Rank & find difference of value in the same column - sql

Question

I have the below table -
Here, I have created the "Order" column by using the rank function partitioned by case_identifier ordered by audit_date.
Now, I want to create a new column as below -
The logic for the new column would be -
select *,
case when [order] = '1' then [days_diff]
else (val of [days_diff] in rank 2) - (val of [days_diff] in rank 1) ...
end as '[New_Col]'
from TABLE
Can you please help me with the syntax? Thanks.

Answer 1

Take a look at the LAG function. It will provide you with what you want.
something like:
declare #temptable TABLE (case_id varchar(2), row_order int, days_diff float)
INSERT INTO #temptable values ('A',1,5)
INSERT INTO #temptable values ('A',2,3)
INSERT INTO #temptable values ('A',3,2)
INSERT INTO #temptable values ('B',1,5)
INSERT INTO #temptable values ('B',2,1)
--select * from #temptable
SELECT case_id,row_order, LAG(days_diff,1) OVER (PARTITION BY case_id ORDER BY row_order) AS prev_row,days_diff,
CASE
WHEN row_order = 1 THEN days_diff
ELSE LAG(days_diff,1) OVER (PARTITION BY case_id ORDER BY row_order) - days_diff
END AS newcolumn
FROM #temptable
order by case_id,row_order asc
SELECT case_id,row_order,LAG(days_diff,1) OVER (PARTITION BY case_id ORDER BY row_order) AS prev_row, days_diff,
COALESCE(LAG(days_diff,1) OVER (PARTITION BY case_id ORDER BY row_order) - days_diff , days_diff)
FROM #temptable
order by case_id,row_order asc
Other answers will use a coalesce in place of the CASE statement. It's probably faster, but I feel like this is clearer.
If you run both and look at the execution plans they are the same.

Answer 2

I believe the following query gets you what you want.
SELECT a.*,
'NEW DAYS DIFF' =
CASE
WHEN a.[order] = 1 THEN a.days_diff
ELSE a.days_diff - b.days_diff
END
FROM dbo.tblCaseDaysDiff a
INNER JOIN dbo.tblCaseDaysDiff b
ON
(b.CASE_ID = a.CASE_ID AND b.[order] + 1 = a.[order] ) -- Get the current row and compare with the next highest order
OR (b.CASE_ID = a.CASE_ID AND b.[order] = 1 AND a.[order] = 1) --WHEN ORDER = 1 Get days_diff value
ORDER BY a.CASE_ID, a.[order]

Answer 3

As it happens, you're already hip-deep in window functions, and as others have pointed out, LAG will do the trick. In general, though, you can always get the difference of two rows by making one row: by joining the table to itself.
with T (CASE_IDENTIFIER, AUDIT_DATE, order, days_diff)
as (
... your query ...
)
select a.*,
a.days_diff - coalesce(b.days_diff, 0) as delta_days_diff
from T as a left join T as b
on a.CASE_IDENTIFIER = b.CASE_IDENTIFIER
and b.days_diff = a.days_diff - 1

Answer 4

LAG METHOD
SELECT
CASE_IDENTIFIER
,AUDIT_DATE
,[order]
,days_diff
,days_diff - ISNULL(LAG(days_diff,1) OVER (PARTITION BY CASE_IDENTIFIER ORDER BY [order]),0) AS New_Column
FROM #Table
SELF JOIN METHOD
SELECT
t1.CASE_IDENTIFIER
,AUDIT_DATE
,t1.[order]
,t1.days_diff
,t1.days_diff - ISNULL(t2.days_diff,0) AS New_Column
FROM
#Table t1
LEFT JOIN #Table t2
ON t1.CASE_IDENTIFIER = t2.CASE_IDENTIFIER
AND t1.[order] - 1 = t2.[order]
I feel like a lot of the other answers are on the right track but there are some nuances or easier ways of writing some of them. Or also some of the answer provide the write direction but had something wrong with their join or syntax. Anyways, you don't need the CASE STATEMENT whether you use the LAG of SELF JOIN Method. Next COALESCE() is great but you are only comparing 2 values so ISNULL() works fine too for sql-server but either will do.