Hello everyone,
I have a doubt interpreting the results obtained when using the sem/medsem commands in Stata.
I am running a model which uses the following variables:
Mediator - median_tenure_promo_top2sf_1st
Dependent Variable - Perc_chang_per_ind_from_more50
Independent Variable - years_at_manag_role_2nd
When I run my mediation code:
The results I obtain state:
From what I understand of mediation analysis, what shows if there is partial mediation, and how much is being mediated, is to see what is the coefficient of the independent variable in a model that does not include the mediator, and then in a model that includes the mediator. If, for models when the mediator is included, the coefficient of the independent variable is lower than the one obtained for models in which the mediator is not included, there is mediation. Please correct me if I am wrong, but this is what I have seen been applied in papers as the difference in coefficients method.
However, for my case, the coefficient I obtain for years_at_manag_role_2nd when running:
is of -.0071366
and when I run:
is of -0.0079394
Therefore, when including the mediator, the coefficient actually increased (in absolute value). The increase is of indeed 11%, but it is an increase, not a decrease. How can I interpret this? How can the results of medsem tell me that all steps of Baron and Kenny approach are significant, and that the mediation be partial, if this is happening?
Any help would be much appreciated!
Thank you very much.
Rui
I have a doubt interpreting the results obtained when using the sem/medsem commands in Stata.
I am running a model which uses the following variables:
Mediator - median_tenure_promo_top2sf_1st
Dependent Variable - Perc_chang_per_ind_from_more50
Independent Variable - years_at_manag_role_2nd
When I run my mediation code:
Code:
sem (median_tenure_promo_top2sf_1st <- years_at_manag_role_2nd ) (Perc_chang_per_ind_from_more50 <- median_tenure_promo_top2sf_1st years_at_manag_role_2nd ) , nocapslatent medsem, indep(years_at_manag_role_2nd) med(median_tenure_promo_top2sf_1st) dep(Perc_chang_per_ind_from_more50) mcreps(500) rit rid
The results I obtain state:
Code:
Baron and Kenny approach to testing mediation
STEP 1 - median_tenure_promo_top2sf_1st:years_at_manag_role_2nd (X -> M) with B=-0.572 and p=0.000
STEP 2 - Perc_chang_per_ind_from_more50:median_tenure_promo_top2sf_1st (M -> Y) with B=-0.001 and p=0.008
STEP 3 - Perc_chang_per_ind_from_more50:years_at_manag_role_2nd (X -> Y) with B=-0.008 and p=0.000
As STEP 1, STEP 2 and STEP 3 as well as the Sobel's test above
are significant the mediation is partial!
RIT = (Indirect effect / Total effect)
(0.001 / 0.007) = 0.112
Meaning that about 11 % of the effect of years_at_manag_role_2nd
on Perc_chang_per_ind_from_more50 is mediated by median_tenure_promo_top2sf_1st!
From what I understand of mediation analysis, what shows if there is partial mediation, and how much is being mediated, is to see what is the coefficient of the independent variable in a model that does not include the mediator, and then in a model that includes the mediator. If, for models when the mediator is included, the coefficient of the independent variable is lower than the one obtained for models in which the mediator is not included, there is mediation. Please correct me if I am wrong, but this is what I have seen been applied in papers as the difference in coefficients method.
However, for my case, the coefficient I obtain for years_at_manag_role_2nd when running:
Code:
xtreg Perc_chang_per_ind_from_more50 years_at_manag_role_2nd , fe cluster(industry)
and when I run:
Code:
xtreg Perc_chang_per_ind_from_more50 years_at_manag_role_2nd median_tenure_promo_top2sf_1st , fe cluster(industry)
Therefore, when including the mediator, the coefficient actually increased (in absolute value). The increase is of indeed 11%, but it is an increase, not a decrease. How can I interpret this? How can the results of medsem tell me that all steps of Baron and Kenny approach are significant, and that the mediation be partial, if this is happening?
Any help would be much appreciated!
Thank you very much.
Rui
