Talk:Slowly changing dimension

From Wikipedia, the free encyclopedia

I don't believe that SCD type six really exists and it is not because what the article is describing is incorrect. In fact the example described in the SCD type 6 is perfectly valid however I do not believe that there is a case where you would need this and it is not a result of lower or different granularity events being aggregated and merged. This is in effect something that can occur quite often and it has to be handled very carefully.

Imagine we have the example that the article is describing; a supply chain process that goes something like order->shipment->delivery. In this case, and based on what the article describes, the order, shipment and delivery can each in effect record different details of the supplier at the time each event occurs. We are in effect looking at three different facts with the same descriptive dimensions but each can have their own dimension values in time.

A better way to address this dimensional modeling problem would be:

Lets say that we use SCD type 2 for the supplier dimension:

Supplier_key Supplier_Code Supplier_Name Supplier_State version

001 ABC Phlogistical Supply Company CA 0

002 ABC Phlogistical Supply Company IL 1

We would have 3 conceptual facts:

Order Fact – Linked to one set of supplier dimension keys Shipment Fact - Linked to another set of supplier dimension keys Delivery Fact - Linked to a final set of supplier dimension keys

At least conceptually, regardless of the implementation, there are 3 different facts. If we wanted to merge these three facts, that are measuring different points of the same supply-chain process, we have to consider the different (but shared) dimensions to become roles.

The Order-Shipment-Delivery Fact would include 3 different roles for the supplier dimension: Order Supplier, Shipment Supplier and Delivery Supplier. In this way the querying becomes extremely easy and flexible. There would be no need to ask the user to use timestamp data or to implement the foreign keys using business keys.

This article just seems to be a bit of a mish-mash of different editor's theories and terminologies. As an encyclopedia article it should only be discussing reliable notable research, eg. Kimball. Anything non-attributed to such research does not belong here. Read WP:NOR to see why this is the case. Remy B 08:12, 7 November 2007 (UTC)

Some additional citations (e.g. Kimball, Haughey, Kalido) have been added to ensure this is not considered as "original research" so hopefully Remy B's criticism is no longer valid - AJD, Feb 08