Recent studies of the impact of child mortality on children ever born have used the "replacement factor" to measure mortality. When microlevel data are used, however, use of the replacement factor (or other variables which are nonlinear in the family's experienced child mortality rate) yields biased coefficients. An alternative model suggests a sequential, rather than a static, decision-making process to relate the decision to have an additional child to the reproductive experience to date. In this case, unbiased coefficients are obtained if the functional form is correctly specified. In the abscence of a priori knowledge of the functional form, it is difficult to untangle true effects from spurious ones.