How many people choosing random answer would it take to muddy the data to the point it is useless?