TSQL Question: Group by on multiple columns

Question

I have a table with multiple columns that represent occurrences of an action.

simpleTableExample

id  type $$$$$ value1 value2 value3 ... value15
--  ----  --   -----  -----  -----      -----
1    "I"   1    "a"    "b"    ""         ""
2    "O"   1    "a"    "d"    "f"        "z"
3    "I"   1    "d"    "b"    ""         ""
4    "O"   1    "g"    "l"    ""         ""
5    "I"   1    "z"    "g"    "a"         ""
6    "I"   1    "z"    "g"    "a"         "a"

not really sure how to represent the data, but the above should show it... Per row, I have multiple columns that I want to group by. Focusing on the "a" part, I'd like the output such that :

Type     Value    Count  Sum
-----    -----    -----  ---
"I"      "A"      "3"     3
"O"      "A"      "2"     1

Just not sure how to go about, or even the best way, to group by with multiple columns. If it was a "tall" table with a sequence number, it'd be a snap...

One thing I'm worried about... is if I have the same value in multiple rows (they have the same action done multiple times), and I stack the columns and try to sum rows. So for example row 6 has 3 and 15 with value of a, and I try to Sum() the totals of that group, I don't want 6 counted twice (shown above

Edit: Trying to clarify here... I have a row of customers with "actions". The actions are collapsed into value1 thru value15. I want to get a count of "type, value" groups.

The count I had earlier was wrong... I want the distinct customers that have had a certain action, not the number of certain actions. So I-A should be a count of 4, using the above visible table.

Final NoteCode Review question for what I came up with. Unpivot to build a list of unique date/value pairs.

Answer 1

Assuming SQL 2005 or greater you can use UNPIVOT and then do your GROUP BYs and COUNTs

Answer 2

Do you have the option to put that table in First Normal Form and break those repeating columns out to a child table? Because, if we put aside the theory of why 1NF is good, the practical reality is that non-normalized examples like this make simple queries tough.

With that being said, if you are stuck with this non-normalized table, you have to code up something like this:

select type
     , 'A' as Column2
     , sum(case when value1='A' or value2='A' or.....
               then 1 else 0 end) as theCount
  from theTable
 group by type

If you want to query for multiple values, not just 'A', you have to pivot the table in the query, which involves oh about 15 unions:

select type,value,count(*) from (
    select type,value1 as value from theTable
    union all
    select type,value2 as value from theTable
    union all
    select type,value3 as value from theTable
    ...and so on...
) x
group by type,value