Two problems motivate the use of instrumental variables. First, using 1997 agricultural data to proxy for agricultural conditions in earlier years introduces measurement error that may bias the estimate toward zero. Second, the OLS estimate will suffer from omitted variable bias if families that prefer girls switch to planting tea after the reform. In this case, the OLS estimate will overestimate the true effect of an increase in the value of female labor because it will confound the aforementioned effect with the sex-preferences of households that switched to planting tea. I address both problems by instrumenting for tea planting with the average slope of each county. Figure IVb shows the slope variation in China, where the darker areas are steeper. The predictive power of slope for tea planting can be seen by comparing the tea planting counties in Figure IVa with the hilly regions in Figure IVb. I use the GIS data pictured in Figure IVb to calculate the average slope for each county and estimate the following first-stage equation, where both the amount of tea planted and the slope is time-invariant. Column (4) of Table III shows the first-stage estimate from equation (4). The estimate for the correlation between hilliness and planting tea, λ, is statistically significant at the 5% level. Column (5) shows the 2SLS estimate from equation (5). The estimate is larger than the OLS estimate and statistically significant. Column (6) shows the 2SLS estimate after controlling for county-level cohort trends. The estimate is similar in magnitude to the OLS estimate, but no longer statistically significant. The estimates with and without trends are not statistically different from each other. The estimate without trends is larger but also less precisely estimated. The 2SLS estimate in column (6) shows that conditional on county-level cohort time trends, the OLS estimate is not biased. Furthermore, the OLS and 2SLS estimates in columns (3) and (6) are almost numerically identical to the initial OLS estimate in column (1). These results give confidence to the robustness of the initial OLS estimates.