It seems like a simple question. With enough safety observations, can a safety professional accurately predict a future accident or injury?
The answer is yes-you can do so with an amazingly high level of accuracy. In this paper we will describe how we came to this conclusion through the development of a site scoring method that accurately identifies sites where workers are most at risk based on safety observations alone. This is not intended to be a technical paper. Rather, it is an informal description of our research, the challenges we have overcome and our results to date.
This question is most important for organizations that have many employees, decentralized management systems and multiple sites that exhibit a vast range of unsafe behaviors and conditions. In addition, those most interested in solving this question will likely deal with the diverse traits of the observers collecting this information, including but not limited to, bias, varying perceptions and a wide range of safety knowledge or competencies.
It is obvious that if we increase the time qualified people have to focus on error-prone situations or high-risk work areas, we are more likely to avoid human error, accidents and injuries. Just as an accurate weather forecast enables a farmer to plan and prepare for imminent foul weather, a safety professional benefits from as much time as possible to educate, coach and provide resources for avoiding catastrophic loss.
Before tackling this question, we had to address three big challenges. The first was how to get a large enough data set to ensure the reliability of information. We then needed to determine which variables would be most indicative of risk. Finally, we were concerned with the quality of an observation and specifically, how we could maximize the objectivity of our information.
For our research, twelve different customers shared their inspection, observation1 and loss information from 1424 different sites (with each site having at least five inspections). Over ten million observations were collected, of which about one million were unsafe observations. On average, less than 15% of the sites had any accidents or injuries to report. From a data collection point of view, the observation lists were not exactly the same but all contained elements of unsafe behaviors and conditions (see Exhibit 1 for a sample list). Because the observations were collected electronically on site using DBO2 SafetyNet,2 this enabled convenient access and analysis for conducting research.
