有标准的方法识别异常值和处理方法,详细见下面两本书
1、 Introduction to Linear Regression Analysis (Douglas C. Montgomery著)的第6章。
2、Neter 和Kutner等人写的教材Applied Linear Regression Models, 4th ed. (ISBN 9780073014661)
winsor是采用随机的方法处理,每次的结果会不同,但是每次的结果应该不会有本质区别的。Winsor很多时候可以作为稳健性检验,比如说1%水平的winsor处理是比较常见的。
这个问题很多人讨论了,我不是这方面的专家,其他人有这样的看法:“CPI自身存在的问题(1) Substitution bias. The CPI ignores the fact that consumers substitute toward goods that have become relatively less expensive. (2) Introduction of new goods. Because the CPI uses a fixed basket of goods, it does not take into account the increased well-being of consumers created when new goods are introduced. (3)Unmeasured quality change. Not all quality changes can be measured.“
CPI只是一篮子的消费商品构成,公布的CPI 和亲身感受的差别大,我觉得一是和各个地区的消费水平有关,二是可能会受房价上涨影响,但房价不计入CPI。