Digital Safety Technology Adoption and Construction Accident Risk: A Multivariable Analysis Using Logistic and Negative Binomial Models

Abstract

Digital innovations increase visibility of and shorten the response time to dangers. However, they do not ensure safer outcomes. This paper analyzes a subset of a Ph.D. construction safety framework and uses appropriate models based on the outcomes to analyze accident frequency and occurrence. The analysis employs the fixed data (n=450, Chennai=250, Coimbatore=200) as reported in the author’s article. Measurement of the adoption of technology was done as a multi-item construct along with proxies of BIM, IoT, wearables, computer vision, mobile safety applications, digital permits, real-time monitoring, drones, RFID, proximity warnings, smart helmets and smart helmets and immersive training. The binary accident experience was modeled with logistic regression, and the accident count was modeled with Poisson and negative binomial regression. Technology adoption was found to decrease accident odds (OR=0.508, 95% CI 0.333–0.775, p=0.002). Safety training (OR=0.433) and PPE (OR=0.554) were protective, whereas financial constraints (OR=1.847) and scaffolding exposure (OR=1.735) were found to increase odds. The logistic model achieved AUC=0.785. For accident frequency, the negative binomial model was found to be better than Poisson (AIC 915.93 vs. 931.69) and protective associations of technology adoption (IRR=0.597), safety culture (IRR=0.647), training (IRR=0.668) and PPE (IRR=0.606) were found along with an increase in expected counts due to financial constraints, height exposure and hazard exposure. The findings indicate a socio-technical meaning, where “Digital tools used to provide visibility of and reduce response time to dangers must be coupled with governance, competence, maintenance and controls such as rapid human response and privacy.” The data are and meant to be presented as a modeling demonstration rather than field data.