of mediating variables are introduced.

This kind of "hierarchical entry" of variables into regression models is widely reported, so you should be able to find examples when you look at the kinds of journals where you are hoping to publish your results.

Thanks for your valuable answer.

Hi Mohamed,

as any other model, mediation models (X --> M-->Y) are a target for misspecifications and bias of direct and indirect effects due to potential 

a) confounding (a set Z of omitted variables) affecting either X and M or M and Y or all three)

b) omitted covariates of both X and M

c) Reverse causation of M (on X) and Y (on M)

Some of these misspecifications will be detected if you analyze the model with SEM, some will be unnoticed. Hence the best ways is to draw a path diagram / causal graph, before study is conducted - containing your and the literatures view on influences on M and Y and incorporate those variables as controls that may introduce bias as confounders or omitted covariates. 

Furthermore, instruments for X and M can even increase the possibilties to exclude a, b, and c above.

In this regard, some knowledge on path tracing rues and d-separation / conditional independence can be very helpful.

See this brilliant introductory paper bei Felix Elwerthttp://www.ssc.wisc.edu/soc/faculty/pages/docs/elwert/Elwert%202013.pdf

Best,

Holger

Thank you.

Maybe we could choose to install INDIRECT or PROCESS Macros of Hayes into SPSS to explain the mediation effect with covariates in model .

from which block of mediation analysis I should interpret the Beta values of control variable