Interpretation of user-written khb program output, reduced OR-full OR != diff OR

[Claire Welsh]

I'm using the user-written package khb in Stata15 on Windows 10. My data are sensitive and cannot be posted/copied. My goal is to a) estimate the effect of a potential mediating variable on the relationship between an ordinal DV and an ordinal IV in ordinal logistic regression with fweighted survey data, then b) to disentangle the effects of a number of such significant mediators to find their relative effects on the DV-IV relationship. These potential mediator variables are a mixture of ordinal, binary and continuous. I'm a bit confused about the khb output, however. The 'Reduced OR' minus the 'Full OR' does not equal the 'Diff OR' reported in any of my models, but it does seem to in every help file example I can find. For example,

[CODE]khb ologit fi4 i.SC || i.smok if sex==0 [fweight=fweighted], concomitant(age_w1) or summary[CODE]

gives the output(partial)...

[CODE]Decomposition using the KHB-Method

Model-Type: ologit Number of obs = 5959788
Variables of Interest: i.SC Pseudo R2 = 0.09
Z-variable(s): i.smok
Concomitant: age_w1
------------------------------------------------------------------------------
fi4 | or Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
1.SC | (base outcome)
-------------+----------------------------------------------------------------
2.SC |
Reduced | 2.393325 .0167703 124.54 0.000 2.36068 2.426421
Full | 2.377004 .0166566 123.56 0.000 2.34458 2.409875
Diff | 1.006866 .0009339 7.38 0.000 1.005038 1.008698
-------------+----------------------------------------------------------------
3.SC |
Reduced | 3.991819 .0283347 195.01 0.000 3.936669 4.047743
Full | 3.915686 .0277999 192.26 0.000 3.861576 3.970553
Diff | 1.019443 .0009531 20.60 0.000 1.017577 1.021313
-------------+----------------------------------------------------------------
[CODE]

I may be misunderstanding something, but I thought that the Diff OR would, by definition, be 'Reduced OR'-'Full OR'? Thanks for any help!

[Claire Welsh]

Just for completion here, I contacted KHB's authors and they set me straight here. When specifying the odds ratios to be printed, 'diff' is actually the Reduced/Full OR, not reduced -full as it is without the OR option specified. Makes sense.