Length-bias Correction in Transformation Models with Supplementary Data
提出一种变换模型的推断方法，处理随机截断点不可观测时的长度偏差问题，仅需截断变量分布函数可从补充数据估计，适用于失业持续时间等应用。
In this article, I propose an inferential procedure of monotone transformation models with random truncation points, which may not be observable. This class includes length-biased samples that are common in duration analysis. The proposed estimator can be applied to more general situations than existing estimators, since it imposes restrictions on neither the transformation function nor the error terms. Furthermore, it does not require observed truncation points either. It is sufficient for point identification to know the cdf of the truncation variable, which can be estimated from supplementary data that are easily found in applications. The estimator converges to a normal distribution at the rate of [image omitted] and Monte Carlo simulations confirm its robustness to error distributions in finite samples. For an empirical illustration, I estimate the effect of unemployment insurance benefits on unemployment duration, using length-biased microdata and supplementary macrodata.