I made a logistic regression model 1 (not fully adjusted) and model 2 (fully adjusted; model 1 + covariate A) with Stata. I am trying to calculate the attenuation percentages between OR.

I used the KHB method to make it correctly but I would want compare my results with a "classic" attenuation percentage (attributable fraction):

1) I don't know if additive scale (OR2-OR1)/(OR1-1)x100 or multiplicative scale [ln(OR2)-ln(OR1)]/[ln(OR1)]x100 is better...

I used the multiplicative scale and the result is 49.9%. To calculate an interval confidence, I made bootstrap with repeats=2000 in model 1 and model 2 (with a same seed) and so I could calculate 2000 attenuation percentages and the mean with 95% IC.

2) Is this method correct?

My IC is : 53.8-55.8 % and my attenuation percentage (49.9) is outside my IC.

3) Must I choose an other method of bootstrap?

  • $\begingroup$ I realized that my method was not correct because the IC of mean and the bootstrapped IC are different (standard errors are different). So I used KHB method which gives reduction percentage and I made a bootstrapped to calculate IC. However my IC is very large (4-92) and bootstrapped standard error is big (22) while my N=27 000 and repeats=2000. I don't understand... $\endgroup$ – user59248 Oct 24 '14 at 15:40
  • $\begingroup$ Please merge your accounts, Emmanuel, so that you may edit and comment on your own question. $\endgroup$ – whuber Oct 24 '14 at 15:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.