Use of SUM and COUNT with GROUP BY

Question

All,
I have a set like below:
State DOB StartDt EndDt Desc Category MemberID SubID Code
NC 2003-01-30 2014-01-01 2014-01-31 Child B 123456 2 38
NC 2003-01-30 2014-01-01 2014-01-31 Child B 123456 2 39
NC 2003-01-30 2014-02-01 2014-05-31 Child B 123456 2 38
NC 2003-01-30 2014-02-01 2014-05-31 Child B 123456 2 39
NC 2003-01-30 2014-06-01 2014-07-31 Child B 123456 2 38
NC 2003-01-30 2014-06-01 2014-07-31 Child B 123456 2 39
NC 2014-01-17 2014-01-01 2014-07-31 Infant S 456789 1 49
NC 2014-02-04 2014-02-01 2014-07-31 Infant S 246376 3 49
-- MemberID and SubID identify 1 member
-- Member 123456 has 2 distinct "Code" i.e. 38,39 but each code has different "StartDt" and "EndDt"
-- Expected Result
State Desc Category CountOfDistinctCode TotalMonthsEnrolled
NC Child B 2 (38 and 39 for same member) 7 months (1 + 4 + 2) (Difference between StartDt and EndDt:1 (1/31/2014 - 1/1/2014) + 4 (5/31/2014 - 2/1/2014).. )
NC Infant S 2 (Same code 49 but different member) 13 months (7+6) (7/31/2014 - 1/1/2014 = 7 months + 7/31/2014 - 2/1/2014 = 6 months)
I tried doing a count of distinct Code and the summing up the member months, grouped by State, Desc, Category, but I am not able to get 2 and 7 for child, it somehow calculates Months as 14 (7+7). Please let me know what you guys suggest.

Answer 1

OK, so we need a different approach. We need a table of numbers, a concept that I discuss here:
http://www.sommarskog.se/arrays-in-sql-2005.html#numbersasconcept
(Only read down to the next header.)
We join the numbers to temp table with BETWEEN over the numbers. Then we count the distinct combination of member ID and number.
We also need to make an adjustment how to build the string for the DISTINCT. I initially assumed that your columns were integer, why I used str() which produces a right-adjusted fixed-length string of 10 characters. (The default, you can specify a different
width as the second parameter.) But str() expects a float value as input, and will fail if for instance member id would be non-numeric. We cannot just concatenate the strings, since in that case (MemberID, SubID) = ('12345', '61') would be
the same as ('123456', '1'). So we must convert to char to get fixed length. All that said, and some more test data added, here is a solution:
CREATE TABLE #Test
[State] CHAR(2),
DOB DATE,
StartDt DATE,
EndDt DATE,
[Desc] VARCHAR(8),
Category CHAR(1),
MemberID VARCHAR(10),
SubID VARCHAR(2),
Code VARCHAR(5)
INSERT INTO #Test
VALUES
('NC', '20130130', '20140101', '20140120', 'Child', 'B', '123456', '2', '38'),
('NC', '20130130', '20140121', '20140131', 'Child', 'B', '123456', '2', '39'),
('NC', '20130130', '20140201', '20140531', 'Child', 'B', '123456', '2', '38'),
('NC', '20130130', '20140401', '20140613', 'Child', 'B', '123456', '2', '39'),
('NC', '20130130', '20140601', '20140731', 'Child', 'B', '123456', '2', '38'),
('NC', '20130130', '20140614', '20140731', 'Child', 'B', '123456', '2', '39'),
('NC', '20130129', '20140601', '20140731', 'Child', 'B', '9123456', '1', '38'),
('NC', '20130129', '20140614', '20140831', 'Child', 'B', '9123456', '1', '39'),
('NC', '20140117', '20140101', '20140731', 'Infant', 'S', '456789', '1', '49'),
('NC', '20140204', '20140201', '20140731', 'Infant', 'S', '246376', '3', '49')
SELECT * FROM #Test ORDER BY MemberID, StartDt, EndDt
SELECT
[State]
, [Desc]
, Category
, COUNT(DISTINCT convert(char(10), MemberID) +
convert(char(2), SubID) +
convert(char(5), Code)) AS TotalCnt
, COUNT(DISTINCT convert(char(10), MemberID) +
convert(char(2), SubID) +
str(N.Number)) AS CalMonths
FROM #Test t
JOIN Numbers N ON N.Number BETWEEN
datediff(MONTH, '19000101', t.StartDt) AND
datediff(MONTH, '19000101', t.EndDt)
GROUP BY [State], [Desc], Category
go
DROP TABLE #Test
Erland Sommarskog, SQL Server MVP, [email protected]