Multidimensional Analysis Factors

One of the distinguishing factors of multidimensional OLAP tools, as opposed to conventional querying tools, is the way they present the information. Measures or facts are usually presented in a multidimensional format, such as columns in a fact table or cells in a cube. These columns and cells contain precalculated numeric data about a functional subject area and are related to business objects (dimension tables) associated with the subject area. For example, Sales Amount, Net Profit Amount, Product Cost, and Monthly Account Fee are numeric data (facts) precalculated by account, by purchase, by customer, and by geography, which are the associated business objects (dimensions). In contrast, a conventional relational table would be a flat matrix of rows and columns, containing numeric data about one and only one business object (dimension). For example, Opening Account Balance, Daily Account Balance, and Account Interest Rate are numeric data of only one business object, namely account.

Figure 12.1 illustrates multidimensionality through a four-dimensional cube with the dimensions customer, account, purchase, and geography. The two examples of geography in this figure are the regions northeast USA and southeast USA.

Figure 12.1. Four-Dimensional Data Representation

Customer profiling and customer profitability are popular multidimensional BI applications. The dimensions of a customer profitability example are listed in Figure 12.2.

Figure 12.2. Multidimensional Customer Profitability

The eight dimensions of customer type, buying behavior, credit rating, region, demographics , psychographics, purchasing history, and product category can be used to analyze the various perspectives of customer profitability.

Some additional examples of complex multidimensional analysis commonly performed in the BI decision-support environment include those listed below.

• Customer information: Buying patterns by product, geography, time, age, gender, number of children, types of cars owned, education level, or income level

• Financial planning: Business analysis on profit margins, costs of goods sold, tax codes, or currency exchange rates

• Marketing: Impact of promotions and marketing programs, pricing, competitors ' initiatives, and market trends

Multivariate Analysis

Another term for multidimensional analysis is multivariate analysis. This term is derived from a specific aspect of this type of analysis, namely, to analyze measures (facts) from the perspective of multiple variables or characteristics. These variables (characteristics) usually describe business objects or dimensions. For example, Product Type describes product and Customer Age describes customer, with product and customer being the business objects or dimensions. Occasionally, the variables can become dimensions in their own right. For example, the variables Product Type and Customer Age can be treated as dimensions. In other words, a dimension can be built for a business object or for a variable of that business object.

These two types of dimensions (the object dimension and the variable dimension) can be illustrated by a simplified example of earthquake analysis. Earthquakes are typically reported by their epicenter, such as the intersection of latitude and longitude coordinates of a location, and by their intensity, such as 7.5 on the Richter scale. Location is normally a business object; thus epicenter can be used as an object dimension, which may be described by variables such as Location Name , Location Address, and Population Size. Intensity, on the other hand, is normally not a business object but a variable that describes the business object earthquake. In this example, however, intensity is treated as an object in its own right and is therefore used as a variable dimension. Another example of a variable dimension might be Shock Type (foreshock, aftershock), which is normally also a variable of the business object earthquake.

Variable dimensions are often " degenerate " dimensions, which means that even though they are being treated as dimensions when precalculating or analyzing the facts, they are not implemented as physical dimension tables. The main reason is that variable dimensions usually do not have other variables describing them ”or they would not be variable dimensions in the first place. For example, intensity is simply a set of numerical values ( numbers on a Richter scale), and there are no other descriptive characteristics about it.