Question

Code included is a simplified version of our situation; the production table equivalent to #MyExample has 20 fields all of which need medians calculating therefore the second part of the script becomes very long – not a huge hard-ship but is there a more compact solution?

I’ve no experience with APPLY or custom FUNCTIONs but is this a situation where we should create a FUNCTION for the median and then use APPLY I’m guessing not as apply is applied to each row?

/*
DROP TABLE #MyExample
DROP TABLE #mediantable
*/

CREATE TABLE #MyExample
        (
        customer char(5),
        amountPeriodA numeric(36,8),
        amountPeriodB numeric(36,8),
        amountPeriodC numeric(36,8)
        )
INSERT INTO #MyExample
        values
        ('a',10,20,30),
        ('b',5,10,15),
        ('c',500,100,150),
        ('d',5,1,1),
        ('e',5,1,15),
        ('f',5,10,150),
        ('g',5,100,1500)




SELECT 
        [Period] = 'amountPeriodA',             
        [Median] = AVG(x.amountPeriodA)         
INTO    #mediantable
FROM (
        SELECT 
                r.customer,
                r.amountPeriodA,
                [RowASC] = ROW_NUMBER() OVER(ORDER BY r.amountPeriodA ASC, customer ASC),
                [RowDESC] = ROW_NUMBER() OVER(ORDER BY r.amountPeriodA DESC, customer DESC)
        FROM #MyExample r 
    ) x
WHERE RowASC IN (RowDESC, ROWDESC-1, ROWDESC+1)

union
SELECT 
        [Period] = 'amountPeriodB',             
        [Median] = AVG(x.amountPeriodB)         
FROM (
        SELECT 
                r.customer,
                r.amountPeriodB,
                [RowASC] = ROW_NUMBER() OVER(ORDER BY r.amountPeriodB ASC, customer ASC),
                [RowDESC] = ROW_NUMBER() OVER(ORDER BY r.amountPeriodB DESC, customer DESC)
        FROM #MyExample r 
    ) x
WHERE RowASC IN (RowDESC, ROWDESC-1, ROWDESC+1)

union
SELECT 
        [Period] = 'amountPeriodC',             
        [Median] = AVG(x.amountPeriodC)         
FROM (
        SELECT 
                r.customer,
                r.amountPeriodC,
                [RowASC] = ROW_NUMBER() OVER(ORDER BY r.amountPeriodC ASC, customer ASC),
                [RowDESC] = ROW_NUMBER() OVER(ORDER BY r.amountPeriodC DESC, customer DESC)
        FROM #MyExample r 
    ) x
WHERE RowASC IN (RowDESC, ROWDESC-1, ROWDESC+1)


SELECT * 
FROM #mediantable

Answer 1

Building on my previous reply I arrived on this which is a lot easier (and shorter) to expand for the number of columns and even runs a bit faster (probably a lot faster in case of 20+ columns!). However, it returns the results horizontally instead of vertically. This can be ‘solved’ again using UNPIVOT.
I’ve done the operation in 2 parts using an intermediate #result table; but you could easily do it in a single statement using a subquery or CTE.

DECLARE @rowcount int
DECLARE @first int
DECLARE @last int
DECLARE @divider numeric(36,8)

SELECT @rowcount = COUNT(*) FROM #MyExample

SELECT @first = (CASE WHEN @rowcount % 2 = 1 THEN (@rowcount + 1) / 2 ELSE (@rowcount / 2)     END),
       @last  = (CASE WHEN @rowcount % 2 = 1 THEN (@rowcount + 1) / 2 ELSE (@rowcount / 2) + 1 END),
       @divider = (CASE WHEN @rowcount % 2 = 1 THEN 1 ELSE 2 END)

 SELECT amountPeriodA = SUM(amountPeriodA) / @divider,
        amountPeriodB = SUM(amountPeriodB) / @divider,
        amountPeriodC = SUM(amountPeriodC) / @divider    
   INTO #result
   FROM
 (
 SELECT amountPeriodA = ((CASE WHEN ROW_NUMBER() OVER(ORDER BY amountPeriodA ASC, customer ASC) IN (@first, @last) THEN amountPeriodA ELSE 0.00 END)),
        amountPeriodB = ((CASE WHEN ROW_NUMBER() OVER(ORDER BY amountPeriodB ASC, customer ASC) IN (@first, @last) THEN amountPeriodB ELSE 0.00 END)),
        amountPeriodC = ((CASE WHEN ROW_NUMBER() OVER(ORDER BY amountPeriodC ASC, customer ASC) IN (@first, @last) THEN amountPeriodC ELSE 0.00 END)) 
   FROM #MyExample
  )t 

and then

  SELECT [Period], [Amount] 
    FROM #result as x
    UNPIVOT ( [Amount] FOR Period IN (amountPeriodA, amountPeriodB, amountPeriodC)) As unpvt