When it is a rank, a probability, or a frequency
In the source system we may find customer ranking, with value = 1,2,3, and so on. This is the customer rank based on their profitability. This rank is not a measure. It is an attribute. So accordingly it should be put in a dimension table, not in a fact table.
A source system may contain a number which is a probability. For example, in credit risk business, we have probability of default (PD). This is not a measure, but an attribute. It is not sumable. And it is not aggregatable. If we have a country with 10 issuers, what’s the PD of that country? You can’t average it. A measure is a measurement from a business event. An attribute is a permanent property of that entity.
If you use Moody’s Analytics (http://www.moodysanalytics.com/), you will come across EDF (https://www.creditedge.com/). EDF is Expected Default Frequency, i.e. how many times we expect the issuer to default in 1 year. If we expect that an issuer might default in the next 5 years with 0.1% chance, then the EDF is 0.02%. An EDF is arguably an attribute. Though many people think it is a measure. Generally speaking a frequency is an attribute. If a machine is expected to breakdown 4 times a year, this frequency number (4) is a property of that machine, hence it’s an attribute. If on average London has a 30% chance of raining on any day, this 30% is the property of London city. It’s an attribute, not a measure. It does not originate from an event.
As usual I welcome your comments.
Vincent Rainardi, 28/1/2012