Outliers treatment methods

[dkanand86]

Hi Everyone,

I have two tasks for each variable in my dataset:

1. Find outliers for each variable

2. Treat the outliers for each variable

Let me know your comments.

Regards,

[ballardw]

Define outlier. There is no actual standard definition, so you'll have to have some rule for that.

proc univariate may get you started with some ideas.

And what do you mean by "treat" the outliers? There are many approaches some can be pretty complex, others are as simple as set the value to missing or remove the observation from the dataset.

[Ksharp]

Yeah.There are many way you can define outliers:

1) [mean-2*std , mean+2*std]  out of this range is outliers.
2) [Q1- 1.5*(Q3-Q1)  ,  Q3 + 1.5*(Q3-Q1)]  out of this range is outliers.
3) Check   proc robustreg  

[Ksharp]

Here can give you outliers:


proc robustreg data=sashelp.cars method=MM ;
model invoice = /diagnostics;
run;

[Rick_SAS]

For a discussion of detecting univariate outliers, you might want to read Detecting outliers in SAS   It explains some of KSharp's suggestions.

[Ksharp]

Oh. Don't format the IML function like : LTS() , LTM()..... Rick's blog has already explained it . And Better check the documentation of IML, There is a special chapter to talk about it .