Effects of previous birth interval length on child outcomes can be estimated in a sibling analysis even when there...

There is much interest in how the length of the previous birth interval affects various child outcomes, and it has become increasingly common to estimate such effects from sibling models. This is because one then controls for unobserved determinants of the outcome that are shared between the siblings and linked to the birth interval length. However, it is a common idea that such effects can only be estimated from data on families with three or more children.

The goal of this paper is to show, through an intuitive argument and a simulation experiment, that it is possible to estimate effects of birth interval only from families with two children.

Observations are simulated from two equations for fertility and one equation for child mortality. The fertility equations include a random term that is assumed to be correlated with the random term in the mortality equation. Mortality models are then estimated from the simulated observations. This is done 1000 times, and the averages of the 1000 sets of estimates are calculated.

The simulation experiment illustrates that it is indeed possible (by using a model specification that takes into account that no birth interval is defined for the first birth) to estimate birth interval effects in sibling models even when the data include only families with two children.

It is good news that families with only two children can contribute to the estimation of birth interval effects. This is because, with a broader basis for the estimation, the precision is improved and there is less reason for concern about the general relevance of the estimates. An important limitation, however, is that it is potentially problematic to control for maternal age in a sibling model estimated only for the first and second child.