2
$\begingroup$

Just a quick one - I am planning on running a negative binomial distribution to test the time spent at the gym in a week, against a series of controls.

I just wanted to double check - would the number of hours spent at the gym, in the past week, be suitable as a 'count' variable? I feel that I am correct in assuming it is but just wanted some confirmation!

$\endgroup$
3
  • $\begingroup$ I'm wondering about the nature of this question. Did you start off wanting (for some reason) to do a negative binomial regression and are now looking for data? Also, time spent at the gym in a week would usually be in hours and minutes, not just hours. I also like Stephen's answer, below. (Count variables are usually things that can't be broken down, like number of cars owned, or number of operations had, or similar). $\endgroup$
    – Peter Flom
    Commented Apr 12 at 18:22
  • $\begingroup$ The data I have is basically a count of how many 'half-hour' periods people had spent at the gym over the course of the week. So I think it should be suitable for the negative binomial? $\endgroup$
    – Vito
    Commented Apr 13 at 22:05
  • $\begingroup$ OK, but I'd use count of half hours for a little bit more precisioin. $\endgroup$
    – Peter Flom
    Commented Apr 14 at 9:44

1 Answer 1

3
$\begingroup$

A number of hours is probably a count, as long as you round to full hours. Once you track minutes, or fractional hours, it's no longer count.

That said, my first distributional assumption would not necessarily be a negbin, rather a zero-inflated (or hurdle) shifted Poisson distribution, since observations are generated by two distinct processes: either I don't go to the gym at all because of sickness or travel (this is what generates the zeros), or I go and spend some time there (which generates the shifted Poisson, assuming that nonzero time at the gym is always at least one, but if you could go and spend time that gets rounded to zero, you could use a non-shifted Poisson). Whether this makes a big difference compared to a straight-up negbin is not obvious, though.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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