Mediation Analysis - Stata sem and medsem commands with fixed effects models

[Rui Agostinho]

Hello everyone,


I am trying to perform a mediation analysis, in which the variables of concern are as follows:

Dependent Variable (Y) - Percentage_changes_per_industry

Independent Variable (X) - tenure_median

Mediator Variable (M) - median_tenure_promo_top2sf_1st

Control Variables - median_age_promo_top2sf_1st gender log_nemp_median numb_firms_div1000 vn_per_employee_median higher_education norte algarve centro alentejo acores madeira


The issue I am having is that I want to include fixed effects models, but do not know how to do so when considering a mediation analysis. The commands for each of the individual regressions would be as follows:

First, regression with the Mediator as the Dependent Variable:

xtset caem2 year

xtreg median_tenure_promo_top2sf_1st  tenure_median median_age_promo_top2sf_1st  gender log_nemp_median numb_firms_div1000 vn_per_employee_median   higher_education  i.year  norte algarve centro alentejo acores madeira, fe cluster(caem2)

Second, regression with the Dependent Variable as the Dependent Variable:

xtset caem2 year

xtreg Percentage_changes_per_industry median_tenure_promo_top2sf_1st tenure_median median_age_promo_top2sf_1st gender log_nemp_median numb_firms_div1000 vn_per_employee_median   higher_education  i.year  norte algarve centro alentejo acores madeira, fe cluster(caem2)

The sem/medsem commands I am using, to perform the actual mediation analysis (and obtain values for the Sobel tests as well), are as follows:

sem    (median_tenure_promo_top2sf_1st <- tenure_median    median_age_promo_top2sf_1st   gender log_nemp_median numb_firms_div1000 vn_per_employee_median   higher_education   norte algarve centro alentejo acores madeira)(Percentage_changes_per_industry <- median_tenure_promo_top2sf_1st tenure_median    median_age_promo_top2sf_1st   gender log_nemp_median numb_firms_div1000 vn_per_employee_median   higher_education   norte algarve centro alentejo acores madeira) , nocapslatent vce(bootstrap, reps(100))
   

medsem, indep(tenure_median) med(median_tenure_promo_top2sf_1st) dep(Percentage_changes_per_industry) mcreps(1000) rit rid

The issue I have is that these sem/medsem commands are performing regular OLS regressions, and not fixed effects (on caem2) regressions, as I would like them to. Is there any way of implementing a fixed effects regressions in the sem models? It would help me solve the issue I am having with implementing these analyses on Stata.


Thank you very much in advance!
Rui

[Lana La]

Hi Rul. I am doing mediation analysis with multiple mediators using panel data and having the same problem as you. I don't know how to include fixed effects with sem/medsem commands. Did you find solutions for the problem? I would appreciate it if you could share with me your solutions. Thank you.