[R] strange behaviour of negative binomial

[Thomas Lumley]

On Wed, 5 Jan 2005, dax42 wrote:

> Dear list,
>
> I ran into a strange behaviour of the pnbinom function - or maybe I just made 
> a stupid mistake.
> First thing is that pnbinom seems to be very slow. The other - more 
> interesting one - is that I get two different curves when I plot the 
> estimated density and the density given by pnbinom. Shouldn't it be the same?

Yes, up to sampling error. That's actually one of the sets of tests that R 
does for the random number generators (though we can't test all possible 
parameters).

> This is only the case, I think, if I use the parameter size = 1. I just tried 
> it for size = 2 and the result seemed correct..
>
> Anyways, this is what I tried:
>
> #parameters
> proba<-0.001039302;
> s<-1;
>
> #random numbers with this distribution
> dom<-rnbinom(10000,s,proba);
>
> #first plot
> plot(ecdf(dom),do.points=FALSE);
>
> #second plot
> x<-seq(0,8700,0.5);
> lines(x,pnbinom(x,s,proba),col="red");
>

I get almost identical curves, on both R 2.0.0 under Linux and R-devel on 
Mac (and it only takes half a second on the local computer and 1.25s 
across a couple of miles of network, so it isn't that slow).

 	-thomas