SQL: Number of occurrences grouped by frequency - sql-server

Question

It's not clear the exact statement for me to use here. I want to know how many times certain occurrences happen in the table when the value is A. So for some sample data:
user | value
1 | A
1 | A
1 | B
4 | A
4 | A
4 | B
5 | A
5 | A
5 | A
Would result in:
Occurrence Frequency
1 0
2 2
3 1
Which reads as: there are 0 users that have 1 value A. There are 2 users that have two value A etc.
I feel like I should use a group by and a count(*) by not clear to me how to construct it.

Answer 1

Since you want the occurrences even for 0 frequencies, you need a recursive cte which return all occurrences from 1 to the max number of occurrences.
Then you join this cte with a LEFT join to a query that aggregates on the table and aggregate once more to get the frequencies:
with
cte as (
select count(*) counter
from tablename
where value = 'A'
group by [user]
),
top_counter as (select max(counter) counter from cte),
occurrences as (
select 1 occurrence
union all
select occurrence + 1
from occurrences
where occurrence < (select counter from top_counter)
)
select o.occurrence, count(c.counter) frequency
from occurrences o left join cte c
on c.counter = o.occurrence
group by o.occurrence
See the demo.
Results:
> occurrence | frequency
> ---------: | --------:
> 1 | 0
> 2 | 2
> 3 | 1

Answer 2

You do use COUNT, just 2 of them:
WITH Counts AS(
SELECT V.[User],
COUNT([Value]) AS Frequency
FROM (VALUES(1,'A'),
(1,'A'),
(1,'B'),
(4,'A'),
(4,'A'),
(4,'B'),
(5,'A'),
(5,'A'),
(5,'A'))V([User],[Value]) --USER is a reserved keyword and should not be used for object names
WHERE V.[Value] = 'A'
GROUP BY V.[user])
SELECT V.I,
COUNT(C.Frequency) AS Frequecy
FROM (VALUES(1),(2),(3))V(I)
LEFT JOIN Counts C ON V.I = C.Frequency
GROUP BY V.I;

Answer 3

Here's my take:
with cte as (
select * from (values
(1, 'A'),
(1, 'A'),
(1, 'B'),
(4, 'A'),
(4, 'A'),
(4, 'B'),
(5, 'A'),
(5, 'A'),
(5, 'A')
) as x([User], [Value])
)
select c, count(*)
from (
select [User], count(*) as c
from cte
where [Value] = 'A'
group by [User]
) as s
group by c;
The common table expression isn't important here - it's just setting up your test data.
What you're after is an aggregation of aggretations. That is, the first level aggregate is a "count of value by user". But then you're going to get a "count of (count of value by user) by (that count)". Note, my set doesn't produce the "0 users that have 1 value A". Nor does it produce "0 users that have 17 value A". If it's important that it produce certain negative results, you'll need a list of which ones you care about and join that list with this set of results with an outer join.