9
$\begingroup$

This is probably an elementary error in either my understanding or my R implementation: I am trying use a Poisson model to make some predictions. The original data is discrete count data. I would expect the predictions to also be discrete outcomes (e.g. 39, 40, 41). Instead - the predictions include decimals (41.2) - which seems odd for count predictions / the Poisson distribution. What am I doing wrong?

Example:

warpbreaks
breaksmodel<-glm(breaks~wool*tension, warpbreaks, family=poisson)
predict(breaksmodel,warpbreaks,type="response") 
$\endgroup$

1 Answer 1

9
$\begingroup$

Predictions from a Poisson model are the mean parameter of a Poisson distribution, which is not constrained to be an integer. (Think of the old "the average family has 2.4 children" conundrum.) Simulated outcomes from your model using those predicted means will of course be integers: for the predictor combination giving a predicted mean of 41.2 you might well get 45, 39, 42, or 40 as simulated outcomes.

$\endgroup$

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.