0
$\begingroup$

If I have a time series that is usually $0$, but occasionally peaks at $1$ (what I'll term an event). If I know the time between 'events', is there a way to define an event probability?

1 Answers 1

0

Yes, if you actually know the underlying distribution of time in between events. A very common distribution for discrete events is the Poisson distribution, where the time in between events (waiting time) is independent of any other waiting time and is distributed as a exponential random variables, i.e., the waiting times are distributed as $\sim e^{-t}.$

This results in the Poisson distribution which tells you the probability of $n$ events happening in some interval that has $\lambda$ events on average:

$$p(n)=\frac{\lambda^ne^{-\lambda}}{n!}$$

This was, however, just an example. If you do not know the underlying distribution of waiting times exactly, you need to make a model that fits your particular data set.

  • 0
    Thanks! If I get a histogram of the waiting time I should be able to see if it's $e^{-t}$ or not.2017-01-18
  • 0
    @James Yes, with a constant $re^{-rt}$, see https://en.wikipedia.org/wiki/Exponential_distribution.2017-01-18