What Do Instrumental Variable Models Deliver with Discrete Dependent Variables?
比较了非参数工具变量模型与线性模型及两阶段最小二乘法在因变量离散时的表现，发现两阶段最小二乘法只能估计局部平均处理效应，而工具变量模型能识别更多结构参数。
We compare nonparametric instrumental variables (IV) models with linear models and 2SLS methods when dependent variables are discrete. A 2SLS method can deliver a consistent estimator of a Local Average Treatment Effect but is not informative about other treatment effect parameters. The IV models set identify a range of interesting structural and treatment effect parameters. We give set identification results for a counterfactual probability and an Average Treatment Effect in a IV binary threshold crossing model. We illustrate using data on female employment and family size (employed by Joshua Angrist and William Evans (1998)) and compare with their LATE estimates.