1

I would like to test an intervention using interrupted time series (segmented regression). I have monthly rate (proportion) data with many of the rates in the <0.10 range. Should I use a standard linear regression approach even though the data are rates? This would allow me to use more standard approaches to account for autocorrelation. Or should I use Poisson regression with an offset for rates? The Poisson regression seems more appropriate for rates, but then how do I test for and adjust the model for autocorrelation?

Mike Hunter
  • 9,682
  • 2
  • 20
  • 43
Yogi
  • 11
  • 1
  • Poisson regression is appropriate for count or integer data and, therefore, is not appropriate for proportions. It may be that the literature on compositional data and models would be informative. There a couple of excellent references: Aitchison's *The Statistical Analysis of Compositional Data* is one and Pawlovsky-Glahn's *Compositional Data Analysis: Theory and Applications* is another. – Mike Hunter Dec 12 '17 at 15:24
  • 3
    It is my understanding that you can model rates with Poisson regression using an offset, as described here: https://stats.stackexchange.com/questions/11182/when-to-use-an-offset-in-a-poisson-regression – Yogi Dec 12 '17 at 16:27

0 Answers0