I have empirical data on a process that I assume is Poisson with a given mean, say $\mu$ (unknown). The data is of the form $(x_i, 1\leq i\leq n)$ for $n$ consecutive time periods.

I am concerned that the process remains stable, by which I mean that $\mu$ does not change over time (as these are "negative" events, I am mainly concerned about the mean switching to a higher value for some unforeseen external reason). Therefore I want to be able to detect if a new occurrence ($x_{n+1}$) exceeds what is reasonably expected (for instance if the empirical history is mostly $0$'s and $1$'s, $x_{n+1}=100$ would be unexpected -- although theoretically possible).

I would like to be able to get:

  1. A predictive $1-\alpha$ confidence interval on $x_{n+1}$, either one sided or two-sided, and

  2. A test of the form "if $x_{n+1}\geq$ some value, reject the hypothesis that $x_{n+1}$ is generated by a Poisson process with the same mean as my historical data.

I have seen a number of papers on the estimation of $\mu$ but this is a bit different.


  • 1
    $\begingroup$ If you believe your Poisson counts to be independent (though in many situations I might find that surprising), you could try searching for Poisson tolerance intervals $\endgroup$
    – Glen_b
    Aug 3, 2019 at 6:34

1 Answer 1


After some research and the help of a comment by @Glen_b I will answer my question. The correct keyword is prediction interval; Wikipedia has this article about the subject.

A recent reference is Improved Closed-Form Prediction Intervals for Binomial and Poisson Distributions (Krishnamoorthy & Peng, 2011) [PDF link].


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.