Presentation

· Contributors · Organizations · Search Program · Flagged · Happening Now

False Confidence and Time-Series Statistics in Oil Spill Injury Assessments

DescriptionTime-series statistics, specifically time-series regression approaches, have been used by Natural Resource Damage Assessment (NRDA) practitioners to determine whether an injury to a natural resource has occurred and, when an injury is thought to have occurred, to quantify the magnitude of the injury. This study evaluates these models using case studies from several recent oil spill injury assessments that used time-series regression methods to evaluate injuries to marine mammals and outdoor recreation services. First, we summarize the general structure and assumptions underlying the time-series models typically used in oil spill injury assessments. Second, we investigate uncertainty using prediction errors derived by applying the modeling procedure to baseline periods in the data (i.e., times when there is no effects of an oil spill). Our proposed approach quantifies the uncertainty arising from the combination of natural variation in the data and uncertainty associated with the modeling process. Compared to other statistical methods commonly used to quantify uncertainty associated with regression models, our approach provides a more robust estimate that better matches the informational needs of an oil spill injury assessment. Applying our approach to the case studies, we find that these time-series regression models can have relatively large prediction errors that limit the ability of analysts to reach firm conclusions regarding injury. Failure to appropriately quantify uncertainty can therefore lead to false confidence. Finally, we discuss how incorporating additional information can reduce uncertainty associated with the use of time-series modeling in this context, leading to higher confidence associated with oil spill injury assessments.

Authors

Heath Byrd

Lead Environmental Scientist - Economics

SWCA