Process-Driven Incremental Effort Estimation

Abstract

Effort estimation has shown its value in process decisions, such as feasibility analysis, resource allocation, risk mitigation, and project planning. In this paper, we propose an incremental effort estimation method that integrates phase-based effort estimation models to solve the concerns about software process compatibility and estimation accuracy, which one may often have when adopting effort estimation methods for project management. We define the process framework for incremental effort estimation in terms of the transition between the phase model that defines the analysis and design activities, the system models that specify a system's behavior and structure at different levels of detail, the sizing model that measures software functional size via transaction analysis, the Bayesian model that statistically models the effects that the size measurements have on project effort, and the phase-based effort estimation models that provide improved effort estimation accuracy over the targeted early phases. The phase-based effort estimation models are calibrated and evaluated based on an empirical dataset of 61 master-level student projects from USC CSSE. The evaluation results show their improvements in out-of-sample estimation accuracy, provide a perspective about how estimation accuracy evolves throughout a software process, and set the practical criteria to decide the investment in analysis and design activities for the return in estimation accuracy.