Ascertainment of the archetype statistical method for incident rate forecasting through forecast performance evaluations

The potential benefits of a safety program are generally, only realized after an incident has occurred. Resource allocation in an organization's safety program has the imperative task of balancing costs and often unrealized benefits. Management can be wary to allocate additional resources to a safety program because it is difficult to estimate the return on investment, especially since the returns are a set of negative outcomes not manifested. One way that safety professionals can provide an estimate of potential return on investment is to forecast how the organizations incident rate can be affected by implementing the different resource allocation strategies, and what the expectation for the incident rate would have been without intervention. Safety professionals often trend the performance of their organization's safety program by benchmarking incident rates against other organizations. Previous studies have employed different statistical forecasting methods to predict how incident rates will react to changes in resource allocation. This paper analyzes the performance of four statistical forecasting methods employed in previous resource allocation studies along another statistical forecasting method, never before used for incident rate prediction, to ascertain the method that provides the highest level of forecast accuracy. By identifying the most accurate forecasting method, the uncertainty of which method a safety professional should utilize for incident rate prediction is reduced. Incident data from the Mine Safety and Health Administration (MSHA) Part 50, was used to forecast both short and long term incident rates. The performance of each of these forecasting methods were evaluated against one another to determine which method has the highest level of accuracy, lowest bias, and best complexityadjusted goodness-of-fit metrics. Evaluation of the performance provides indications that the double exponential smoothing statistical forecasting method can provide the most accurate incident rate predictions. Analysis of forecast bias indicated that the error for the double exponential smoothing method is unbiased, within the acceptable range for tracking signal, and had a level of prediction accuracy above 70%. The results of this observational study indicate that the double exponential smoothing method should be the method to consider for incident rate prediction. Consistent use of the same forecasting methodology amongst safety professionals as part of their safety program's resource allocation process, will allow for more consistent benchmarking of incident rate prediction.