Question

Is it possible to calculate the average time to reply to a message just with the following columns:

id | ref | client | admin | date | message

• id is the unique message number
• ref is the message reference number, which is not unique (searching for ref, ordering by date will show a conversation)
• client is ID of client, if it is a client message, else 0 if not a client
• admin is ID of admin, if it is an admin message, else 0 if not a client
• date is set up using datetime being the time of the message
• message being the message sent

Example Data:

1  | 1   | 1      | 0     | 2011-11-07 01:00:00 | ABC
2  | 1   | 1      | 0     | 2011-11-07 01:01:00 | DEF
3  | 1   | 0      | 1     | 2011-11-07 01:05:00 | abc
4  | 2   | 2      | 0     | 2011-11-07 01:10:00 | 123
5  | 3   | 1      | 0     | 2011-11-07 01:11:00 | abc
6  | 2   | 0      | 1     | 2011-11-07 01:20:00 | a
7  | 3   | 0      | 2     | 2011-11-07 02:11:00 | b

Ideally looking for the average time period between a client message and an admin message, though if there are 2 client messages from the same client (ie. admin didn’t reply to first message before client added their 2nd message) with the same ref.

From example, time for (1) = 5 minutes, (2) = 10 minutes, (3) = 60 minutes … average = 25 minutes (1500 seconds – happy to get work with seconds)

I’m not sure how to even begin working on this…. I do hope someone can help :S

Answer 1

Your question is well formulated but leaves room for interpretation. This is one interpretation:

SELECT avg(TIMESTAMPDIFF(SECOND, c.c_date, a.a_date) AS avg_time_to_response
FROM   (
    SELECT ref, min(date) AS c_date
    FROM   tbl
    WHERE  client > 0
    GROUP  BY 1
    ) c
JOIN  (
    SELECT ref, min(date) AS a_date
    FROM   tbl
    WHERE  admin > 0
    GROUP  BY 1
    ) a USING (ref)
WHERE a.a_date > c.c_date;

Gives you the average time that passes between the first client posting and the first admin posting per thread (message reference number).

Unanswered messages are ignored. Threads started by Admins would confuse the result with negative durations, so I excluded those. Only the first response time per thread goes into the result. Additional postings on the same thread are ignored here.

Read the manual here about TIMESTAMPDIFF().
_{Thanx to @MrJ and @Vincent for pointing out the mistake with the subtraction of timestamps!}

Concerning GROUP BY 1

I quote the manual here:

Columns selected for output can be referred to in ORDER BY and GROUP
BY clauses using column names, column aliases, or column positions.
Column positions are integers and begin with 1:

Emphasis mine. So I group on the first column that is selected (ref in both cases). Just a notational shortcut.