Probability, Formation, and Prediction of Large-Size Construction Cost Overruns Governed by a Power-Law Distribution

Abstract

Cost overruns persist in the current construction industry. Large-size cost overruns are especially detrimental to projects, but the literature has treated such overruns no differently from regular ones. This paper argues that strategies should be specifically developed to address large-size cost overruns because their occurrence is higher than perceived and their impact is more severe. To get a full picture of these events, the authors collected cost overrun data from 30,950 UK construction projects and plotted them into a probability distribution curve, which featured a bell-shaped major part representing regular cost overruns and a long right tail representing the large-size ones. Taking a bootstrap method based on the minimum Kolmogorov–Smirnov method, the authors demonstrated that the bell and the tail were governed by different laws. This suggested that large-size overruns should not be considered an extension of regular ones, and management strategies for large-size overruns should be specifically tailored. To this end, the authors further investigated 368 construction projects to understand the specific formation mechanism for large-size cost overruns, based on which a Bayesian classifier was established. The classifier was then tested by 431 real construction projects, showing the model achieved an accuracy rate of 97.2% and F1 score of 72.0% in practice. For general management strategies, improvements in plans and specifications and risk identification during project initiation and enhancement of communication throughout the project were suggested. For project-specific avoidance strategies, the Bayesian classifier could predict the occurrence of large-size cost overruns and simulate the effect of available improvements to assist in identifying the optimal avoidance strategy. This paper corrects the general perception of the severity of construction cost overruns, completes the understanding of the formation mechanism of large-size cost overruns, and provides a predictive model to help avoid large-scale cost overruns on a case-by-case basis.