Removing Outlier

[MD ARIFUL ISLAM]

Dear all,

I want to remove the outlier above the yellow line. Is there any Stata command by which I can only remove the outlier from the Y variable on the basis of each X variable. For example, I want to remove the extreme value of Y when X is 1,2,3,......,27.

Thanks and kind regards,

Ariful

[Marcos Almeida]

You may use the ‘if’ clause to combine situations.

That being said, I just wish to underline a concept that already became a letany: simply excluding outliers is not the most appropriate strategy.

[William Lisowski]

You need to describe how the code should choose which observations are outliers. For X=1 there are two points above the yellow line; for X=2 there is a single point above the yellow line, and for X=3 there are no points above the yellow line.

[William Lisowski]

Expanding on Marcos's suggestion, the following example code demonstrates a technique that could start you on your way.

sort X Y
generate outlier = 0
by X (Y): replace outlier = 1 if X==1 & _n>=_N-1
by X (Y): replace outlier = 1 if X==2 & _n==_N
...

and then in your subsequent commands you can exclude the outliers

scatter Y X if outlier==0

This will enable you to do what you seek. It is poor statistical technique, as Marcos suggests,

[Nick Cox]

What are the variables any way? The plot to me looks consistent with a log-linear relationship for which logarithmic llink function would give good results.

[MD ARIFUL ISLAM]

Thanks a lot. I solve the problem. Thanks William, Nick and Marcos.