Dear Statalisters,

I hope you can help me with the following question, that I have regarding the khb method (https://journals.sagepub.com/doi/pdf...867X1101100306). In the following example I want to decompose the effect of parental social status on school graduation. I want to know which of my mediators (academic ability and sex) contributes most to the confounding. To answer the question I use the disentangle option, which requests an additional table that shows the contribution of each mediators separately. I also use the ape option to report average partial effects.

The last column of the disentangle table (Components of Difference) shows how much of the total effect of the parental social status (2.isei) is due to confounding of the respective mediator; for the case that the social status has the value 2 this last column sums up to 17.06% (17.41-0.35). But the second table (Summary of confounding) shows that the total percentage of confounding for 2.isei is 7.81% not 17.06%. If use Odds-Ratios instead of the ape option, both table show equal percentages.

So my question are:
  1. Why is it not possible to sum up the last column of the last table (Compontens of Difference) to get the total percentage of confounding?
  2. What can I do to know how much each of my mediators contriubte in to the confounding?
I hope that my questions are clear enough.

HTML Code:
. khb logit school i.isei || c.abil i.sex, disentangle summary ape

Decomposition using the APE-Method

Model-Type:  logit                                 Number of obs     =    2934
Variables of Interest: i.isei                      Pseudo R2         =    0.13
Z-variable(s): abil i.sex
------------------------------------------------------------------------------
      school |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
2.isei       |
     Reduced |   .0539641   .0144083     3.75   0.000     .0257243    .0822038
        Full |   .0497509   .0151335     3.29   0.001     .0200898     .079412
        Diff |   .0042131          .        .       .            .           .
-------------+----------------------------------------------------------------
3.isei       |
     Reduced |   .2745123   .0173679    15.81   0.000     .2404718    .3085528
        Full |   .2352349    .017566    13.39   0.000     .2008062    .2696636
        Diff |   .0392774          .        .       .            .           .
------------------------------------------------------------------------------
 Note: Standard errors of difference not known for APE method

Summary of confounding

        Variable | Conf_ratio    Conf_Pct   Dist_Sens  
    -------------+-------------------------------------
         1b.isei |          .           .           .  
          2.isei |  1.0846845        7.81   1.0199339  
          3.isei |  1.1669711       14.31   1.0053791  
    ---------------------------------------------------

Components of Difference

      Z-Variable |      Coef    Std_Err     P_Diff  P_Reduced  
    -------------+---------------------------------------------
    1b.isei      |                                             
            abil |         0          0          .          .  
         _i_sex2 |         0          0          .          .  
    2.isei       |                                             
            abil |  .0093947   .0043612     222.99      17.41  
         _i_sex2 | -.0001882   .0009283      -4.47      -0.35  
    3.isei       |                                             
            abil |   .036283   .0051923      92.38      13.22  
         _i_sex2 | -.0014555   .0010581      -3.71      -0.53
I´m using Stata 16.1.