Multiple treatment/exposure variables in mediation analysis?

[WJ Wang]

Good morning all,

I am currently working on some data that might be best understood through mediation analysis. I have a binary outcome variable, a set of continuous mediators, a group of covariates, and multiple treatment/exposure variables. My original thought was to create an index with the group of treatment/exposure variables, but that did not work out. So far I've searched -gsem-, -medeff-, and -paramed-, and none of them seems to provide an option for multiple treatment/exposure variables. I was wondering if there is any Stata package available for mediation analysis with multiple treatment/exposure variables? If not, what might be some Stata techniques that can handle this situation?

Thanks a lot in advance!

[Mike Lacy]

The community-contributed program -khb- (-ssc describe khb-) will handle multiple continuous mediators with a binary response variable.

[WJ Wang]

The community-contributed program -khb- (-ssc describe khb-) will handle multiple continuous mediators with a binary response variable.

Thanks so much Mike! This is yet another package to explore and I really appreciate the info! However, can this package handle multiple treatment/exposure variables?

[Mike Lacy]

I don't happen to have done this before, but after consulting -help khb-, I have created an example that is substantively silly, but which I believe does what you want:

sysuse auto
khb logit foreign headroom trunk || weight length, disentangle

Decomposition using the KHB-Method

Model-Type:  logit                                 Number of obs     =      74
Variables of Interest: headroom trunk              Pseudo R2         =    0.36
Z-variable(s): weight length
------------------------------------------------------------------------------
     foreign |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
headroom     |
     Reduced |  -.2920597    .556822    -0.52   0.600    -1.383411    .7992914
        Full |  -.1234467   .5573297    -0.22   0.825    -1.215793    .9688994
        Diff |  -.1686129   .2925238    -0.58   0.564    -.7419491    .4047232
-------------+----------------------------------------------------------------
trunk        |
     Reduced |  -.2592075   .1160622    -2.23   0.026    -.4866852   -.0317297
        Full |   .0431317   .1240469     0.35   0.728    -.1999957    .2862591
        Diff |  -.3023391   .3007911    -1.01   0.315    -.8918789    .2872006
------------------------------------------------------------------------------

Components of Difference

      Z-Variable |      Coef    Std_Err     P_Diff  P_Reduced  
    -------------+---------------------------------------------
    headroom     |                                             
          weight | -.1776006   .3154841     105.33      60.81  
          length |  .0089876   .0923432      -5.33      -3.08  
    trunk        |                                             
          weight | -.3217269   .2003008     106.41     124.12  
          length |  .0193878   .1965004      -6.41      -7.48  
    -----------------------------------------------------------

[WJ Wang]

I don't happen to have done this before, but after consulting -help khb-, I have created an example that is substantively silly, but which I believe does what you want:

sysuse auto
khb logit foreign headroom trunk || weight length, disentangle

Decomposition using the KHB-Method

Model-Type: logit Number of obs = 74
Variables of Interest: headroom trunk Pseudo R2 = 0.36
Z-variable(s): weight length
------------------------------------------------------------------------------
foreign | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
headroom |
Reduced | -.2920597 .556822 -0.52 0.600 -1.383411 .7992914
Full | -.1234467 .5573297 -0.22 0.825 -1.215793 .9688994
Diff | -.1686129 .2925238 -0.58 0.564 -.7419491 .4047232
-------------+----------------------------------------------------------------
trunk |
Reduced | -.2592075 .1160622 -2.23 0.026 -.4866852 -.0317297
Full | .0431317 .1240469 0.35 0.728 -.1999957 .2862591
Diff | -.3023391 .3007911 -1.01 0.315 -.8918789 .2872006
------------------------------------------------------------------------------

Components of Difference

Z-Variable | Coef Std_Err P_Diff P_Reduced
-------------+---------------------------------------------
headroom |
weight | -.1776006 .3154841 105.33 60.81
length | .0089876 .0923432 -5.33 -3.08
trunk |
weight | -.3217269 .2003008 106.41 124.12
length | .0193878 .1965004 -6.41 -7.48
-----------------------------------------------------------

Thanks so much Mike! I'll check the help document and see it does the job. Thanks again for your help!