Question

I’m working on a problem of finding mean processing times. I’m trying to eliminate outlier data by essentially performing a average on only the best 80% of the data.

I am struggling trying to adapt existing Top N per Group solutions to perform averaging per group. Using SQL Server 2008.

Here is a sample of what the table looks like:

    OpID  |  ProcessMin  |  Datestamp
    2     |  234         |  2012-01-26 09:07:29.000
    2     |  222         |  2012-01-26 10:04:22.000
    3     |  127         |  2012-01-26 11:09:51.000
    3     |  134         |  2012-01-26 05:02:11.000
    3     |  566         |  2012-01-26 05:27:31.000
    4     |  234         |  2012-01-26 04:08:41.000

I want it to take the lowest 80% of the ProcessMin for each OpID, and take the average of that array. Any help would be appreciated!

* UPDATE *

Given the following table:

OpID  ProcessMin   Datestamp
602   33           46:54.0
602   36           38:59.0
602   37           18:45.0
602   39           22:01.0
602   41           36:43.0
602   42           33:00.0
602   49           03:48.0
602   51           22:08.0
602   69           39:15.0
602   105          59:56.0
603   13           34:07.0
603   18           07:17.0
603   31           57:07.0
603   39           01:52.0
603   39           01:02.0
603   40           40:10.0
603   46           22:56.0
603   47           11:03.0
603   48           40:13.0
603   56           25:01.0

I would expect this output:

OptID   ProcessMin
602     41
603     34.125

Notice that since there are 10 data points for each OpID, it would only average the lowest 8 values (80%).

Answer 1

You can use ntile

select OpID,
       avg(ProcessMin) as ProcessMin
from
  (
    select OpID,
           ProcessMin,
           ntile(5) over(partition by OpID order by ProcessMin) as nt
    from YourTable
  ) as T
where nt <= 4
group by OpID

SE-Data

If ProcessMin is an integer you can do avg(cast(ProcessMin as float)) as ProcessMin to get the decimal average value.