Multiple joins to single table in SQL Server query throwing off counts - sql

Question

I am writing a stored procedure in SQL Server Management Studio 2005 to return a list of states and policy counts for two different time periods, month to date and year to date. I have created a couple of views to gather the required data and a stored procedure for use in a Reporting Services report.
Below is my stored procedure:
SELECT DISTINCT
S.[State],
COUNT(HP_MTD.PolicyID) AS PolicyCount_MTD,
COUNT(HP_YTD.PolicyID) AS PolicyCount_YTD
FROM tblStates S
LEFT OUTER JOIN vwHospitalPolicies HP_MTD ON S.[State] = HP.[State]
AND HP.CreatedDate BETWEEN DATEADD(MONTH, -1, GETDATE()) AND GETDATE()
LEFT OUTER JOIN vwHospitalPolicies HP_YTD ON S.[State] = HP.[State]
AND HP.CreatedDate BETWEEN DATEADD(YEAR, -1, GETDATE()) AND GETDATE()
GROUP BY S.[State]
ORDER BY S.[State] ASC
The problem I am running into is my counts are bloating when a second LEFT OUTER JOIN is added, even the COUNT() that isn't referencing the second join. I need a left join since not all states will have policies for the given period, but they should still appear on the report.
Any suggestions would be greatly appreciated!

Answer 1

It sounds like you need:
COUNT(DISTINCT HP_MTD.PolicyID) AS PolicyCount_MTD,
COUNT(DISTINCT HP_YTD.PolicyID) AS PolicyCount_YTD
instead of:
COUNT(HP_MTD.PolicyID) AS PolicyCount_MTD,
COUNT(HP_YTD.PolicyID) AS PolicyCount_YTD
Your original query is including the number of matching rows in the second join. Adding a DISTINCT clause inside the COUNT limits it to unique occurrences of the PolicyID.

Answer 2

I prefer CTEs for this sort of work - they're basically a sort of inline view.
Your actual problem is that some policies are being counted twice - once for the month-to-date, and once for the year-to-date - in both count columns. So, if you have source tables like this:
year-to-date
state policy
==================
1 1
1 2
1 3
2 4
month-to-date
state policy
==================
1 1
1 2
The result table for the JOINs (before COUNT is assesed) looks like this:
temp
state monthPolicy yearPolicy
================================
1 1 1
1 1 2
1 1 3
1 2 1
1 2 2
1 2 3
2 - 4
Clearly not what you want. Often, the solution is to use a CTE or other table reference to present summed-up records for the final join. Something like this:
WITH PoliciesMonthToDate (state, count) as (
SELECT state, COUNT(*)
FROM vwHospitalPolicies
WHERE createdDate >= DATEADD(MONTH, -1, GETDATE())
AND createdDate < DATEADD(DAY, 1, GETDATE())
GROUP BY state),
PoliciesYearToDate (state, count) as (
SELECT state, COUNT(*)
FROM vwHospitalPolicies
WHERE createdDate >= DATEADD(YEAR, -1, GETDATE())
AND createdDate < DATEADD(DAY, 1, GETDATE())
GROUP BY state)
SELECT a.state, COALESCE(b.count, 0) as policy_count_mtd,
COALESCE(c.count, 0) as policy_count_ytd
FROM tblStates a
LEFT JOIN PoliciesMonthToDate b
ON b.state = a.state
LEFT JOIN PoliciesYearToDate c
ON c.state = a.state
ORDER BY a.state
(minor nitpick - don't use prefixes for tables and views. It's noise, and if for some reason you switch between them, it would require a code change. That, or the name would then be misleading)