y****2
发帖数: 46 | 1 Dear all
I have a question about outlier identification.
I have 2 groups of biological samples, 9 items per groups.
There is a couple values is far away from mean and also change the data
distribution dramatically.
I tried different methods to identify outliers, grub's method, median+/- 1.5
of interquarntile, median+/-5.2 MAD, Rubb method.
However, none of them can remove all the "obvious" outliers.
Only if I use the outlier package in R, which based on which data point has
most distance from mean.
But do you think this statement is reasonable enough?
Thanks | c********h
发帖数: 330 | 2 I remember there is an outlier test in linear regression. Basically, you
want to look at whether deleting one instance changes the coefficients
dramatically or not. Cook's distance may also be relevant here.
But your sample size is toooooo small, 18? I don't think deleting any sample
is a good strategy... |
|
