Exposing Useful Trends in Metric Data Through Group Level Analysis

In this paper the results of experiments which applied both structure and code metrics to three large scale systems are presented. This metric research is distinct in that trends in the data are uncovered through the use of group level analysis. Components are partitioned into groups based on their various metric values and on observed measures of complexity (ie. errors, coding time). Crosstabulation data is given which indicates that trends between some of the metrics and the observed data do exist. Code metrics typically formed groups of increasing complexity which corresponded to increases in the mean values of the observed data. The strength of the Information Flow metric and the Invocation measure is their ability to form a group containing highly complex components which was found to be populated by outliers in the observed data.