Averages are too high when getting data from over a month - sql

Question

I was asked to alter a query to work with data from a given date selection instead of just the current month. The query should get the average sales per hour during that date range. It appears to work just fine when selecting one month of data, but when I try go to over a month, the averages appear to be higher than they ought to.
I think the problem may have to do with grouping by the day, since the day would be doubled up when data is over a month, but how would I go about fixing it? Thanks in advance.
DECLARE #Start DATETIME
DECLARE #End DATETIME
SET #Start = '6/15/2015'
SET #End = '8/15/2015'
SELECT TheHour, AVG(TheCount) AS SalesPerHour
FROM
(SELECT DATEPART(DAY, DateTimeCreated) AS TheDay,
DATEPART(HOUR, DateTimeCreated) AS TheHour,
COUNT(*) AS TheCount
FROM OrderHeader
WHERE Deleted = 0
AND OrderType = 1
AND BranchID = 4
AND BackOrderedFromID IS NULL
AND DateTimeCreated >= #Start
AND DateTimeCreated < #End
GROUP BY DATEPART(DAY, DateTimeCreated), DATEPART(HOUR, DateTimeCreated)) AS T
GROUP BY TheHour
ORDER BY TheHour
SAMPLE DATA for 6/15/2015 to 7/15/2015
TheHour SalesPerHour
5 2
6 5
7 6
8 5
9 4
10 4
11 2
12 2
13 3
14 2
15 2
16 1
SAMPLE DATA for 7/15/2015 to 8/15/2015
TheHour SalesPerHour
5 1
6 7
7 6
8 5
9 4
10 4
11 4
12 2
13 4
14 2
15 1
SAMPLE DATA for 6/15/2015 to 8/15/2015 (most values are too high?)
TheHour SalesPerHour
5 2
6 10
7 11
8 8
9 7
10 6
11 5
12 3
13 5
14 4
15 2
16 1

Answer 1

Don't use datepart(day). This gives the day of the month. When your time frame spans multiple months, datepart(day) returns the same value for different days (for instance, "1" on the first of any month).
Instead, simply cast the value to a date to remove the time component. The rest of the query remains the same:
SELECT TheHour, AVG(TheCount) AS SalesPerHour
FROM (SELECT CAST(DateTimeCreated as Date) AS TheDay,
DATEPART(HOUR, DateTimeCreated) AS TheHour,
COUNT(*) AS TheCount
FROM OrderHeader
WHERE Deleted = 0 AND OrderType = 1 AND BranchID = 4 AND
BackOrderedFromID IS NULL AND
DateTimeCreated >= #Start
DateTimeCreated < #End
GROUP BY CAST(DateTimeCreated as Date), DATEPART(HOUR, DateTimeCreated)
) dh
GROUP BY TheHour
ORDER BY TheHour;
Alternatively, you can do this without the double aggregation:
SELECT DATEPART(HOUR, DateTimeCreated) as TheHour,
(COUNT(*) * 1.0 /
COUNT(DISTINCT CAST(DateTimeCreated as Date))
) as SalesPerHour
FROM OrderHeader oh
WHERE Deleted = 0 AND OrderType = 1 AND BranchID = 4 AND
BackOrderedFromID IS NULL AND
DateTimeCreated >= #Start
DateTimeCreated < #End
GROUP BY DATEPART(HOUR, DateTimeCreated);
Also, note that AVG() of an integer value does an integer average. So, the average of 1 and 2 is 1 in SQL Server, not 1.5. In this version the query multiplies the count by 1.0 to get decimal places -- that may or may not be desirable.

Answer 2

To round a datetime down to it's nearest whole hour, use DATEADD and DATEDIFF together:
DECLARE #Start DATETIME
DECLARE #End DATETIME
SET #Start = '6/15/2015'
SET #End = '8/15/2015'
SELECT DATEPART(hour,RoundedHour) as Hour, AVG(TheCount) AS SalesPerHour
FROM
(SELECT DATEADD(hour,DATEDIFF(hour,0,DateTimeCreated),0) as RoundedHour,
COUNT(*) AS TheCount
FROM OrderHeader
WHERE Deleted = 0
AND OrderType = 1
AND BranchID = 4
AND BackOrderedFromID IS NULL
AND DateTimeCreated >= #Start
AND DateTimeCreated < #End
GROUP BY DATEADD(hour,DATEDIFF(hour,0,DateTimeCreated),0)) AS T
GROUP BY DATEPART(hour,RoundedHour)
ORDER BY DATEPART(hour,RoundedHour)
That way you don't have to think about all of the larger components (day, month, year) that you'd also want to group by, for larger ranges.

Answer 3

Since your query is using DAY as the datepart, you're effectively adding the number of sales in each hour on each day before getting the averages. For example, if a salesperson has 10 sales in the 5pm hour on Jan. 1st and 12 sales on Feb. 1st in the 5pm hour then you're going to get an intermediate value of 22 sales for "day 1". You end up averaging these over the days of each individual month, but then not over the days themselves.
You could use the DATEPART of DY (day of year) instead, but then your query would experience the same issue if you started to span years. Instead, just CAST the DATETIME as a DATE to get rid of the time portion, or even better yet, use a windowed function to get your numbers, like so:
;WITH CTE_HourBreakdown AS
(
SELECT
DATEPART(HOUR, DateTimeCreated) AS hr,
COUNT(*) OVER (PARTITION BY (YEAR(DateTimeCreated), DATEPART(DY, DateTimeCreated), DATEPART(HOUR, DateTimeCreated)) AS cnt
FROM
OrderHeader
)
SELECT
hr,
AVG(CAST(cnt AS DECIMAL(10, 2)))
FROM
CTE_HourBreakdown
GROUP BY
hr
There's likely a better way to do this with windowed functions, but this was the first thing that came to me. Also, note that if there are no sales in an hour this method does NOT average that into the results. For example, if on one day between 4pm and 5pm there are no sales and the next day there are 2 sales this will show an average of 2 sales between 4pm and 5pm instead of 1 sale on average. If you want to account for that then you'll need a method to distinguish zero-sale hours from hours when no one is working.