I am attempting to model the finishing position (independent variable) of runners in a race based upon height, weight, age, gender, past results (dependent variables). Thusfar I have performed an ordered logit regression from which I can generate predicted probabilities of a runner finishing in a given position. However, the probabilities of all runnings coming first do not sum to one, nor do sum of the probabilities of a runner finishing in any position. So I'm not sure how useful these predicted values are. My preferred software package is Stata, but also have R and Matlab. My best attempt thus far is to run

ologit y x1 x2... xn

and then scale down the predicted values so they sum to one, but this seems a bit rough and I suspect the resulting numbers are biased in some way. My question is: How do a perform a conditional ordered logit / probit regression?

  • $\begingroup$ What do you mean that the probabilities do not sum to one? Can you paste in some of the problematic output? $\endgroup$ Jun 2 '15 at 3:12
  • $\begingroup$ To clarify: Ordered logit regression of runner characteristics on finishing position will generate the predicted probability of a each runner finishing 1st. But the sum of the these probabilities will not necessarily equal 1, unless I force them to be. I know how to force this for regular logit (use clogit command in stata), but not for ordered logit. $\endgroup$ Jun 2 '15 at 5:17
  • $\begingroup$ In OLR, the probabilities for all outcomes should sum to 1 for each individual, not the probabilities for all individuals for the same place in the finishing order. $\endgroup$ Jun 2 '15 at 14:25

Here's an example in R:

to fit ordered logit model and store results 'm'

m <- MASS::polr(y ~ x1 + x2 + xn, data = dat, Hess=TRUE)

view a summary of the model


Here's a link to a detailed walkthrough of ordered logistic regression in R.

  • $\begingroup$ Answer needs to be explained more. $\endgroup$
    – Learner
    Jun 2 '15 at 4:55
  • 1
    $\begingroup$ Thanks but that is regular ordered logit - which i can do easily in Stata. I am after conditional ordered logit... $\endgroup$ Jun 2 '15 at 7:10

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.