Identify Outliers, Understand the Process

Identifying atypical performance in a software process or atypical entities in software data is important for statistically analyzing processes and products and for statistical process control (SPC) of the software process. In this research, data from the Personal Software Process (PSP) was analyzed using two different techniques for identifying atypical observations. It was found that simple techniques such as interquartile limits are approximately as effective as XmR control charts for identifying outliers in the absence of causal analysis. When outliers are appropriately removed from a process, the remaining data characterizes the “commoncause” system of typical implementation. The review rate in the PSP common-cause system was found to be faster than the recommended 100 to 200 lines of code per hour. Following recommended practice is a precursor to SPC.