4

I'm using a negative binomial GLM to predict individual level scores on a psychometric test that ranges from 0-27 and presenting the coefficient "incidence rate ratios".

It seems a little strange to use the term "incidence rate ratio" when actually it is a psychometric score ratio. Should I still use the term "incidence rate ratio" in my manuscript for easy recognition/interpretation? If not, is there another, more generic term that is used or should I make up my own domain specific term (e.g. PGSI ratio where PGSI is the psychometric index I am modelling).

fmark
  • 4,666
  • 5
  • 35
  • 51
  • I agree with @EpiGrad's answer. I am wondering where this term cropped up in your situation - did someone publish a paper on the test using this term? – Michelle Mar 13 '12 at 07:16
  • @Michelle Yep, another paper in the field whose methods we are replicating. I have seen applied glm guides that present this estimate as an IRR regardless of field, which is where the confusion probably stems from. – fmark Mar 14 '12 at 04:41
  • 1
    looks like the use comes from `Stata`, e.g. here http://www.ats.ucla.edu/stat/stata/faq/relative_risk.htm and here http://data.princeton.edu/wws509/stata/c4s1.html. But I'm hard pressed to see how the number of "children ever born" can be viewed as an incidence (prevalence - maybe). – Michelle Mar 14 '12 at 05:18

1 Answers1

3

I don't think its appropriate to use the term "incidence rate ratio" and your instincts are correct. You're reporting, as far as I can tell, neither an incident event, nor an actual rate, and rather than retain it because "that's what usually comes out of a negative binomial GLM" I think coming up with a new term will add clarity. I know if I hit "IRR" or the like and you weren't reporting incident cases, my first thought would be "The hell...?"

I vote for a domain-specific term, and a one line discussion of how its obtained, ala "obtained by the exponentiation of the beta1 in a negative binomial GLM..."

Fomite
  • 21,264
  • 10
  • 78
  • 137