Understanding the Effects of Correlation in Modelling Construction Cost Uncertainty

Abstract

Estimating costs for construction projects is a complex and often uncertain process. Given the inherent uncertainty of estimating and the unique risk profile of each construction project, many owners use cost uncertainty analysis to understand a project’s range of potential costs. Since results of the analysis can determine whether a project proceeds or how a project’s budget is set, it is essential that model results represent the actual range of costs that exists for a project. Various studies of actual and estimated costs have found cost underestimation, which occurs when estimates fail to adequately anticipate potential cost overruns, to be a persistent issue. While the causes of cost underestimation are numerous, this thesis focuses on two contributing methodology deficiencies: the inaccurate representation of uncertainty in model inputs and failing to account for correlation between model inputs. To address the former deficiency, the research examines elicitation biases, explores leading elicitation protocols, and demonstrates the application of the Sheffield elicitation framework (SHELF) protocol to cost uncertainty analysis for a construction project. To address the latter deficiency, this research proposes a Monte Carlo simulation experiment to test the hypothesis that since cost uncertainty (measured by standard deviation) is affected by correlation between model inputs, modelling correlated inputs independently results in underestimation of cost uncertainty. The experiment involves generating correlated cost item data, collecting statistics and generating independent samples, applying three correlation modelling methods, and evaluating those methods relative to the correlated data under a variety of scenarios. Experimental results found that independent modelling yielded an underestimation of cost uncertainty in most cases and that correlation modelling methods generated less error on average. A case study is also presented for the application of correlation modelling methods to cost uncertainty analysis of a light rail transit (LRT) project.