0

I use poisson regression (glm() with family='poisson') and would like to give the answers in the form:

"Each predicted value has an error from the interval [low, high] with probability 0.95".

  1. Is it possible to do? If yes, how?
  2. I see few ways.

1) Compute residuals and get the low and high values. Which distribution should I use?

2) Use function predict.glm(model, type = 'link', se.fit=TRUE). Compute confidence intervals for normal distribution. Use 

model$family$linkinv()

to figure out low and high values.

I have implemented the second case. The question is how can I give the answers in the form described above?

dStack
  • 73
  • 1
  • 7
  • 5
    Make your example [reproducible](http://stackoverflow.com/q/5963269/1478381) and you will probably get an answer very quickly... – Simon O'Hanlon Mar 18 '13 at 17:20
  • This appears to be a problem more from a lack of statistical background than from the fact that you are using R. Suggest you request the moderators to migrate to CrossValidated. – IRTFM Mar 18 '13 at 18:23
  • your case #2 is essentially correct: it's not clear what else you need for a complete answer to your question. – Ben Bolker Mar 19 '13 at 01:29

0 Answers0