KHB - Question on negative confounding percentage

[Lydia Palumbo]

Dear Stata users,

I have a question on the interpretation of KHB command output. The output is the following.

khb reg futrj ib0.nssec3 || i.mfamily314 [pw=lweight] if (age>15&age<22)&myouthq==1 , summary disentangle concomitant(religion i.ukborn i.sex i.educ3 i.yowave)

Decomposition using Linear Probability Models

Model-Type: regress Number of obs = 9681
Variables of Interest: ib0.nssec3 R-squared = 0.05
Z-variable(s): i.mfamily314
Concomitant: religion i.ukborn i.sex i.educ3 i.yowave
------------------------------------------------------------------------------
| Robust
futrj | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
0.nssec3 | (base outcome)
-------------+----------------------------------------------------------------
1.nssec3 |
Reduced | -4.359166 .6862733 -6.35 0.000 -5.704237 -3.014095
Full | -4.908575 .6929247 -7.08 0.000 -6.266682 -3.550467
Diff | .5494088 .1108678 4.96 0.000 .3321119 .7667058
------------------------------------------------------------------------------

Summary of confounding

Variable | Conf_ratio Conf_Pct Resc_Fact
-------------+-------------------------------------
0b.nssec3 | . . .
1.nssec3 | .88807161 -12.60 1
---------------------------------------------------

I cannot understand the presence of this negative confounding percentage.
Should it represent a suppression effect, such that my mediator, i.mfamily314, increases (by little) the effect of nssec on the dependent variable (futrj), when introduced in the regression ?

If yes, I cannot get it, as my variable nssec3 has a positive effect on the mediator and the mediator has a positive effect on the dependent variable, thereby generating a positive indirect effect. If this positive indirect effect were summed up to the negative direct effect of nssec3 on futrj, I would expect that it results into a smaller total effect, not into a bigger one. Since this expectation may depend on a missinterpretation of KHB output, I would appreciate if someone could clarify.


Thank you very much.
Best regards,
Lydia

[Mike Lacy]

Unfortunately, you have posted your output without choosing a nonproportional font for it or putting it in code delimiters, which makes it very hard to read, so I won't comment on your specific results. I can say, though, that a negative "confounding percentage" from -khb- (-ssc describe khb-) can result from a suppressor effect, that is, a causal nexus in which the direct and indirect effects of a variable are opposite in sign.