Economic models of home production predict a tradeoff between a mother's fertility and
her labor supply. Recent empirical work suggests that while these outcomes are correlated,
the causal impact is negligible when estimated through Instrumental Variables (sex preference,
twinning). I nd that in Taiwan, intense son preference produces 2SLS estimates of a mother's
labor response that are larger than OLS estimates, suggesting that previous IV analyses may
focus on mothers with lower than average costs to childbearing. I present a model of a mother's
joint determination of fertility and labor supply allowing for unobserved heterogeneity in both
the benets and costs of children. The model predicts that IV estimates will rise in proportion
to the intensity of the instrument, and provides a framework for assessing the bias in OLS and
IV estimators. Estimation of the structural model predicts that the average causal effect of
a third child on a mother's probability of working is -10 percent in Taiwan and -12 percent
in the United States for mothers between 34 and 36 years of age. The results imply that
the IV estimate for the US (-7 percent) is lower than the average causal effect, but the IV
estimate in Taiwan (-11 percent) is slightly larger than the average causal effect. I present a
set of simulations to consider how IV estimates derived from simulated data change when the
strength of the instrument varies, and nd that in both Taiwan and the US, weaker instruments
provide estimates lower than the average causal effect, and stronger instruments yield estimates
closer to the average causal effect.