In general, the matrix will have a basis of eigenvectors, and one will have an eigenvalue with modulus strictly bigger than the others (this occurs with probability 1) (*). Call this eigenvector v.
If you then choose a random initial vector and represent it using the eigenvector basis, then with probability 1, the coefficient of v will be non-zero (**).
Then repeatedly multiplying by the matrix causes that coefficient to grow to dominate all others, and so the vector tends towards the eigenline corresponding to the biggest eigenvector.
If (*) or (**) turn out not to be true (because you are unlucky, or because you have an especially fiendish examiner (in which case I guess you're still unlucky)), then all bets are off. (For any specific case, it's unlikely to be that hard to work out what happens, but a general solution/description is going to be tricky).
I