How to get difference between two rows for a column field?

Viewed 234310

I have a table like this:

rowInt  Value
2       23
3       45
17      10
9       0
....

The column rowInt values are integer but not in a sequence with same increament. I can use the following sql to list values by rowInt:

SELECT * FROM myTable ORDER BY rowInt;

This will list values by rowInt. How can get get the difference of Value between two rows with the result like this:

rowInt   Value Diff
2        23    22    --45-23
3        45    -35   --10-45
9        0     -45   --0-45
17       10    10    -- 10-0
....

The table is in SQL 2005 (Miscrosoft)

8 Answers
SELECT
   [current].rowInt,
   [current].Value,
   ISNULL([next].Value, 0) - [current].Value
FROM
   sourceTable       AS [current]
LEFT JOIN
   sourceTable       AS [next]
      ON [next].rowInt = (SELECT MIN(rowInt) FROM sourceTable WHERE rowInt > [current].rowInt)

EDIT:

Thinking about it, using a subquery in the select (ala Quassnoi's answer) may be more efficient. I would trial different versions, and look at the execution plans to see which would perform best on the size of data set that you have...


EDIT2:

I still see this garnering votes, though it's unlikely many people still use SQL Server 2005.

If you have access to Windowed Functions such as LEAD(), then use that instead...

SELECT
  RowInt,
  Value,
  LEAD(Value, 1, 0) OVER (ORDER BY RowInt) - Value
FROM
  sourceTable
SELECT rowInt, Value,
       COALESCE(
       (
       SELECT TOP 1 Value
       FROM myTable mi
       WHERE mi.rowInt > m.rowInt
       ORDER BY
             rowInt
       ), 0) - Value AS diff
FROM  myTable m
ORDER BY
      rowInt

SQL Server 2012 and up support LAG / LEAD functions to access the previous or subsequent row. SQL Server 2005 does not support this (in SQL2005 you need a join or something else).

A SQL 2012 example on this data

/* Prepare */
select * into #tmp
from
(
    select 2  as rowint,      23 as Value
    union select 3,       45
    union select 17,      10
    union select 9,       0
) x


/* The SQL 2012 query */
select rowInt, Value, LEAD(value) over (order by rowInt) - Value  
from #tmp

LEAD(value) will return the value of the next row in respect to the given order in "over" clause.

If you really want to be sure of orders, use "Row_Number()" and compare next record of current record (take a close look at "on" clause)

T1.ID + 1 = T2.ID

You are basically joining next row with current row, without specifying "min" or doing "top". If you have a small number of records, other solutions by "Dems" or "Quassanoi" will work fine.

with T2 as (
    select  ID = ROW_NUMBER() over (order by rowInt),
            rowInt, Value
    from    myTable
)
select  T1.RowInt, T1.Value, Diff = IsNull(T2.Value, 0) - T1.Value
from    (   SELECT  ID = ROW_NUMBER() over (order by rowInt), *
            FROM    myTable ) T1
        left join T2 on T1.ID + 1 = T2.ID
ORDER BY T1.ID

Does SQL Server support analytic functions?

select   rowint,
         value,
         value - lag(value) over (order by rowint) diff
from     myTable
order by rowint
/

I'd just make a little function for that. Toss in the two values you need to know the difference between and have it subtract the smaller from the larger value. Something like:

CREATE FUNCTION [dbo].[NumDifference] 
    (   @p1 FLOAT,
        @p2 FLOAT )
RETURNS FLOAT
AS
BEGIN
    DECLARE @Diff FLOAT
    IF @p1 > @p2 SET @Diff = @p1 - @p2 ELSE SET @Diff = @p2 - @p1
    RETURN @Diff
END

In a query to get the difference between column a and b:

SELECT a, b, dbo.NumDifference(a, b) FROM YourTable
Related