Proxying Ability by Family Background in Returns to Schooling Estimations is Generally a Bad Idea*
当教育年限测量有误差且缺乏能力数据时，用家庭背景变量代理能力会增大测量误差偏差，使教育回报率估计值偏离真实值更远。
Abstract When schooling is measured with error and data on ability are lacking, return to schooling estimates will be subject to positive omitted variable bias (OVB) and negative measurement error bias (MEB). We investigate how these biases are affected when ability is proxied by family background variables. We show that the effect on OVB is uncertain, while MEB invariably increases in magnitude. Our empirical analysis demonstrates that MEB generally dominates OVB. With more background variables or increased measurement error, the total bias rapidly becomes negative and increasing in magnitude, thereby driving the return estimate further and further away from the true value.