主  题：Item Non-response and imputations

主讲人：Christopher R. Bollinger教授

主持人：李伊 副教授

时  间：2014年6月11日（星期三）上午10:00

地  点：通博楼B212学术会议室

主办单位：统计学院   科研处

 

内容提要：

While enumerators seek to elicit responses from all survey participants,item non-response for earnings and income is increasingly a problem for large  surveys.  

One solution to this has been the use of “imputations.” In this research we examine the use of imputations in regression contexts, particularly when it is used as the dependent variable. We show that imputations may induce bias into regression estimates. We further examine whether the assumption of missing at random, as needed for both imputations and other suggested solutions to the non-response problem, is supported.Using both statistical sample selection   corrections and matched administrative data, we find that non-response is not   random, but may be symmetric.