tsql UNION without nulls - sql-server

Question

I have the following query
SELECT MONTH, COUNT(DISTINCT VISITS) AS BRAND_VISITS, NULL AS NONB_VISITS
FROM Table1
WHERE KEYWORD_TYPE = BRAND(
AND DATE >= '2013-01-01'
GROUP BY MONTH
UNION ALL
SELECT MONTH, NULL, COUNT(DISTINCT VISITS) AS NONB_VSTS
FROM Table1
WHERE KEYWORD_TYPE = NON-BRAND
AND DATE >= '2013-01-01'
GROUP BY MONTH
I get the following results:
1 352540 NULL
2 309834 NULL
3 228764 NULL
4 236054 NULL
5 218096 NULL
6 172527 NULL
1 NULL 5337
2 NULL 14120
3 NULL 9954
4 NULL 23755
5 NULL 19771
6 NULL 30797
However, what I want is inline results without NULLS
1 352540 5337
2 309834 14120
3 228764 9954
4 236054 23755
5 218096 19771
6 172527 30797

Answer 1

You can do this with using a single statement with CASE or with an JOIN on month instead of a UNION. If you take the join approach you may need to account for null values (no visist for a keyword in a month). You will want to profile them to see which is faster with your data and table structure. It is really all about the indexes and the amount of data you need to aggregate.
Assuming you don't have to worry about nulls based on the counts in your example, here is what you want.
SELECT brand.month, brand.brand_visits,nonbrand.non_brand_visits
FROM (SELECT month, COUNT(visits) AS brand_visits
FROM Table1
WHERE keyword_type = 'BRAND'
AND date >= '2013-01-01'
GROUP BY month) brand
INNER JOIN
(SELECT month, COUNT(visits) AS non_brand_visits
FROM Table1
WHERE keyword_type = 'NON-BRAND'
AND date >= '2013-01-01'
GROUP BY month) nonbrand
ON brand.month=nonbrand.month
Here is the CASE approach. You should profile based on your actual data you are aggregating and your indexes to see which method is faster.
SELECT month,
SUM(CASE WHEN keyword_type = 'BRAND' THEN 1 ELSE 0 END) AS brand_visits,
SUM(CASE WHEN keyword_type = 'NON-BRAND' THEN 1 ELSE 0 END) AS non_brand_visits
FROM Table1
WHERE date >= '2013-01-01'
GROUP BY month
Finally, you did not provide table structure or example data so I made some assumptions above. I strongly believe you did not need the COUNT(DISTINCT in your original statement. I have removed it and verified the two statement above yield the same results. If COUNT(DISTINCT is required then the CASE approach will not work but the join approach will still work fine.

Answer 2

Using your columns:
SELECT month,
count(distinct CASE WHEN keyword_type = 'BRAND' THEN visits END) AS BRAND_VISITS,
count(distinct CASE WHEN keyword_type = 'NON-BRAND' THEN visits END) AS NONB_VSTS
FROM Table1
WHERE date >= '2013-01-01'
and keyword_type in ('BRAND','NON-BRAND')
GROUP BY month
Am tempted to believe that month is simply the month from the date column, I would prefer this solution, it con cover more years than 1 and the same query will still be valid in the year 2014
SELECT cast(dateadd(month, datediff(month, 0, date), 0) as date) month,
count(distinct CASE WHEN keyword_type = 'BRAND' THEN visits END) AS BRAND_VISITS,
count(distinct CASE WHEN keyword_type = 'NON-BRAND' THEN visits END) AS NONB_VSTS
FROM Table1
WHERE date >= '2013-01-01'
and keyword_type in ('BRAND','NON-BRAND')
GROUP BY datediff(month, 0, date)
If you want to stick with your old script, you can fix it this way:
SELECT MONTH, max(BRAND_VISITS) BRAND_VISITS, max(NONB_VISITS) NONB_VISITS
FROM
(
SELECT MONTH, COUNT(DISTINCT VISITS) AS BRAND_VISITS, NULL AS NONB_VISITS
FROM Table1
WHERE KEYWORD_TYPE = 'BRAND'
AND DATE >= '2013-01-01'
GROUP BY MONTH
UNION ALL
SELECT MONTH, NULL, COUNT(DISTINCT VISITS) AS NONB_VSTS
FROM Table1
WHERE KEYWORD_TYPE = 'NON-BRAND'
AND DATE >= '2013-01-01'
GROUP BY MONTH
) a
GROUP BY MONTH