Why nonconservative schemes converge to wrong solutions: error analysis