s******o
发帖数: 283 | 1 用box plot的时候，为啥用h = 1.5IQR 来 detect outliers, 系数为啥是1.5?
One other relevant question is when we use 3sigma for outliers? why the
coefficient is 3 here?
Thanks so much in advance. |
p***r
发帖数: 920 | 2 h=1.5 IQR, because this is the range for 25% (24.65%) on each side of the
center 50% under normal distribution.
chance to be outliers (.35%+.35%) on either side under normality assumption.
【在 s******o 的大作中提到】
: 用box plot的时候，为啥用h = 1.5IQR 来 detect outliers, 系数为啥是1.5?
: One other relevant question is when we use 3sigma for outliers? why the
: coefficient is 3 here?
: Thanks so much in advance.
|
p*******r
发帖数: 1951 | 3 这个看看正态分布曲线就明白了。至于是3 sigma，还是 2.95 sigma 完全是个约定俗
成的东西了。
【在 s******o 的大作中提到】
: 用box plot的时候，为啥用h = 1.5IQR 来 detect outliers, 系数为啥是1.5?
: One other relevant question is when we use 3sigma for outliers? why the
: coefficient is 3 here?
: Thanks so much in advance.
|
s******o
发帖数: 283 | 4 thanks a lot for all of your help.
So both the Box-plot 1.5IQR and 3sigma methods all have fixed portion of
data to be outliers, the difference between them is the value of the
percentage.
is that correct?
Also when the sample size increase, the number of outliers also increase ? |
a****t
发帖数: 1007 | 5 还是不大懂，我记得书本上是说1.5IQR以外算是outlier，那大家在工作中，是用1.
5IQR还是3sigma?
【在 s******o 的大作中提到】
: thanks a lot for all of your help.
: So both the Box-plot 1.5IQR and 3sigma methods all have fixed portion of
: data to be outliers, the difference between them is the value of the
: percentage.
: is that correct?
: Also when the sample size increase, the number of outliers also increase ?
|
s******o
发帖数: 283 | 6 I want to know too.
【在 a****t 的大作中提到】
: 还是不大懂，我记得书本上是说1.5IQR以外算是outlier，那大家在工作中，是用1.
: 5IQR还是3sigma?
|
