[Felix Bittmann]

These values look OK. In your problematic model, you include even more control variables:

AGE BMI ANX_preop cs ETHNICITY ANX_preop DEP_preop cs_previous HO_OTHERCHRON_PAIN

I suspect the problem lies there. I would carefully assess each control and attempt to reduce the number of controls used. Your sample size is not that large. Start with a very simple model (in KHB) without many controls and see whether the results become more sensible.

[Asish Subedi]

Dear Felix Bittmann Sir,
I kept only the maximum pain at 6 weeks and postpartum depression as mediator variable whereas max pain at 72 hours as a confounder and got the results as shown below.
[CODE] khb logit cpsp intraop || pain6wk postdep, conc(AGE BMI ETHNICITY anxiety cs chronicpain predep pain72) summary dis or

Decomposition using the KHB-Method

Model-Type: logit Number of obs = 414
Variables of Interest: intraop Pseudo R2 = 0.66
Z-variable(s): pain6wk postdep
Concomitant: AGE BMI ETHNICITY anxiety cs chronicpain predep pain72
------------------------------------------------------------------------------
cpsp | or Std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
intraop |
Reduced | 32.09183 25.80521 4.31 0.000 6.636497 155.1851
Full | 15.82269 11.73416 3.72 0.000 3.698509 67.6915
Diff | 2.028215 .6307756 2.27 0.023 1.102529 3.73111
------------------------------------------------------------------------------

Summary of confounding

Variable | Conf_ratio Conf_Pct Resc_Fact
-------------+-------------------------------------
intraop | 1.256082 20.39 1.2553123
---------------------------------------------------

Components of Difference

Z-Variable | Coef Std_Err P_Diff P_Reduced
-------------+---------------------------------------------
intraop |
pain6wk | .5223103 .2628304 73.86 15.06
postdep | .184846 .1744171 26.14 5.33
-----------------------------------------------------------
/CODE] thank you in advance

[Felix Bittmann]

I think the results are acceptable yet you should think about the theoretical relations of your variables in detail. There are multiple empirical challenges as well, such as the low number of cases, the scaling of your variables and their high intercorrelation. Maybe this is the best you can do. In any case, make sure to report your findings transparently and discuss the implications and limitations of your research.

[Asish Subedi]

Dear Felix Bittmann Sir,

In this model, I kept the severity of intraoperative pain as exposure variable instead of presence/absence of intraoperative pain (binary)

[CODE] khb logit cpsp severity_intraop || pain72 pain6wk postdep, conc(AGE cs chronicpain predep ) summary dis or

Decomposition using the KHB-Method

Model-Type: logit Number of obs = 414
Variables of Interest: severity_intraop Pseudo R2 = 0.65
Z-variable(s): pain72 pain6wk postdep
Concomitant: AGE cs chronicpain predep
----------------------------------------------------------------------------------
cpsp | or Std. err. z P>|z| [95% conf. interval]
-----------------+----------------------------------------------------------------
severity_intraop |
Reduced | 5.82613 1.558264 6.59 0.000 3.449187 9.841099
Full | 1.856794 .31095 3.70 0.000 1.337262 2.578168
Diff | 3.137735 .6722714 5.34 0.000 2.061784 4.775176
----------------------------------------------------------------------------------

Summary of confounding

Variable | Conf_ratio Conf_Pct Resc_Fact
-------------+-------------------------------------
severity_i~p | 2.8477797 64.88 1.6560235
---------------------------------------------------

Components of Difference

Z-Variable | Coef Std_Err P_Diff P_Reduced
-------------+---------------------------------------------
severity_i~p |
pain72 | .4213935 .1756077 36.85 23.91
pain6wk | .683433 .2127939 59.77 38.78
postdep | .0386749 .0359568 3.38 2.19
-----------------------------------------------------------
/CODE]

[Felix Bittmann]

If you can defend this model theoretically, the empirical results seem fine to me.

[Asish Subedi]

Dear Felix Bittmann Sir,
Thank you. Earlier you suggested some code for finding the significance of each mediating variable statistically . I tried to run those, but it did not display the data. Please do suggest. Thank you once again.
Asish

[Felix Bittmann]

cap program drop khbboot
program define khbboot, rclass
    khb logit cpsp severity_intraop || pain72 pain6wk postdep, conc(AGE cs chronicpain predep) summary dis or
    return scalar med1 = e(disentangle)[1,4]
    return scalar med2 = e(disentangle)[2,4]
    return scalar med2 = e(disentangle)[3,4]
end

*** Test ***
khbboot
return list

*** Bootstrap ***
bootstrap r(med1) r(med2) r(med3), reps(1000) seed(123) dots(10): khbboot
estat bootstrap, bc

[Asish Subedi]

Dear Felix Bittmann Sir,
For the first code I got the same result as before. In addition the following
[CODE]
scalars:
r(med2) = 2.194502881792457
r(med1) = 23.91084480460459
/CODE]
How do I interpret them?

For the bootstrap code, it was inconclusive
[CODE] bootstrap r(med1) r(med2) r(med3), reps(1000) seed(123) dots(10): khbboot
(running khbboot on estimation sample)
'r(med3)' evaluated to missing in full sample
r(322);

. estat bootstrap, bc
last estimates not found
r(301);
/CODE]
thank you once again.

[Felix Bittmann]

My bad, there is a typo in my code. The correct code is:

cap program drop khbboot
program define khbboot, rclass
    khb logit cpsp severity_intraop || pain72 pain6wk postdep, conc(AGE cs chronicpain predep) summary dis or
    return scalar med1 = e(disentangle)[1,4]
    return scalar med2 = e(disentangle)[2,4]
    return scalar med3 = e(disentangle)[3,4]
end

*** Test ***
khbboot
return list

*** Bootstrap ***
bootstrap r(med1) r(med2) r(med3), reps(1000) seed(123) dots(10): khbboot
estat bootstrap, bc

[Asish Subedi]

Dear Felix Bittmann Sir,
Thank you.