How to select the top and bottom 100 observations for 5 variables each in a dataset?

[Hank]

Hi,

I have searched around online for answers on the fastest way, in a database of around a million rows, to select the top and bottom observarions each for a selected few variables. It could either be the top/bottom 100 obs, or selected based on some criteria (for example; select all observations above/below a certain value, or above/below 1 standard deviation from the mean etc.).

To clarify, I want to create a new dataset for each of the variables I want the tob/bottom values from . Does anyone have some code-examples to share?

Best regards,

Hank

[LinusH]

PROC RANK?

[gnanau]

PROC SURVEYSELECT?

[Hank]

I have tried it and maybe my coding is poor, but it only gives you the ranking of the variables. To proceed, I would then have to select the top and bottom of the ranks which in my dataset adds no value. Not familiar with proc surveyselect, but will check it out.

[data_null__]

I don't see how SURVEYSELECT will help.

How do you define top and bottom.

• order by var; 100 from the top 100 from end (could be less than 100 values if there are ties)
• rank by var; and take top 100 ranks and bottom 100 ranks.  (could be more obs if there are ties)