I have a variable R that represents the risk of disease in each subject. I
regress this variable against predictors of risk:
regress R sex age obese diabetic
In this model obese and diabetic are dichotomous variables. I want to
calculate the proportion of R that is attributable to obesity (P), after
adjustment by sex, age and diabetes. Therefore, I estimated the predicted R
in the whole population and in those without obesity using adjust:
adjust, gen(adj1 seadj1) se
adjust if obese==0, gen(adj2 seadj2) se
Then, I calculate P: gen P=(adj1-adj2)/adj1
However, I need to calculate the variance of P. Since P is a ratio, there
seems to be no analytical way to estimate it. Therefore, I made 10000
simulations of P and got the variance from the simulated values, as follows:
gen Ipop = adj1 + seadj1 * invnorm(uniform())
gen Inull = adj2 + seadj2'* invnorm(uniform())
gen P = (Ipop-Inull)/Ipop
sum P
gen Pmean=r(mean) (Mean of P)
gen Pvar=r(Var) (Var of P)
Is this a reasonable approach?
Thanks!
Leonelo E. Bautista
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/