Dimensions of a measure

Dimensions are points of view from where a user can look at the measures stored in the cube. They form the structure of the multidimensional model. A measure without dimensions does not make sense. This can be compared with a record with numbers having no corresponding primary key attributes or secondary explanatory data attributes. It is important, as indicated by the title of this blog, to allocate measures to a dimension, because not all measures do hold sensible values for a dimension.

Derived measures as category dimension

Derived measures, which are composed of more than one data attribute, can be relatively simple allocated to a dimension by adding the dimensions of each base measure. The measure Price of a product is dimensioned only by product. The location does here not make sense. Across all cities in the United States, the same price is calculated.

One major exception

Customer dimensions the measure Quantity ordered. If we make now a derived measure Sales value we can simply add the Customer dimension by the Product dimension and we get the dimensions that make sense to the measure Sales value. However, there is one major exception. Imagine the construction of a derived measure Total revenue which is based upon two measures: Other revenue and Traffic revenue. The measure Other revenue is part of entity x and the measure Traffic revenue is part of entity y. Those entities have a one-to-many relation (one instance of entity x has many instances in entity y). The union of the dimensions of the two base measures cannot dimension the derived measure Total revenue.

A natural way to navigate

The combined steering model is top-down oriented. This provides the user with a natural way to navigate through the performance management information. The user can drill-down from mission to objective and from objective to measure. For the first part of the model, strategies and objectives are often dimensioned by a relatively small number of dimensions. Once the user arrives at the level of measures, a problem can arise. Measures can be dimensioned by a relatively large number of dimensions compared to objectives.

Drill-down operations

When the user performs a drill-down operation from a key performance indicator to a lower level, which dimension and measure should be displayed first? The procedure below might be used to derive which measure and dimensions should then be displayed:

• if there is one, always show the time dimension first;
• organize the other dimensions according to how remarkable they are; the dimensions that reveal the most variation or deviation should be on top or easiest to slice-and-dice.

Ordering of dimensions and measures

Ordering the dimensions and measures according to how remarkable they are might ease the navigation actions the manager has to undertake in order to detect deviations from normal situations. In addition, it might reduce the time spent in search of useful information.