Combining multiple rows in TSQL query - sql

Question

I have a table with data like this
Road Item Response added_on
1 82 Yes 7/11/16
1 83 Yes 7/11/16
1 84 Yes 7/11/16
2 82 Yes 8/11/16
2 83 No 8/11/16
2 85 Yes 8/11/16
This reflects an assessment of a road where 'item' is things being assessed.
Some items will always be done during an assessment (82, 83) where others are optional (84, 85).
I want to return something that combines all of the assessment results for a road/date, returning null if that item was not assessed. And also only returning last month's results. For example
Road 82 83 84 85 added_on
1 Yes Yes Yes 7/11/16
2 Yes No Yes 8/11/16
I have tried a multiple self joins like this but it's returning nothing.
FROM assess AS A
JOIN assess AS B
ON A.road = B.road AND a.added_on = B.added on
JOIN assess AS C
ON A.road = C.road AND a.added_on = C.added on
JOIN assess AS D
ON A.road = D.road AND a.added_on = D.added on
WHERE A.item = '81'
AND B.item = '82'
AND (C.item = '83' OR C.item IS NULL)
AND (D.item = '84' OR D.item IS NULL)
AND datepart(month,A.added_on) = datepart(month,getdate()) -1
To clarify,
-no road is assessed more than once a day
-each item is only assessed once, and sometimes is NULL i.e. not applicable
-multiple roads are assessed each day
-this table has other assessments but we aren't worried about those.
Any ideas? Using SQL server 2008. Thanks.

Answer 1

Assuming you need to go Dynamic
Declare #SQL varchar(max)
Select #SQL = Stuff((Select Distinct ',' + QuoteName(Item) From YourTable Order By 1 For XML Path('')),1,1,'')
Select #SQL = 'Select [Road],' + #SQL + ',[added_on]
From YourTable
Pivot (max(Response) For Item in (' + #SQL + ') ) p'
Exec(#SQL);
Returns
EDIT - The SQL Generated is as follows. (just in case you can't go
dynamic)
Select [Road],[82],[83],[84],[85],[added_on]
From YourTable
Pivot (max(Response) For Item in ([82],[83],[84],[85]) ) p

Answer 2

Another way of achieving this is less elegant, but uses basic operations if you don't want to use pivot.
Load up test data
create table #assess ( road int, item varchar(10), response varchar(3), added_on date )
insert #assess( road, item, response, added_on )
values
(1, '82', 'Yes', '2016-07-11' )
, (1, '83', 'Yes', '2016-07-11' )
, (1, '84', 'Yes', '2016-07-11' )
, (2, '82', 'Yes', '2016-08-11' )
, (2, '83', 'No', '2016-08-11' )
, (2, '85', 'Yes', '2016-08-11' )
Process the data
-- Get every possible `item`
select distinct item into #items from #assess
-- Ensure every road/added_on combination has all possible values of `item`
-- If the combination does not exist in original data, leave `response` as blank
select road, added_on, i.item, cast('' as varchar(3)) as response into #assess2
from #items as i cross join #assess AS A
group by road, added_on, i.item
update a set response = b.response
from #assess2 a inner join #assess b on A.road = B.road AND a.added_on = B.added_on AND a.item = b.item
-- Join table to itself 4 times - inner join if `item` must exist or left join if `item` is optional
select a.road, a.added_on, a.response as '82', b.response as '83', c.response as '84', d.response as '85'
FROM #assess2 AS A
INNER JOIN #assess2 AS B ON A.road = B.road AND a.added_on = B.added_on
LEFT JOIN #assess2 AS C ON A.road = C.road AND a.added_on = C.added_on
LEFT JOIN #assess2 AS D ON A.road = D.road AND a.added_on = D.added_on
WHERE A.item = '82'
AND B.item = '83'
AND (C.item = '84' OR C.item IS NULL)
AND (D.item = '85' OR D.item IS NULL)
--AND datepart(month,A.added_on) = datepart(month,getdate()) -1
The resultset is:
road added_on 82 83 84 85
1 2016-07-11 Yes Yes Yes
2 2016-08-11 Yes No Yes

Answer 3

I would do this using conditional aggregation:
select road,
max(case when item = 82 then response end) as response_82,
max(case when item = 83 then response end) as response_83,
max(case when item = 84 then response end) as response_84,
max(case when item = 85 then response end) as response_85,
added_on
from t
group by road, added_on
order by road;
For the month component, you can add a where clause. One method is:
where year(date_added) * 12 + month(date_added) = year(getdate())*12 + month(getdate()) - 1
Or, you can use logic like this:
where date_added < dateadd(day, 1 - day(getdate()), cast(getdate() as date)) and
date_added >= dateadd(month, -1, dateadd(day, 1 - day(getdate()), cast(getdate() as date)))
The second looks more complicated but it is sargable, meaning that an index on date_added can be used (if one is available).