AVG and COUNT in SQL Server - sql

Question

I have a rating system in which any person may review other. Each person can be judged by one person more than once. For the calculation of averages, I would like to include only the most current values​​.
Is this possible with SQL?
Person 1 rates Person 2 with 5 on 1.2.2011 <- ignored because there is a newer rating of person 1
Person 1 rates Person 2 with 2 on 1.3.2011
Person 2 rates Person 1 with 6 on 1.2.2011 <-- ignored as well
Person 2 rates Person 1 with 3 on 1.3.2011
Person 3 rates Person 1 with 5 on 1.5.2011
Result:
The Average for Person 2 is 2.
The Average for Person 1 is 4.
The table may look like this: evaluator, evaluatee, rating, date.
Kind Regards
Michael

Answer 1

It's perfectly possible.
Let's assume your table structure looks like this:
CREATE TABLE [dbo].[Ratings](
[Evaluator] varchar(10),
[Evaluatee] varchar(10),
[Rating] int,
[Date] datetime
);
and the values like this:
INSERT INTO Ratings
SELECT 'Person 1', 'Person 2', 5, '2011-02-01' UNION
SELECT 'Person 1', 'Person 2', 2, '2011-03-01' UNION
SELECT 'Person 2', 'Person 1', 6, '2011-02-01' UNION
SELECT 'Person 2', 'Person 1', 3, '2011-03-01' UNION
SELECT 'Person 3', 'Person 1', 5, '2011-05-01'
Then the average rating for Person 1 is:
SELECT AVG(Rating) FROM Ratings r1
WHERE Evaluatee='Person 1' and not exists
(SELECT 1 FROM Ratings r2
WHERE r1.Evaluatee = r2.Evaluatee AND
r1.evaluator=r2.evaluator AND
r1.date < r2.date)
Result:
4
Or for all Evaluatee's, grouped by Evaluatee:
SELECT Evaluatee, AVG(Rating) FROM Ratings r1
WHERE not exists
(SELECT 1 FROM Ratings r2
WHERE r1.Evaluatee = r2.Evaluatee AND
r1.evaluator = r2.evaluator AND
r1.date < r2.date)
GROUP BY Evaluatee
Result:
Person 1 4
Person 2 2
This might look like it has an implicit assumption that no entries exist with the same date;
but that's actually not a problem: If such entries can exist, then you can not decide which of these was made later anyway; you could only choose randomly between them. Like shown here, they are both included and averaged - which might be the best solution you can get for that border case (although it slightly favors that person, giving him two votes).
To avoid this problem altogether, you could simply make Date part of the primary key or a unique index - the obvious primary key choice here being the columns (Evaluator, Evaluatee, Date).

Answer 2

declare #T table
(
evaluator int,
evaluatee int,
rating int,
ratedate date
)
insert into #T values
(1, 2, 5, '20110102'),
(1, 2, 2, '20110103'),
(2, 1, 6, '20110102'),
(2, 1, 3, '20110103'),
(3, 1, 5, '20110105')
select evaluatee,
avg(rating) as avgrating
from (
select evaluatee,
rating,
row_number() over(partition by evaluatee, evaluator
order by ratedate desc) as rn
from #T
) as T
where T.rn = 1
group by evaluatee
Result:
evaluatee avgrating
----------- -----------
1 4
2 2

Answer 3

This is possible to do, but it can be REALLY harry - SQL was not designed to compare rows, only columns. I would strongly recommend you keep an additional table containing only the most recent data, and store the rest in an archive table.
If you must do it this way, then I'll need a full table structure to try to write a query for this. In particular I need to know which are the unique indexes.