Dear Statalisters,

I have a question on the user-written command khb (https://journals.sagepub.com/doi/pdf...867X1101100306). I often find people highlighting the significance levels of the % mediated by each mediator (if more than one).
For instance in the following example, they would be able to say whether 6.61 is a significant % mediated, although limited.

Code:
Z-Variable    Coef    Std_Err    P_Diff    P_Reduced     
                    
1b.nssec2par                     
_i_mfam~1122    0    0    .    .     
_i_mfam~1123    0    0    .    .     
_i_mfam~1124    0    0    .    .     
_i_mcolle~22    0    0    .    .     
2.nssec2par                     
_i_mfam~1122    .0221344     .0091995    23.18     34.01     
_i_mfam~1123    .0613838     .0146694    64.27     94.33     
_i_mfam~1124    -.0004717    .0011844    -0.49      - 0.72     
_i_mcolle~22    .0124569       .0080881    13.04     19.14     
3.nssec2par                     
_i_mfam~1122    .0317158    .0095893    14.70        6.61     
_i_mfam~1123    .1584903    .0318423    73.47       33.01     
_i_mfam~1124    .0030143    .0042229    1.40          0.63     
_i_mcolle~22      .0224871    .0137921    10.42        4.68
However, I am not so sure how they tested it. My guess is that I just need to use a test of hypothesis with a normal distribution (this is my understanding from the paper).

However, one of my mediators, mfamily112, is a categorical variable with more than one category, so I am not sure whether it is the correct distribution.

I also tried looking at the indirect effect in a model keeping only one category as the mediator (say 3.mfamily112) and the other as confounders ( in the concomitant part) but I have never been able to replicate the results as if I were including all the categories as mediators. Do you have any clue how I can find significant mediation in this case? Thank you and best,
Lydia