Question

I have a dataset that analogously looks like this:

X | U | datetime
-------------
1 | 1 | 1/1/12
1 | 2 | 1/1/12
1 | 2 | 1/1/12
1 | 2 | 1/1/12
1 | 4 | 1/1/12
2 | 2 | 2/1/12
2 | 3 | 2/1/12
1 | 3 | 3/1/12
2 | 4 | 3/1/12
3 | 2 | 4/1/12

it is a log of visits. X is the id of the thing visited, and U is the user id

I need to compute two statistics.

Given a value for X (x):

1. "new visitors": count the number of unique users that made their first visit for any X to x.

use cases:

• A user has only visited x once -> counts as 1
• A user has only visited !x once -> counts as 0
• A user has only visited x twice -> counts as 1
• A user has only visited !x twice -> counts as 0
• A user has visited many Xs, where their first visit of any X is x -> counts as 1
• A user has visited many Xs, where their first visit of any X is !x -> counts as 0

Examples from above data:

X | Count
---------
1 | 3
2 | 1
3 | 0

2. "returning visitors": count the number of unique users that have visited x more than once OR have visited x once, but have visited another X previously (i.e. visits made after their single visit to x do not count)

Examples from above data:

X | Count
---------
1 | 3
2 | 2
3 | 1 

I’m using SQL Server 2008.

Update

This appears to answer Q1, although it’s not very fast 🙁

select x.X, COUNT(1)
from (
    select t1.X
    from @t t1
    group by t1.X, t1.U
    having (select COUNT (1) from @t t2 where t2.u= t1.U and t2.OccurredOn < MIN(t1.OccurredOn)) =0 
) x 
group by x.X

Update 2

I think this is (2)

select t.X, COUNT(1)
from @t t
left join (
    select t.U, MIN(t.OccurredOn) as O
    from @t t
    group by t.U
) x on t.U = x.U and t.OccurredOn <= x.O
where x.U is null
group by t.X

Answer 1

For the first case, you need a sub-query to join to that will filter out all user-thing visits that aren’t the first of their kind. So you’ll have something like

select X, count(*) [First Visits]
from table t1
     join (select U, min(datetime) firstvisit 
           from table 
           group by U) t2 on t1.datetime = t2.firstvisit and t1.U = t2.U
group by X

Edit: I think your solution to the second one is fine, except that the join will be faster if you replace the <= with simply =.