Improving predictability in effort estimation by task size and productivity measurement

Abstract

The accurate estimation of project development time is one of the most important keys for the success of a project. Many studies use some method to enhance both estimation accuracy, such as Function Point (FP) and the estimation ability of the engineers, like Personal Software Process (PSP). While those studies focused on accuracy of the initial estimation, the current thesis considers the re-estimating process and shows how effort should be estimated and measured, and why. The study began as a studio project involving students in a Master of Software Engineering (MSE) program. Participants would use Team Software Process (TSP) as a managerial process and receive informal Personal Software Process (PSP) training. During the project, all team members estimated the amount of effort required for their own tasks and then collected actual effort data. Despite continuous adjustments using estimated and actual effort data, estimation did not become more accurate as time went on. In this thesis, it was determined that the problem arose because effort estimation was not treated as having two components: task size and individual productivity. By measuring task size and productivity independently, the tendency between estimated and actual data gets clearer and re-estimation of effort is more successful. Furthermore, the specific reasons for inaccurate estimation can be identified in detail and controlled. Thus it becomes easier to find relevant solutions from related studies for enhancing estimation accuracy.