I am trying to to run a logistic regression (case-control) and the variable of interest is categorical, taking the values 0 to 6. For a subset of individuals, I do not have the exact value (0, 1 .. or 6) but a probability distribution : P(V=0)=0.03 P(V=1)=0.2 .. P(V=6) = 0.3

I need an unbiased estimate of the OR and confidence interval both when I test this variable as an ordinal one or when I am testing all the levels as dummy variables (ORs of levels 1, 2, 3 ,4 ,5 and 6 vs level 0).

This looks like I need to introduce the probabilities in the likelihood maximization and I was wondering whether this is already implemented in a R script or any other program.

  • $\begingroup$ This seems well-suited to multiple imputation. There are plenty of textbooks on it, I find Andrew Gelman's [chapter][1] on it to be the most readable. I'd reply as an answer, but I'm not immediately sure how one would use an MI framework to combine multiple variance-covariance matrices in a logistic regression setting. www.stat.columbia.edu/~gelman/arm/missing.pdf $\endgroup$ – generic_user Apr 2 '13 at 2:22

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.