Now we have discussed the upper region of the steering model I will take a look at the lower region of the model. The physical storage for the measures and dimensions are often called cubes. Cubes are three-dimensional. Many people are familiar with two dimensional spreadsheet software packages, which establish a co-ordinate system by using rows and columns.
Information needed for analysis
The cube-concept builds on this concept. A multi-dimensional model is a way to look at information needed for in-depth analyses and decision making. Outside the cube, we find the points of view, called dimensions. Inside the cube we see a space that is spanned by the dimensions called the storage capacity for measures.
Intersection of the dimensions
In the cube, at the point of intersection of the dimensions, we find a single cell. A cell represents a measure, usually a number, dimensioned by all relevant point of views. Measures can even be one single standard that represents an industry, country or world average. Those normative measures cannot be directly derived from the organizations’ mission statement. However, they are often related to measures that can be successfully linked to the mission of the organization.
Furthermore, not all measures will make sense across all dimensions. This can blow up the cube. To reduce the size of the cube, the OLAP cube creation facility has to deal with sparse data.
Sparsity
Sparsity of data is highest at the lowest level of a hierarchy. For instance, one customer buys only a few products while different customer groups might have bought from diverse product groups. If the user of the information system wants to retrieve a measure that shows a lot of sparsity the interface could decide to eliminate those products and customers that hold no sales figures.
Measures can be clustered in related groups
Finally, measures appear together in related groups that are often linked by a formula. Actual figures, budget figures and the difference between those are an example of this occurrence. The order within those groups can at first glance be derived from the formula. The second step should be to manually organize the base measures to make sure that the sequence corresponds with the wishes of the user. Thus, measures should be enriched with a property that indicates at which order the measures should appear on the screen.
Types of measures
Base measures are representing raw data and can be directly retrieved from the On-Line Transaction Processing (OLTP) database or the data warehouse. Information in the data warehouse can in turn be aggregated and be made up of several attributes of the OLTP system. Derived measures reveal variations in four ways and can be discriminated by their expressions. A derived measure can be a:
- value that is aggregated from the OLTP database concerning one attribute;
- value that is calculated from more than one attribute for example the invoice amount divided by the quantity;
- combination of the prior derived measures (summarized as well as calculated);
- standard or norm (derivation takes place by averaging industries, branches and countries figures).
For all types of measures, the user should be able to retrieve the formula which the value is based on. One can imagine that novice users will see automatically the formula when a particular measure is retrieved. Experienced users should undertake explicit action to retrieve the formula.
Kind of measures
Measures have another significant feature in common that should be discussed because it is very crucial for many user interface design issues:
- normal measures (sales in a particular period);
- flow measures (off and by);
- stock measures (old stock previous period + by – off);
- ratio’s and percentages.
Normal measures might be added over time as well as flow measures. Stock measures cannot be added over time. Those measures must be averaged or the closing position must be taken for the underlying period. In addition, not every actual measure can be allocated to those three kinds, but sometimes it will make only sense to one of them. The measure Subscriptions can be both allocated to the stock (for example 600.000 subscribers) and to the flow of the measure (for example: we get 300 new subscribers and 500 subscribers cancel their subscription). The kind of the measure only makes sense to time-related dimensions. All other dimensions do not have this distinction.
When the two (off and by) flow measures as well as the belonging stock measure exist, the user of the system should be able drill-down from the stock measure to the flow measures. Thus, drill-down operations should be available only if both measures are available. Likewise, normal and flow measures can be added up to a total where stock measure cannot. Thus, add operations should be disabled at stock measures.
Furthermore, each kind of measure has its own specific best matching graphical presentations. Normal measures can be presented in all type of graphs. In addition, stock measures cannot be presented in stacked bars since they cannot be added up to a total. Certain stocks like supplies can better be showed in a vertical bar diagram instead of a horizontal bar diagram since supplies are expressed and perceived in height and not in length.
The appearance of the measure
Whether measures are base or derived, we distinguish measures based on their appearance. As mentioned before, measures are mostly representing quantities or numbers, but this does not always hold true because sometimes it is possible that someone is only interested in qualified measures. Qualified measures are special derived measures in a sense that a concrete number is converted to some textual indicator. Quantitative measures can be converted to their textual counterparts. In turn, the textual presentation can be converted to a visual presentation by applying rules or some other kind of intelligence (Roth, 1997; Mackinlay, 1986).
In one of the next blogs, I provide a detailed overview of which elements of the data might be used to make a proper graphical counterpart.
The appearance of the measure should not be confused with derived measures, which only calculate a target value. The appearance of the measure is the form of the measure which will be presented on the screen. Quantifiable measures might be formatted in several ways:
- by showing the exact number;
- by showing a percentage;
- by showing a number, but then using a scale (for example all occurrences of a measure in the cross-table must be multiplied by a million);
- by using some graphical component like a graph (pie, bar, line) which will show its value more clearly related to the other values;
- By converting it into a qualified textual measure.
Standard measures
OLAP tools have made it possible to compose multi-dimensional questions in an easy manner and with the guarantee of fast response times. Such multi-dimensional questions comprise variance analyses and time series (Buytendijk, 1996a). Senior managers might, for example, compare the previous year with how they are now performing. These comparative analyses can be classified into two groups:
- variances;
- time series.
Variances can be used as a method to employ progressive disclosure in the interface. Novice users are provided with a user interface that shows, at first, only the variance. When they learn more about the Business Intelligence system, the interface can expose the underlying base measures by enabling, for instance, drill-down operations.
Not all type of time series are necessarily part of the steering model since managers are possibly not interested in all types of time series-measures. Nevertheless, they can be very useful in determining whether a significant deviation exists or not. Functions like prior year to date and last month are excellent methods to decide whether a measure should be placed more outstanding on the screen. Although such functions are not part of the user interface they can be exploited by the user interface to enable more effective communication.
No comments:
Post a Comment