SQL Server Rounding Issue Looking for Explanation - sql

Question

I've solved this issue but I'm just wondering why this works the way it does. I have a temporary table I am selecting from and am looking to display a a name, the number of records that match this name, and the percentage for that name of the total records. This is the way I originally had it:
SELECT name, number,
CASE WHEN number = 0 THEN 0 ELSE
convert(Numeric(10,2), number / CONVERT(decimal(5,2),SUM(number)) * 100)
END as "Percentage of Total"
FROM #names
group by name, number
The results I received were:
name number Percentage of Total
------------------------- ----------- ---------------------------------------
Test 1 0 0.00
Test 2 22 100.00
Test 3 28 100.00
When I change the query to this, the results are correct:
declare #total decimal(5,2)
select #total = SUM(number) FROM #names
SELECT name, number, convert(Numeric(10,2), number/ #total * 100) as "Percentage of Total"
FROM #names
group by name, number
Correct Results:
name number Percentage of Total
------------------------- ----------- ---------------------------------------
Test 1 22 44.00
Test 2 0 0.00
Test 3 28 56.00
Can someone explain what is going on, I would like to understand this better. Thanks!
Jon

Answer 1

You first query groups by number.
Since you don't have duplicates of numbers, number / SUM(number) is equivalent to the 1 / COUNT (except when the number is 0).
You second query does not group by number, it calculates total sum.
Use this instead:
SELECT name, number * 100.0 / SUM(number) OVER ()
FROM #names
When used with OVER clause, SUM becomes the analytical function rather than the aggregate one.
It does not shrink several records into one: instead, it returns the total value along with each record:
-- This is an aggregate function. It shrinks all records into one record and returns the total sum
WITH q (name, number) AS
(
SELECT 'test1', 0
UNION ALL
SELECT 'test2', 22
UNION ALL
SELECT 'test3', 28
)
SELECT SUM(number)
FROM q
--
50
-- This is an analytical function. It calcuates the total sum as well but does not shrink the records.
WITH q (name, number) AS
(
SELECT 'test1', 0
UNION ALL
SELECT 'test2', 22
UNION ALL
SELECT 'test3', 28
)
SELECT SUM(number) OVER ()
FROM q
--
50
50
50