A Simple Correction for Fertility Selection

Mark Pitt, Brown University

Failure to account for the possibility that unmeasured determinants of health may affect the fertility decision, and thus the composition of those born classified by health, may result in biased estimates of the determinants of child health. Parameter identification is the first practical problem in controlling for fertility selection. If parents care about the health outcomes of potential births, then any exogenous variable that affects health also affects the fertility decision. The availability of longitudinal data on births permits modeling fertility selection as arising from the correlation between a random effect influencing the probability of a birth and another influencing the health of those born. The second problem is computational complexity. This paper sets out some methods for controlling and testing for fertility selection that can be simply implemented using only a few lines of code in standard software packages, and provides simulation results demonstrating the efficacy of these methods.