Precursors of Construction Fatalities. II: Predictive Modeling and Empirical Validation

Fatalities continue to plague the construction industry. To address this ongoing concern, researchers have begun to develop and test proactive methods to improve construction safety, such as risk analysis, leading indicators, and predictive analytics. The present study aims to build upon these current methodologies by creating and testing the first objective precursor-analysis program for construction fatalities. Specifically, the following hypothesis was tested: The probability of fatal and disabling events can be predicted by a small number of precursors that can be identified prior to an incident. Testing this hypothesis involved obtaining case data using a precursor-analysis protocol described in the companion to this paper, using principal component analysis to reduce the dimensions of the data set, building a mathematical predictive model using generalized linear modeling, and testing the predictive validity of the model with independent validation cases. The results indicated that there are 16 principal precursors that, when organized into a generalized linear model, are able to predict the outcome of new cases far better than random (p<0.001). With further validation and testing, this new methodology can serve as the foundation for the first objective and valid precursor-analysis program for construction. Such a program will enable the construction industry to predict the potential for a high-impact event during construction operations and ultimately improve its safety performance. (C) 2017 American Society of Civil Engineers.