MySQL XOR produces incorrect results

The problem is the subquery, where you combine in the select list both an aggregated (max(cr.posted_dt)) and non-aggregated expressions/columns from the tables.

MySQL allows you to run this kind of inconsistent queries - with default settings - which basically means it depends on the developer to write consistent queries and not fire themselves on the foot. You can change the sql mode to ONLY_FULL_GROUP_BY and see what happens if you try to run your query.

Now to solve the issue, it seems you want a [greatest-n-per-group] type of query. There are several ways to do this, all quite complicated in MySQL (because it lacks window functions). Check the relevant tags in this site and in the SO main site.

Here's one way:

SELECT c.id,
       c.title,
       c.member_id,
       cr.mem_id    AS last_replier,
       CONCAT(cr.posted_date, 'T', cr.posted_time) AS last_posted_dt
FROM complaints AS c
  LEFT JOIN complaints_reply AS cr
    ON  cr.id = 
        ( SELECT crl.id
          FROM complaints_reply AS crl
          WHERE crl.complaint_id = c.id
            AND crl.mem_id <> c.member_id
          ORDER BY posted_date DESC, posted_time DESC
          LIMIT 1
        ) ;

It will give you consistent results and also be quite efficient, if you add an index on (complaint_id, posted_date, posted_time, mem_id)

Tested at SQLfiddle.

Given this sample data:

root@localhost:playground > select * from t;
+------+------+--------------+
| id   | name | phone_number |
+------+------+--------------+
|    1 | A    | 111111111111 |
|    2 | B    | 222222222222 |
|    3 | C    | 222222222222 |
|    4 | D    | 111111111111 |
|    5 | E    | 222222222222 |
+------+------+--------------+
5 rows in set (0.00 sec)

first you determine which rows you want to keep.

root@localhost:playground > select min(id), min(name), phone_number from t group by phone_number;
+---------+-----------+--------------+
| min(id) | min(name) | phone_number |
+---------+-----------+--------------+
|       1 | A         | 111111111111 |
|       2 | B         | 222222222222 |
+---------+-----------+--------------+

With this query you can check which rows will be deleted:

root@localhost:playground > select * from t left join (select min(id) id, min(name) name, phone_number from t group by phone_number) a on a.id = t.id and a.name = t.name and a.phone_number = t.phone_number;
+------+------+--------------+------+------+--------------+
| id   | name | phone_number | id   | name | phone_number |
+------+------+--------------+------+------+--------------+
|    1 | A    | 111111111111 |    1 | A    | 111111111111 |
|    2 | B    | 222222222222 |    2 | B    | 222222222222 |
|    3 | C    | 222222222222 | NULL | NULL | NULL         |
|    4 | D    | 111111111111 | NULL | NULL | NULL         |
|    5 | E    | 222222222222 | NULL | NULL | NULL         |
+------+------+--------------+------+------+--------------+
5 rows in set (0.00 sec)

The delete query:

root@localhost:playground > delete t.* from t left join (select min(id) id, min(name) name, phone_number from t group by phone_number) a on a.id = t.id and a.name = t.name and a.phone_number = t.phone_number where a.phone_number is null;
Query OK, 3 rows affected (0.00 sec)

root@localhost:playground > select * from t;
+------+------+--------------+
| id   | name | phone_number |
+------+------+--------------+
|    1 | A    | 111111111111 |
|    2 | B    | 222222222222 |
+------+------+--------------+
2 rows in set (0.00 sec)

Another feasible approach would be this:

CREATE TABLE t_tmp LIKE t;

INSERT INTO t_tmp 
SELECT MIN(id), MIN(name), phone_number
FROM t
GROUP BY phone_number;

RENAME TABLE t TO t_backup, t_tmp TO t;