Dear Stata Users,

I am using the user-written command KHB to perform a mediation analysis with a multinomial logit.

When I report the results, I show the effects of the reduced (total), full (direct) and difference (indirect) models,
plus the "percentage reduction due to confounding" associated with all mediators (conf_pct) or to each mediator (P_reduced).

However, I am doubtful about how to interpret the results for the mediaton analysis below (meduc is the mediated variable and college and mfamily the mediators).

Firstly, the reduced effect is not statistically signficant at any conventional level (but the indirect effect is). So, I was wondering whether it is even the case of performing mediation.

I know this could be the result of a suppression effect (Zhao et al., 2010) , as it seems to be the case here. However, the percentage of mediation (suppression) results (-1159) and
I am having a hard time in understanding whether this is a reliable result or whether a mediation effect is simply not present in this case.

Therefore, my questions are:

1) Would you report the results of the mediation, even if the total effect is not statistically signficant at conventional level?
2) How would interpret and report to the readers the huge suppression percentage?

Thank you and best regards,
Lydia Palumbo



Code:
Decomposition using the KHB-Method

Model-Type:  mlogit    Number of obs     =    5758
Variables of Interest: ib1.medupar11_3    Pseudo R2         =    0.06
Z-variable(s): i.mcollege i.mfamily311
Concomitant: i.age20 i.religionsp i.ukborn    i.sex i.yowave i.nbrothers i.nchild2 i.lmst5    i.health1
Results for outcome 1 and base outcome 3
    
Robust
famint         Coef.   Std. Err.    z    P>z     [95% Conf. Interval]
    
1.medupar11_3    (base outcome)
    
2.medupar11_3  
Reduced    -.007056    .110393    0.06   0.949    -.2234223    .2093103
Full       -.088845   .1138897    0.78   0.435    -.3120647    .1343748
Diff        .081789   .0317808    2.57   0.010     .0194998    .1440781
    

Summary of confounding

Variable       Conf_ratio    Conf_Pct    Resc_Fact  
    
1b.medupar~3           .           .    .  
2.medupar1~3   .07941924    -1159.14    .91671067  
    

Components of Difference

Z-Variable       Coef    Std_Err    P_Diff  P_Reduced  
    
1b.medupar~3                        
_i_mcollege2          0          0    .          .  
_i_mfam~3112          0          0    .          .  
2.medupar1~3                        
_i_mcollege2  -.0150851   .0135398    -18.44     213.79  
_i_mfam~3112   .0968741   .0281088    118.44   -1372.93