I want to understand the appropriate way to regress a duration variable on a count variable. For instance: I have data on turn around time of a process in a hospital ward. The ward admits patients with diverse clinical specialties. I have the daily average turnaround time of the process as the dependent variable. The independent variable is the count of medical specialties on a given day. The count variable has non-zero values ranging from 1 to 12.


What is good depends on the relationship in practice.

There is no reason I can imagine to rule out using the count as it comes if the relationship is linear.

As I understand it the intercept is unattainable, as zero specialties won't ever be observed, but that's not a novelty. If people's weights are predictable from their heights, we don't object that zero height cannot occur.

We would need to see the data to say more; I wouldn't be surprised if say log of time versus square root of count worked a bit better. Nor would I be surprised by enormous scatter and a weak relationship.


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.