Matt Miller (#4) (12/17/2014)

---

I am feeling silly. There's a much simpler implementation, but I basically never use this function.

The trick is to leverage the NTILE aggregate function rather than ROW_NUMBER. In which case the query would look a lot more like djj's.

with cte as

(

select *,

NTILE(4) over(PARTITION by MainGroup order by value DESC) as 'GroupN'

--the NTILE parameter divvies up the partitioned groups into 4 parts

--in this case NTILE #1 is the top 25% and NTILE 4 = bottom 25%

from #mytable

)

select * from cte

where GroupN=4