6.3 Data Analysis Techniques

• Before starting with design measurement, be aware of your measurement goals. You do not measure for the sake of measurement, but to help plan and control the development process, and/or to improve the quality of the product. Design measurement can help here by supporting effort estimation, monitoring progress, and identifying the areas in your design where improvements are likely to have a high payoff.

• Review the set of design metrics for completeness - are there structural features of your designs you deem important that SDMetrics does not yet cover? For instance, you may want to include metrics to count certain stereotyped elements that have a special meaning in your development environment (e.g., stereotypes to mark variation points in a reference architecture of a product line). Use SDMetrics' flexibility to define new metrics.

• Use descriptive statistics (Section 6.3.1 "Descriptive Statistics") and techniques such as PCA (Section 6.3.2 "Dimensional Analysis") to identify a minimal, nonredundant set of metrics that is meaningful for your design practices.

• This reduced set of metrics can be used for class/package rankings to identify areas with potential design problems (Section 6.3.3 "Rankings").

• Over time, you can build up a design measurement database and establish quality benchmarks (Section 6.3.4 "Quality Benchmarks").

• In more mature organizations that regularly perform process measurement (fault tracking, effort data), you can use design measurement for quality predictions (Section 6.3.5 "Prediction Models").