Examples from the help file
The examples in help sgmediation2, run in full
The examples at the end of help sgmediation2, run in order: the basic test, then control variables, survey weights, cluster-robust standard errors, and bootstrapped confidence intervals. The overview page explains what each part of the output is.
use "https://tdmize.github.io/data/data/cda_ah4", clear(cda_ah4.dta | Add Health Wave 4 - | )drop if missing(health, edyrs, income, race, woman, age)(131 observations deleted)sgmediation2 health, iv(edyrs) mv(income)Model with dv regressed on iv (path c)
regress health edyrs , vce()
Source | SS df MS Number of obs = 4,983
-------------+---------------------------------- F(1, 4981) = 264.55
Model | 210.169008 1 210.169008 Prob > F = 0.0000
Residual | 3957.0993 4,981 .794438728 R-squared = 0.0504
-------------+---------------------------------- Adj R-squared = 0.0502
Total | 4167.26831 4,982 .836464936 Root MSE = .89131
------------------------------------------------------------------------------
health | Coefficient Std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
edyrs | 0.089 0.005 16.265 0.000 0.078 0.100
_cons | 2.393 0.079 30.241 0.000 2.238 2.548
------------------------------------------------------------------------------
Model with mediator regressed on iv (path a)
regress income edyrs , vce()
Source | SS df MS Number of obs = 4,983
-------------+---------------------------------- F(1, 4981) = 502.00
Model | 329198.443 1 329198.443 Prob > F = 0.0000
Residual | 3266426.93 4,981 655.777341 R-squared = 0.0916
-------------+---------------------------------- Adj R-squared = 0.0914
Total | 3595625.38 4,982 721.723279 Root MSE = 25.608
------------------------------------------------------------------------------
income | Coefficient Std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
edyrs | 3.519 0.157 22.405 0.000 3.211 3.827
_cons | -17.676 2.273 -7.776 0.000 -22.133 -13.220
------------------------------------------------------------------------------
Model with dv regressed on mediator and iv (paths b and c')
regress health income edyrs , vce()
Source | SS df MS Number of obs = 4,983
-------------+---------------------------------- F(2, 4980) = 168.16
Model | 263.63457 2 131.817285 Prob > F = 0.0000
Residual | 3903.63374 4,980 .783862197 R-squared = 0.0633
-------------+---------------------------------- Adj R-squared = 0.0629
Total | 4167.26831 4,982 .836464936 Root MSE = .88536
------------------------------------------------------------------------------
health | Coefficient Std. err. t P>|t| [95% conf. interval]
-------------+----------------------------------------------------------------
income | 0.004 0.000 8.259 0.000 0.003 0.005
edyrs | 0.075 0.006 13.108 0.000 0.064 0.086
_cons | 2.464 0.079 31.165 0.000 2.309 2.619
------------------------------------------------------------------------------
Sobel-Goodman Mediation Tests
| Est Std_err z P>|z|
---------------------+-----------------------------------------------
Sobel | 0.014 0.002 7.749 0.000
Aroian | 0.014 0.002 7.742 0.000
Goodman | 0.014 0.002 7.756 0.000
Indirect, Direct, and Total Effects
| Est Std_err z P>|z|
---------------------+-----------------------------------------------
a_coefficient | 3.519 0.157 22.405 0.000
b_coefficient | 0.004 0.000 8.259 0.000
Indirect_effect_aXb | 0.014 0.002 7.749 0.000
Direct_effect_c' | 0.075 0.006 13.108 0.000
Total_effect_c | 0.089 0.005 16.265 0.000
Proportion of total effect that is mediated: 0.160
Ratio of indirect to direct effect: 0.191
Ratio of total to direct effect: 1.191Add control variables:
sgmediation2 health, iv(edyrs) mv(income) cv(i.race i.woman age)Model with dv regressed on iv (path c)
regress health edyrs i.race i.woman age, vce()
Source | SS df MS Number of obs = 4,983
-------------+---------------------------------- F(6, 4976) = 56.32
Model | 264.985975 6 44.1643291 Prob > F = 0.0000
Residual | 3902.28234 4,976 .784220727 R-squared = 0.0636
-------------+---------------------------------- Adj R-squared = 0.0625
Total | 4167.26831 4,982 .836464936 Root MSE = .88556
----------------------------------------------------------------------------------
health | Coefficient Std. err. t P>|t| [95% conf. interval]
-----------------+----------------------------------------------------------------
edyrs | 0.093 0.005 16.979 0.000 0.083 0.104
|
race |
Black | -0.111 0.030 -3.747 0.000 -0.169 -0.053
Native American | -0.171 0.145 -1.179 0.238 -0.454 0.113
Asian | -0.201 0.073 -2.735 0.006 -0.345 -0.057
|
woman |
Woman | -0.172 0.025 -6.756 0.000 -0.222 -0.122
age | -0.013 0.007 -1.829 0.068 -0.026 0.001
_cons | 2.817 0.214 13.179 0.000 2.398 3.236
----------------------------------------------------------------------------------
Model with mediator regressed on iv (path a)
regress income edyrs i.race i.woman age, vce()
Source | SS df MS Number of obs = 4,983
-------------+---------------------------------- F(6, 4976) = 171.22
Model | 615297.309 6 102549.551 Prob > F = 0.0000
Residual | 2980328.07 4,976 598.940528 R-squared = 0.1711
-------------+---------------------------------- Adj R-squared = 0.1701
Total | 3595625.38 4,982 721.723279 Root MSE = 24.473
----------------------------------------------------------------------------------
income | Coefficient Std. err. t P>|t| [95% conf. interval]
-----------------+----------------------------------------------------------------
edyrs | 3.836 0.152 25.246 0.000 3.538 4.134
|
race |
Black | -5.922 0.821 -7.215 0.000 -7.531 -4.313
Native American | 0.113 3.997 0.028 0.977 -7.723 7.949
Asian | 4.917 2.030 2.422 0.015 0.937 8.897
|
woman |
Woman | -13.135 0.704 -18.664 0.000 -14.515 -11.755
age | 1.167 0.192 6.086 0.000 0.791 1.543
_cons | -47.033 5.906 -7.963 0.000 -58.612 -35.454
----------------------------------------------------------------------------------
Model with dv regressed on mediator and iv (paths b and c')
regress health income edyrs i.race i.woman age, vce()
Source | SS df MS Number of obs = 4,983
-------------+---------------------------------- F(7, 4975) = 55.12
Model | 299.936161 7 42.848023 Prob > F = 0.0000
Residual | 3867.33215 4,975 .777353196 R-squared = 0.0720
-------------+---------------------------------- Adj R-squared = 0.0707
Total | 4167.26831 4,982 .836464936 Root MSE = .88168
----------------------------------------------------------------------------------
health | Coefficient Std. err. t P>|t| [95% conf. interval]
-----------------+----------------------------------------------------------------
income | 0.003 0.001 6.705 0.000 0.002 0.004
edyrs | 0.080 0.006 13.797 0.000 0.069 0.092
|
race |
Black | -0.091 0.030 -3.061 0.002 -0.149 -0.033
Native American | -0.171 0.144 -1.187 0.235 -0.453 0.111
Asian | -0.218 0.073 -2.975 0.003 -0.361 -0.074
|
woman |
Woman | -0.127 0.026 -4.845 0.000 -0.178 -0.076
age | -0.017 0.007 -2.406 0.016 -0.030 -0.003
_cons | 2.978 0.214 13.906 0.000 2.558 3.398
----------------------------------------------------------------------------------
Sobel-Goodman Mediation Tests
| Est Std_err z P>|z|
---------------------+-----------------------------------------------
Sobel | 0.013 0.002 6.481 0.000
Aroian | 0.013 0.002 6.476 0.000
Goodman | 0.013 0.002 6.485 0.000
Indirect, Direct, and Total Effects
| Est Std_err z P>|z|
---------------------+-----------------------------------------------
a_coefficient | 3.836 0.152 25.246 0.000
b_coefficient | 0.003 0.001 6.705 0.000
Indirect_effect_aXb | 0.013 0.002 6.481 0.000
Direct_effect_c' | 0.080 0.006 13.797 0.000
Total_effect_c | 0.093 0.005 16.979 0.000
Proportion of total effect that is mediated: 0.141
Ratio of indirect to direct effect: 0.164
Ratio of total to direct effect: 1.164Add survey weights already set with svyset:
sgmediation2 health, iv(edyrs) mv(income) cv(i.race i.woman age) prefix(svy:)Model with dv regressed on iv (path c)
svy: regress health edyrs i.race i.woman age, vce()
(running regress on estimation sample)
Survey: Linear regression
Number of strata = 1 Number of obs = 4,983
Number of PSUs = 132 Population size = 21,417,540
Design df = 131
F(6, 126) = 65.63
Prob > F = 0.0000
R-squared = 0.0714
----------------------------------------------------------------------------------
| Linearized
health | Coefficient std. err. t P>|t| [95% conf. interval]
-----------------+----------------------------------------------------------------
edyrs | 0.101 0.006 16.787 0.000 0.089 0.113
|
race |
Black | -0.093 0.042 -2.204 0.029 -0.176 -0.009
Native American | -0.105 0.088 -1.196 0.234 -0.280 0.069
Asian | -0.255 0.086 -2.969 0.004 -0.424 -0.085
|
woman |
Woman | -0.161 0.027 -5.856 0.000 -0.215 -0.106
age | -0.011 0.008 -1.311 0.192 -0.028 0.006
_cons | 2.648 0.280 9.475 0.000 2.095 3.201
----------------------------------------------------------------------------------
Model with mediator regressed on iv (path a)
svy: regress income edyrs i.race i.woman age, vce()
(running regress on estimation sample)
Survey: Linear regression
Number of strata = 1 Number of obs = 4,983
Number of PSUs = 132 Population size = 21,417,540
Design df = 131
F(6, 126) = 118.21
Prob > F = 0.0000
R-squared = 0.1706
----------------------------------------------------------------------------------
| Linearized
income | Coefficient std. err. t P>|t| [95% conf. interval]
-----------------+----------------------------------------------------------------
edyrs | 3.690 0.195 18.899 0.000 3.304 4.077
|
race |
Black | -7.333 1.051 -6.979 0.000 -9.411 -5.254
Native American | 3.006 3.656 0.822 0.412 -4.225 10.238
Asian | 4.844 3.015 1.607 0.111 -1.120 10.807
|
woman |
Woman | -13.312 0.849 -15.682 0.000 -14.991 -11.633
age | 1.230 0.267 4.608 0.000 0.702 1.759
_cons | -46.830 8.828 -5.305 0.000 -64.295 -29.366
----------------------------------------------------------------------------------
Model with dv regressed on mediator and iv (paths b and c')
svy: regress health income edyrs i.race i.woman age, vce()
(running regress on estimation sample)
Survey: Linear regression
Number of strata = 1 Number of obs = 4,983
Number of PSUs = 132 Population size = 21,417,540
Design df = 131
F(7, 125) = 57.38
Prob > F = 0.0000
R-squared = 0.0778
----------------------------------------------------------------------------------
| Linearized
health | Coefficient std. err. t P>|t| [95% conf. interval]
-----------------+----------------------------------------------------------------
income | 0.003 0.001 3.919 0.000 0.001 0.004
edyrs | 0.090 0.007 13.012 0.000 0.076 0.104
|
race |
Black | -0.071 0.041 -1.718 0.088 -0.153 0.011
Native American | -0.114 0.093 -1.234 0.220 -0.298 0.069
Asian | -0.269 0.086 -3.143 0.002 -0.438 -0.100
|
woman |
Woman | -0.121 0.027 -4.441 0.000 -0.175 -0.067
age | -0.015 0.008 -1.761 0.081 -0.031 0.002
_cons | 2.787 0.278 10.010 0.000 2.236 3.338
----------------------------------------------------------------------------------
Sobel-Goodman Mediation Tests
| Est Std_err z P>|z|
---------------------+-----------------------------------------------
Sobel | 0.011 0.003 3.837 0.000
Aroian | 0.011 0.003 3.832 0.000
Goodman | 0.011 0.003 3.842 0.000
Indirect, Direct, and Total Effects
| Est Std_err z P>|z|
---------------------+-----------------------------------------------
a_coefficient | 3.690 0.195 18.899 0.000
b_coefficient | 0.003 0.001 3.919 0.000
Indirect_effect_aXb | 0.011 0.003 3.837 0.000
Direct_effect_c' | 0.090 0.007 13.012 0.000
Total_effect_c | 0.101 0.006 16.787 0.000
Proportion of total effect that is mediated: 0.108
Ratio of indirect to direct effect: 0.121
Ratio of total to direct effect: 1.121Obtain cluster-robust variance estimates for clustering on occcat:
sgmediation2 health, iv(edyrs) mv(income) cv(i.race i.woman age) vce(cluster occcat)Model with dv regressed on iv (path c)
regress health edyrs i.race i.woman age, vce(cluster occcat)
Linear regression Number of obs = 4,959
F(6, 8) = 431.43
Prob > F = 0.0000
R-squared = 0.0638
Root MSE = .88549
(Std. err. adjusted for 9 clusters in occcat)
----------------------------------------------------------------------------------
| Robust
health | Coefficient std. err. t P>|t| [95% conf. interval]
-----------------+----------------------------------------------------------------
edyrs | 0.094 0.005 19.542 0.000 0.083 0.105
|
race |
Black | -0.110 0.040 -2.738 0.026 -0.203 -0.017
Native American | -0.169 0.109 -1.558 0.158 -0.420 0.081
Asian | -0.199 0.116 -1.718 0.124 -0.465 0.068
|
woman |
Woman | -0.172 0.011 -15.251 0.000 -0.198 -0.146
age | -0.012 0.006 -2.098 0.069 -0.026 0.001
_cons | 2.806 0.209 13.419 0.000 2.324 3.289
----------------------------------------------------------------------------------
Model with mediator regressed on iv (path a)
regress income edyrs i.race i.woman age, vce(cluster occcat)
Linear regression Number of obs = 4,959
F(6, 8) = 645.55
Prob > F = 0.0000
R-squared = 0.1711
Root MSE = 24.512
(Std. err. adjusted for 9 clusters in occcat)
----------------------------------------------------------------------------------
| Robust
income | Coefficient std. err. t P>|t| [95% conf. interval]
-----------------+----------------------------------------------------------------
edyrs | 3.838 0.312 12.295 0.000 3.118 4.557
|
race |
Black | -5.885 1.132 -5.197 0.001 -8.497 -3.274
Native American | 0.112 4.084 0.027 0.979 -9.306 9.529
Asian | 4.964 2.969 1.672 0.133 -1.883 11.811
|
woman |
Woman | -13.179 0.697 -18.920 0.000 -14.785 -11.572
age | 1.160 0.307 3.776 0.005 0.452 1.869
_cons | -46.824 11.890 -3.938 0.004 -74.242 -19.405
----------------------------------------------------------------------------------
Model with dv regressed on mediator and iv (paths b and c')
regress health income edyrs i.race i.woman age, vce(cluster occcat)
Linear regression Number of obs = 4,959
F(7, 8) = 753.00
Prob > F = 0.0000
R-squared = 0.0721
Root MSE = .88161
(Std. err. adjusted for 9 clusters in occcat)
----------------------------------------------------------------------------------
| Robust
health | Coefficient std. err. t P>|t| [95% conf. interval]
-----------------+----------------------------------------------------------------
income | 0.003 0.001 6.769 0.000 0.002 0.005
edyrs | 0.080 0.004 18.574 0.000 0.070 0.090
|
race |
Black | -0.090 0.044 -2.061 0.073 -0.191 0.011
Native American | -0.170 0.099 -1.720 0.124 -0.397 0.058
Asian | -0.216 0.122 -1.763 0.116 -0.497 0.066
|
woman |
Woman | -0.127 0.012 -10.308 0.000 -0.155 -0.099
age | -0.016 0.006 -2.734 0.026 -0.030 -0.003
_cons | 2.966 0.203 14.582 0.000 2.497 3.436
----------------------------------------------------------------------------------
Sobel-Goodman Mediation Tests
| Est Std_err z P>|z|
---------------------+-----------------------------------------------
Sobel | 0.013 0.002 5.930 0.000
Aroian | 0.013 0.002 5.915 0.000
Goodman | 0.013 0.002 5.945 0.000
Indirect, Direct, and Total Effects
| Est Std_err z P>|z|
---------------------+-----------------------------------------------
a_coefficient | 3.838 0.312 12.295 0.000
b_coefficient | 0.003 0.001 6.769 0.000
Indirect_effect_aXb | 0.013 0.002 5.930 0.000
Direct_effect_c' | 0.080 0.004 18.574 0.000
Total_effect_c | 0.094 0.005 19.542 0.000
Proportion of total effect that is mediated: 0.140
Ratio of indirect to direct effect: 0.163
Ratio of total to direct effect: 1.163Use bootstrapping to obtain standard errors and confidence intervals. 1,000 replications are recommended for serious use; 100 are used here for illustration.
bootstrap r(ind_eff) r(dir_eff) r(tot_eff), reps(100) seed(2026): ///
sgmediation2 health, iv(edyrs) mv(income) cv(i.race i.woman age)(running sgmediation2 on estimation sample)
regress health edyrs i.race i.woman age, vce()
regress income edyrs i.race i.woman age, vce()
regress health income edyrs i.race i.woman age, vce()
Bootstrap replications (100): .........10.........20.........30.........40.........50.........60....
> .....70.........80.........90.........100 done
Bootstrap results Number of obs = 4,983
Replications = 100
Command: sgmediation2 health, iv(edyrs) mv(income) cv(i.race i.woman age)
_bs_1: r(ind_eff)
_bs_2: r(dir_eff)
_bs_3: r(tot_eff)
------------------------------------------------------------------------------
| Observed Bootstrap Normal-based
| coefficient std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
_bs_1 | 0.013 0.002 7.532 0.000 0.010 0.017
_bs_2 | 0.080 0.005 14.984 0.000 0.070 0.091
_bs_3 | 0.093 0.005 17.746 0.000 0.083 0.104
------------------------------------------------------------------------------Obtain bias-corrected and percentile confidence intervals based on the bootstrapped samples:
estat bootstrap, bc percentileBootstrap results Number of obs = 4,983
Replications = 100
Command: sgmediation2 health, iv(edyrs) mv(income) cv(i.race i.woman age)
_bs_1: r(ind_eff)
_bs_2: r(dir_eff)
_bs_3: r(tot_eff)
------------------------------------------------------------------------------
| Observed Bootstrap
| coefficient Bias std. err. [95% conf. interval]
-------------+----------------------------------------------------------------
_bs_1 | .01313674 .0001615 .00174404 .0100487 .0166817 (P)
| .0095707 .0165165 (BC)
_bs_2 | .08021862 .0003122 .00535347 .0689332 .0904137 (P)
| .0676583 .08868 (BC)
_bs_3 | .09335536 .0004736 .00526056 .0822798 .1035904 (P)
| .0815194 .1035485 (BC)
------------------------------------------------------------------------------
Key: P: Percentile
BC: Bias-corrected