Subjective De-Biasing of Data Sets: A Bayesian Approachby M. Elisabeth Paté-Cornell, Stanford Univ, Stanford, United States,
Abstract: In this paper, we examine the relevance of data sets (for instance, of past incidents) for risk management decisions when there are reasons to believe that all types of incidents have not been reported at the same rate. Our objective is to infer from the data reports what actually happened in order to assess the potential benefits of different safety measures. We use a simple Bayesian model to correct (`de-bias') the data sets given the nonreport rates, which are assessed (subjectively) by experts and encoded as the probabilities of reports given different characteristics of the events of interest. We compute a probability distribution for the past number of events given the past number of reports. We illustrate the method by the cases of two data sets: incidents in anesthesia in Australia, and oil spills in the Gulf of Mexico. In the first case, the `debiasing' allows correcting for the fact that some types of incidents, such as technical malfunctions, are more likely to be reported when they occur than anesthetist mistakes. In the second case, we have to account for the fact that the rates of oil spill reports in different incident categories have increased over the years, perhaps at the same time as the rates of incidents themselves.
Subject Headings: Bayesian analysis | Data processing | Hazardous materials spills | Risk management | Probability | Information management | Probability distribution | Professional societies | Australia | Gulf of Mexico
Services: Buy this book/Buy this article
Return to search