Combination of expert opinion is frequently used to produce estimates in software projects, and various combination techniques are advocated in software engineering textbooks and courses. However, empirical evidence as to how the various methods perform is rarely offered. This is surprising, since there exists plenty of evidence from other research communities detailing the properties of group processes. For example, a common finding is that group discussion increases individual biases.
This talk will provide a summary of empirical studies on combination of software estimates. It will also draw on research from other areas, mainly psychology. In short, you will learn if, when and how to combine estimates in order to increase prediction accuracy in software projects.