If one case is eliminated, you can't assume that the others still all have the same propability.

However, I did a statistical test with this C++ program:
Click to reveal..

Code:
#include <cstdlib>
#include <iostream>

using namespace std;

#define BOY 0
#define GIRL 1

int main(int argc, char *argv[])
{
    int child1, child2;
    int i;
    int hits;
    int tries;
    
    srand ( time(NULL) );
    
    
    cout << "\nMan:";
    hits = 0;
    tries = 0;
    for(i=0;i<1000;i++)
    {
        child1 = BOY;
        child2 = rand() % 2;
         
        if((child1 == BOY) || (child2 == BOY)) tries++;
        if((child1 == BOY) && (child2 == BOY)) hits++;
    }
    cout << "\n" << hits << "/" << tries << " = " << static_cast<double>(hits)/tries;
    
    
    cout << "\nWoman:";
    hits = 0;
    tries = 0;
    for(i=0;i<1000;i++)
    {
        child1 = rand() % 2;
        child2 = rand() % 2;
         
        if((child1 == BOY) || (child2 == BOY)) tries++;
        if((child1 == BOY) && (child2 == BOY)) hits++;
    }
    cout << "\n" << hits << "/" << tries << " = " << static_cast<double>(hits)/tries;
    
    
    cout << "\n";
    
    system("PAUSE > NUL");
    return EXIT_SUCCESS;
}



It shows that you are actually right and I was wrong, but I still don't see why. This seems pretty strange to me, because that "three possibilities" argument is not quite valid in my eyes. Still, the propability seems to be 1:3.

You can try the program I posted while I reread your and Quadraxas' posts and rack my brain about this...

Kubetris
LBGUI