Likelihood-Preserving Normalization in Multiple Equation Models