This problem arose from an analysis services (SSAS 2005) course I instructed last week. Since I am probably under NDA I can only reference the MSDN article which has the same example:

http://msdn2.microsoft.com/en-us/library/ms166553.aspx

In this example, we set up a dimension hierarchy and specify some attribute relationships. The problem surfaces if you try to use the dimension with a measure group - e.g Internet Sales - Total Sales.

The key point is that the subtotals for regions, countries etc are not the same before and after applying the relationships.

Moreover, the totals are correct before applying the relationships, and incorrect after (we checked this by executing SQL queries on the underlying datawarehouse).

Specifically, the problem occurs with customers who have the same name (e.g Jaime Hernandez)

There are 2 Jaime Hernandez - one in South Australia, one in Berkeley CA.

After applying the FullName- Postal Code attribute relationship, the Berkeley Sales figures go down, and up in the South Australia totals.

There are a couple of points - first, the relationships do not need to be applied in this example. The data is correct without them.

Secondly, be VERY careful with applying relationships. It can completely screw your data. I would imagine that if this surfaced to an end-user your datawarehouse project might come to an untimely end.