SQL Query to calculate median and group by - sql

Question

I have following table.
DECLARE #TBL_RESULT Table
(
ID varchar(10),
CreateDate DateTime,
PEOPLE_CODE_ID varchar(10),
CONVERSION_DATE DateTime,
CAMPUS varchar(20),
DAYS_TOOK int
);
This table has records from January 01,2013 to date of all the leads that were received and converted.
I initially needed to find the Median time it took to convert leads that arrived in last 10 weeks and group them by Campus I was able to do that Using the SQL Query below
WITH CTE_RESULT
AS ( SELECT *
FROM #TBL_RESULT
WHERE CreateDate > DATEADD(WEEK, -10, GETDATE())
)
SELECT Campus ,
AVG(DAYS_TOOK) AS MedianTime
FROM ( SELECT CAMPUS ,
Days_Took ,
ROW_NUMBER() OVER ( PARTITION BY Campus ORDER BY Days_Took ASC ) AS AgeRank ,
COUNT(*) OVER ( PARTITION BY CAMPUS ) AS CampusCount
FROM CTE_RESULT
) x
WHERE x.AgeRank IN ( x.CampusCount / 2 + 1, ( x.CampusCount + 1 ) / 2 )
GROUP BY x.Campus
I now need to plot this trend on a graph i.e. find records the previous 10 weeks buckets and plot the median on a line chart - where each line is one campus. (Grouped by campus)
Is cursor my only option? where I will find the leads of first 10 week starting from Jan 01, do the above SQL query to get median, push it to a temp table and then find the next 10 weeks and so on.
Or is there anything better i can do?

Answer 1

Without trying to optimise your query, if you need to produce the same result across multiple 10-WEEK periods, you can expand your current (10 week ago to today) ranges to as many ranges as required, threading a PeriodEndDate throughout the query as shown below.
SQL Fiddle
MS SQL Server 2012 Schema Setup:
Query 1:
DECLARE #TBL_RESULT Table
(
ID varchar(10),
CreateDate DateTime,
PEOPLE_CODE_ID varchar(10),
CONVERSION_DATE DateTime,
CAMPUS varchar(20),
DAYS_TOOK int
);
-- fill the table with some dummy data from 2013-01-01
INSERT #TBL_RESULT (CreateDate, Campus, Days_Took)
SELECT DATEADD(D, A.Number, '20130101'), 'Campus' + Right(B.Number, 10),
ABS(CAST(NEWID() AS binary(6)) % 130) + 1
FROM master..spt_values A
JOIN master..spt_values B on B.type='P' and B.number < 50 -- 50 campuses
WHERE A.type='P'
AND DATEADD(D, A.Number, '20130101') <= GetDate();
-- This first CTE is used to create the required number of 10-week periods
WITH N(NUMBER) AS (
SELECT 0
union all
select number+1 from N
where Number <= DATEDIFF(WEEK, '20130101', GETDATE())
),
-- and from below here it's your query with the PeriodEndDate threaded through
CTE_RESULT AS (
SELECT DATEADD(WEEK, -Number, GETDATE()) PeriodEndDate,
T.*
FROM #TBL_RESULT T
CROSS JOIN N
-- you see the range built up dynamically here
WHERE CreateDate > DATEADD(WEEK, -Number-10, GETDATE())
AND CreateDate < DATEADD(WEEK, -Number, GETDATE()) +1
)
SELECT PeriodEndDate, Campus ,
AVG(DAYS_TOOK) AS MedianTime
FROM (
SELECT PeriodEndDate, CAMPUS ,
Days_Took ,
ROW_NUMBER() OVER ( PARTITION BY PeriodEndDate, Campus ORDER BY Days_Took ASC ) AS AgeRank ,
COUNT(*) OVER ( PARTITION BY PeriodEndDate, CAMPUS ) AS CampusCount
FROM CTE_RESULT
) x
WHERE x.AgeRank IN ( x.CampusCount / 2 + 1, ( x.CampusCount + 1 ) / 2 )
GROUP BY x.PeriodEndDate, x.Campus
ORDER BY x.PeriodEndDate, x.Campus;

Answer 2

It seems that you solved the hard part of the problem.
To get what you want, you need to introduce a grouping variable. In this case, I measure the number of weeks in the past and divide by 10 (SQL Server does integer division so this produces an integer).
You just then judiciously use this in the partition by and group by statements:
WITH CTE_RESULT AS (
SELECT t.*,
DATEDIFF(week, CreateDate, GETDATE()) / 10 as groupnum
FROM #TBL_RESULT t
)
SELECT Campus, groupnum, MIN(CreateDate), MAX(CreateDate),
AVG(DAYS_TOOK) AS MedianTime
FROM (SELECT t.*,
ROW_NUMBER() OVER (PARTITION BY groupnum, Campus ORDER BY Days_Took ASC ) AS AgeRank ,
COUNT(*) OVER (PARTITION BY groupnum, CAMPUS) AS CampusCount
FROM CTE_RESULT t
) x
WHERE x.AgeRank IN ( x.CampusCount / 2 + 1, ( x.CampusCount + 1 ) / 2 )
GROUP BY x.Campus, groupnum
I haven't tested this, so it might have a syntax error or two.