Consider a female population disaggregated by single years of age. Suppose we have available a forecast that is thought to be unbiased for the logarithm of the population vector. The problem we consider is, what is the contribution of the different vital processes to the forecast error of different ages? Alho and Spencer (1991) provided approximate propagation of error formulas needed in such an analysis. The derivation assumed that certain covariance terms would be so small that they could be neglected. A related issue is, therefore, how accurate are these approximate formulas in different circumstances?