Principles of Sequencing and Scheduling, 2nd Edition

18Project Analytics

18.1 Introduction

The methods of analytics constitute the main bridge from theory to practice. Those methods analyze data to reveal facts that must be addressed by any practicable theory. If such facts are congruent with the theory, we say the theory is valid, and we expect valid theory to guide us toward successful implementation. In our context, a project scheduling framework that does not involve validation is simply not credible. Classical PERT assumptions, including statistical independence and reliance on the beta distribution, are examples of what we now recognize as invalid theory. Project analytics replace the flawed PERT assumptions and support project management in general. In this chapter, we argue that the beta distribution should be replaced by the lognormal, and the independence assumption should be replaced by a model of dependence such as linear association. To fully justify these claims, however, we must show how to resolve various implementation challenges.

Unfortunately, the bulk of stochastic scheduling research has focused on mathematically convenient distributions that have rarely been validated. That is, with few exceptions, no attempt has been made to prove that the stochastic models fit actual observations. Similarly, the vast majority of that research also relies on the assumption that processing times are statistically independent. We discuss evidence that the lognormal distribution is often valid but that estimates may be subject ...