Generating Arrival Counts Given Time

Question

I'm looking to generate arrival counts given a length of time for a Poisson process. This is similar to another question, however rather than generate the particular arrivals, I'd like to just get the final count.

As an example, given a Poisson distribution where we expect 1 arrival per second, I'd like to generate instances where over a period (let's say 10 seconds) we then yield arrival counts (such as 9, 13, 10, ...).

Currently, I'm simulating the inter-arrival times between arrivals, and when I reach my desired length of time, add up how many arrivals happened. However this seems inefficient if there is a way to just arrive at the final count.

Is this possible, and if so, how would I go about doing it?

Answer 1

In R, the function rpois will take a random sample of specified size from a Poisson distribution of specified mean. If you expect $1$ arrival per second, then the mean arrival rate for 10 seconds is $\lambda = 10.$

You don't say for how many independent (i.e, disjoint) 10-second periods you want to generate arrivals. For $n = 15$ the simulation would be as follows:

set.seed(2019) # for reproducibility
lambda = 10;  n = 15
rpois(n, lambda)
 [1] 12  8  4 17  7 11 11  9 10 14
[11] 11 12  4  9 12

Numbers in brackets give the index of the first count on each line of output.

With a larger number of time periods, you could make a histogram of results and illustrate goodness of fit of the simulated sample to the population distribution $\mathsf{Pois}(\lambda = 10).$ (With an even larger number of simulated counts, the fit of the histogram to the exact Poisson distribution would tend to be better.)

Centers of open red circles are exact probabilities of the distribution $\mathsf{Pois}(\lambda = 10).$ R-code for making the figure is shown below.

set.seed(402);  lam = 10;  n = 5000
x = rpois(n, lam)
cutp = (-1:max(x))+.5
lbl = "Histogram of 1000 Counts from POIS(10)"
hist(x, prob=T, br=cutp, col="skyblue2", main=lbl)
  y = 0:30;  PDF = dpois(y, lam)
  points(y, PDF, col="red")

Note: Setting a seed for the pseudorandom number generator allows one to replicate a simulation experiment exactly. Omit the set.seed statement for a fresh simulation based on an unpredictable seed.