SQL Query - SUM(CASE WHEN x THEN 1 ELSE 0) for multiple columns

Question

I'm looking to see if there is a better approach to the query below. What I'm trying to do is create a summary report, compiling stats by date.

 SELECT CAST(Detail.ReceiptDate AS DATE) AS 'DATE'
, SUM(CASE WHEN Detail.Type = 'TotalMailed' THEN 1 ELSE 0 END) AS 'TOTALMAILED'
, SUM(CASE WHEN Detail.Type = 'TotalReturnMail' THEN 1 ELSE 0 END) AS 'TOTALUNDELINOTICESRECEIVED'
, SUM(CASE WHEN Detail.Type = 'TraceReturnedMail' THEN 1 ELSE 0 END) AS 'TRACEUNDELNOTICESRECEIVED'
FROM
(
select SentDate AS 'ReceiptDate', 'TotalMailed' AS 'Type'
from MailDataExtract
where sentdate is not null
UNION ALL
select MDE.ReturnMailDate AS 'ReceiptDate', 'TotalReturnMail' AS 'Type'
from MailDataExtract MDE
where MDE.ReturnMailDate is not null
UNION ALL
select MDE.ReturnMailDate AS 'ReceiptDate', 'TraceReturnedMail' AS 'Type'
from MailDataExtract MDE
    inner join DTSharedData.dbo.ScanData SD ON SD.ScanDataID = MDE.ReturnScanDataID
where MDE.ReturnMailDate is not null AND SD.ReturnMailTypeID = 1
) AS Detail
GROUP BY CAST(Detail.ReceiptDate AS DATE)
ORDER BY 1   

This is only a sample of the query (which is used in a report) as there are a number of other columns and the logic for the other stats are way more complicated. Is there a more elegant approach to getting this kind of information/writing this kind of report?

Answer 1

I would change the query in the following ways:

1. Do the aggregation in subqueries. This can take advantage of more information about the table for optimizing the group by.
2. Combine the second and third subqueries. They are aggregating on the same column. This requires using a left outer join to ensure that all data is available.
3. By using count(<fieldname>) you can eliminate the comparisons to is null. This is important for the second and third calculated values.
4. To combine the second and third queries, it needs to count an id from the mde table. These use mde.mdeid.

The following version follows your example by using union all:

SELECT CAST(Detail.ReceiptDate AS DATE) AS "Date",
       SUM(TOTALMAILED) as TotalMailed,
       SUM(TOTALUNDELINOTICESRECEIVED) as TOTALUNDELINOTICESRECEIVED,
       SUM(TRACEUNDELNOTICESRECEIVED) as TRACEUNDELNOTICESRECEIVED
FROM ((select SentDate AS "ReceiptDate", COUNT(*) as TotalMailed,
              NULL as TOTALUNDELINOTICESRECEIVED, NULL as TRACEUNDELNOTICESRECEIVED
       from MailDataExtract
       where SentDate is not null
       group by SentDate
      ) union all
      (select MDE.ReturnMailDate AS ReceiptDate, 0,
              COUNT(distinct mde.mdeid) as TOTALUNDELINOTICESRECEIVED,
              SUM(case when sd.ReturnMailTypeId = 1 then 1 else 0 end) as TRACEUNDELNOTICESRECEIVED
       from MailDataExtract MDE left outer join
            DTSharedData.dbo.ScanData SD
            ON SD.ScanDataID = MDE.ReturnScanDataID
       group by MDE.ReturnMailDate;
      )
     ) detail
GROUP BY CAST(Detail.ReceiptDate AS DATE)
ORDER BY 1;

The following does something similar using full outer join:

SELECT coalesce(sd.ReceiptDate, mde.ReceiptDate) AS "Date",
       sd.TotalMailed, mde.TOTALUNDELINOTICESRECEIVED,
       mde.TRACEUNDELNOTICESRECEIVED
FROM (select cast(SentDate as date) AS "ReceiptDate", COUNT(*) as TotalMailed
      from MailDataExtract
      where SentDate is not null
      group by cast(SentDate as date)
     ) sd full outer join
    (select cast(MDE.ReturnMailDate as date) AS ReceiptDate,
            COUNT(distinct mde.mdeID) as TOTALUNDELINOTICESRECEIVED,
            SUM(case when sd.ReturnMailTypeId = 1 then 1 else 0 end) as TRACEUNDELNOTICESRECEIVED
     from MailDataExtract MDE left outer join
          DTSharedData.dbo.ScanData SD
          ON SD.ScanDataID = MDE.ReturnScanDataID
     group by cast(MDE.ReturnMailDate as date)
    ) mde
    on sd.ReceiptDate = mde.ReceiptDate
ORDER BY 1;