rand for statistical purposes

[exiledAussie]

I read that rand should not be used for serious statistical purposes. Does anyone have an explanation for this point of view? What are more suitable alternatives?

[Moschops]

It's not very random.

If you want better, look at Boost's random number library, or perhaps randomlib (nice interface for a Mersenne twister). There are many more; google will present lots.

[exiledAussie]

Thanks - I'll have a look.

Additional question: Do you know the algorithm that rand uses? Is it implementation specific?

[Moschops]

It is implementation specific. The standard does not specify anything beyond "The rand function returns a pseudo-random integer."

It gives this as a possible implementation, so you can see the potentially shoddy quality of random numbers you might get out of calling rand(). :)

static unsigned long int next = 1;
int rand(void)
// RAND_MAX assumed to be 32767
{
next = next * 1103515245 + 12345;
return (unsigned int)(next/65536) % 32768;
}

void srand(unsigned int seed)
{
next = seed;
}

[cire]

http://en.cppreference.com/w/cpp/numeric/random

is a good place to start, if you'd care to stick to standard C++.

other useful links:

http://www2.research.att.com/~bs/C++0xFAQ.html#std-random
http://stackoverflow.com/questions/7114043/random-number-generation-in-c11-how-to-generate-how-they-work