Three methods for quantifying software development effort uncertainty

Abstract

Abstract Software development effort estimates have several major sources of uncertainty. Among these uncertainties are the size of the project, the development attribute ratings, and the error of the estimation model. This paper presents three methods which quantify the effects of these uncertainties on development effort estimates. One method takes advantage of the invertibility of the nonlinear effort models to approximate the effort probability distribution. In the case of a single software configuration item, this methods yields the exact probability distribution. A second method uses Taylor series to estimate mean and variance of effort, and then specifies its probability distribution by invoking the Central Limit Theorem. The third method, specific to the Constructive Cost Model (COCOMO), invokes a Monte Carlo simulation technique to approximate the effort probability distribution. The results of case studies based on the COCOMO model are presented and compared. The mathematical details are provide...