报告人/Speaker: Liqun Wang, University of Manitoba
报告题目/Title: Instrumental Variable Estimation in OrdinalProbit Models with Predictor Measurement Error
时间/Date& Time: May 22, 2018,14:30-15:30
Regression models with categorical responses arewidely used in many fields. In practice, many real applications involvepredictor variables which cannot be measured precisely and instead only proxyobservations are available to estimate model parameters. It is well-known thatstatistical methods ignoring such measurement error result in biased andinconsistent estimates and therefore misleading conclusions
We propose an instrumental variable approach toestimation of a probit model with ordinal response and mismeasured predictorvariables. We obtain likelihood based and method of moments estimators whichare consistent and asymptotically normally distributed under generalconditions. These estimators are easy to compute, perform well and are robustagainst the normality assumption for the measurement errors in our simulationstudies. The proposed method is applied to some real datasets.
LiqunWang is a professor in Department of Statistics, University of Manitoba,Canada. He obtained his Doctoral degree in statistics and econometrics fromVienna University of Technology, Austria. Liqun Wang's research areas includeapplied probability and statistical theory and methodology for complex dataanalysis. His current research focuses on boundary crossing probability (firstpassage time) for diffusion processes, identification and estimation innonlinear measurement error models and in longitudinal data models,high-dimensional variable selection and data assimilation, and simulationmethods in statistical computation and optimization.