Question

SQL Server 2005

I have an SQL Function (ftn_GetExampleTable) which returns a table with multiple result rows

EXAMPLE

ID    MemberID MemberGroupID   Result1    Result2 Result3 Year Week
1     1        1               High Risk  2       xx      2011 22
2     11       4               Low Risk   1       yy      2011 21 
3     12       5               Med Risk   3       zz      2011 25
etc.

Now I do a count and group by on a table above this for Result 2 for instance so I get

SELECT MemberGroupID, Result2, Count(*) AS ExampleCount, Year, Week 
FROM ftn_GetExampleTable 
GROUP BY MemberGroupID, Result2, Year, Week

MemberGroupID  Result2  ExampleCount Year Week
1              2        4            2011 22
4              1        2            2011 21
5              3        1            2011 25

Now imagine when I go to graph this new table between Weeks 20 and 23 of Year 2011, you’ll see that it won’t graph 20 or 23 or certain groups or even certain results in this example as they are not in the included data, so I need “false data” inserted into this table which has all the possibilities so they at least show on a graph even if the count is 0, does this make sense?

I am wondering on the easiest and kind of most dynamic way as it could be Result1 or Result3 I want to Graph on (different column types).

Thanks in advance

Answer 1

It looks like your dimensions are: MemberGroupID,Result2, and week (Year,Week).

One approach to solving this is to generate a list of all values you want for all the dimensions, and produce a cartesian product of them. As an example,

SELECT m.MemberGroupID, n.Result2, w.Year, w.Week
  FROM (SELECT MemberGroupID FROM ftn_GetExampleTable GROUP BY MemberGroupID) m
 CROSS
  JOIN (SELECT Result2       FROM ftn_GetExampleTable GROUP BY Result2      ) n
 CROSS
  JOIN (SELECT Year, Week FROM myCalendar WHERE ... ) w

You don’t necessarily need a table named myCalendar. (That approach does seem to be the popular one.) You just need a row source from which you can derive a list of (Year, Week) tuples. (There are answers to the question elsewhere in Stackoverflow, how to generate a list of dates.)

And the list of MemberGroupID and Result2 values doesn’t have to come from the ftn_GetExampleTable rowsource, you could substitute another query.

With a cartesian product of those dimensions, you’ve got a complete “grid”. Now you can LEFT JOIN your original result set to that.

Any place you don’t have a matching row from the “gappy” result query, you’ll get a NULL returned. You can leave the NULL, or replace it with a 0, which is probably what you want if it’s a “count” you are returning.

SELECT d.MemberGroupID
     , d.Result2
     , d.Year
     , d.Week
     , IFNULL(r.ExampleCount,0) as ExampleCount
  FROM ( <dimension query from above> ) d
  LEFT
  JOIN ( <original ExampleCount query> ) r
    ON r.MemberGroupID  = d.MemberGroupID
    AND r.Result2       = d.Result2
    AND r.Year          = d.Year
    AND r.Week          = d.Week

That query can be refactored to make use of Common Table Expressions, which makes the query a little easier to read, especially if you are including multiple measures.

; WITH d AS ( /* <dimension query with no gaps (example above)> */ 
       )
     , r AS ( /* <original query with gaps> */
              SELECT MemberGroupID, Result2, Count(*) AS ExampleCount, Year, Week
              FROM ftn_GetExampleTable
              GROUP BY MemberGroupID, Result2, Year, Week
       )
SELECT d.*
     , IFNULL(r.ExampleCount,0)
  FROM d
  LEFT 
  JOIN r 
    ON r.Year=d.Year AND r.Week=d.Week AND r.MemberGroupID = d.MemberGroupID
       AND r.Result2 = d.Result2

This isn’t a complete working solution to your problem, but it outlines an approach you can use.