从二项式（K，p）进行采样，结果意外

Question

In the help files for rbinom , size argument is a number of trials (incl. a zero) but it doesn't say if this can also be a vector.

The correct way of using this function is

table(rbinom(n = 1000, size = 1, prob = 0.2))

  0   1 
809 191

But what is happening here?

table(rbinom(n = 1000, size = 0:1, prob = 0.2))

  0   1 
894 106 

Answer 1

Argument recycling of the size argument is the prime cause.

Because n is 1000, 0:1 is recycled until you get 500 0 's and 500 1 's (alternating).

All the 0-size ones give 0 :

   > rbinom(10,size=0,prob=0.2)  
     [1] 0 0 0 0 0 0 0 0 0 0

Resulting in 500 0 's + 500 Bernoulli trials with p=0.2, resulting in about 100 1 's out of 1000 elements.

[Your results didn't seem surprising to me, but argument recycling can bite if you're not looking for it, and - while there are reasons why the number of successes in 0 Bernoulli trials should be defined as 0 - it may not seem obvious at first either.]

Answer 2

Documentation bug:

If 'size' is not an integer, 'NaN' is returned. [my emphasis]

You are giving it more than one integer, so the documentation would imply that you would get NaN .

Its confusing because it explicitly states where other arguments can be vectors but not size . I'd file a documentation bug with the maintainer, which in this case probably means the main R bug tracker.