Conformed Dimensions

Hi,

There is great emphasis placed on ‘Conforming dimensions’ when designing a DW. I’m a little confused, as having conformed dimensions seems the most obvious thing do and I struggle to answer the question ‘Why would you not have conformed dimensions’. Most of the examples I see speak of Conformed Dimensions and how they can be usefully put to work. What I think I’m missing is an example of ‘Non Conformed Dimensions’ and their associated downsides.

It would be much appreciated if someone can offer a quick explanation/example ?

Regards

Philip

A couple of weeks ago I saw a data warehouse that had six date dimensions - one for each fact table.

When I asked why, I found that each star had been developed over time, for different business functions, and each business area wanted to see different information in the date dimension.

Rather than doing it properly, which "may" have disrupted reports, the developers took the lazy way out.

I've also seen cases with multiple versions of product, customer and organisation dimensions. They represent the different perspectives that key users have on their data. It is wrong, even stupid, but the developers weren't strong enough to overcome the business pressure to do it that way.

What this leads to is the oft-stated situation where no two users can reconcile their reports, because the data in each report is categorised and aggregated on different rules.

It is much more common than you might expect, given its importance in dimensional modelling.

Thanks for that Ron,

Am I correct in saying then that given a fair wind and business co-operation there should be no reason NOT to have all dimensions as conformed dimensions? Also for further clarification, if I have 2 dimensions Company and Organisation say, are they conformed if they have the same set of attributes, with the same meaning and the same values even though the dimension names are different? Conversely if I have 2 dimensions called Company which have a different set of attributes, then they are non conformed dimensions?

Thanks again for your input

Regards

Your first example is a little tricky, because it might apply in the case of role-playing dimensions.

For example, imagine that we have created a Calendar dimension, which has dates, hierarchies, etc. We want to use this dimension in an Order fact, where it has three roles - Order Date, Ship Date and Invoice Date. We might implement this as three views on the Calendar dimension, it is still a conformed dimension, but it might appear to the user as three separate dimensions.

Your second example is a non-conformed dimension, as long as a Company is a Company. If you really meant "Supplier" and "Customer", and each had distinct attributes, then there would be an argument for keeping them separate, but naming them appropriately.

Conformance is a broad term. To define it as everyone sharing the same row in a table is a narrow view. Conformance is also about the same representation of attributes across dimensions.

Take the Date dimension as an example. Very often summaries are created for time periods, such as weeks, months and fiscal periods. It is a common practice to create period dimensions, such as a week dimension, to support such a summary. The week dimension is actually a subset of the date dimension with its own key. Week is considered conforming since the attribute values it holds are the same as those held by date. This allows you to combine date based facts with week based facts by summarizing the date based facts on common attributes held by both the date and week dimensions.

For further reading, there is a great chapter on Conformed Dimensions in Christopher Adamson's book, "Star Schema - The Complete Reference".

I went back to it last night, and found well presented examples of the points made by ngalemmo.

You're not missing anything. Siloed organizations building redundant data warehouses means nothing can be integrated. It's just common sense that requires extensive elucidation by Ralph because of all the dummies running IT organizations.

Thanks everyone for your replies - its helped to clarify things for me