DRU: What is a 'Better' Prediction System? Combining Statistical and Economic Metrics of Prediction Quality

点击翻译按钮获取中文摘要

What is a 'Better' Prediction System?Combining Statistical and Economic Metrics of Prediction QualityPIs: A. Small, III (PI), J. Evans, K. Keller, A. Kleit, A. ThompsonOrganizations: The Pennsylvania State University (lead), Howard UniversityThis research project concerns the ways that people use predictions and forecasts to make decisions. Almost everyone has had the experience of checking the weather forecast before deciding whether or not to pack an umbrella on an outing. More dramatically, emergency management officials in coastal regions who confront an on-coming hurricane will keep a close eye on the forecasts of hurricanes track and intensity, when deciding whether or not to order an evacuation of a large region. In these cases and in countless others, decision-makers rely on prediction systems, or forecasting models, to form ideas about what the future holds. Often, prediction systems take the form of computer simulation models that have been developed in advance by technical specialists. To develop a prediction system or forecasting model, a developer must have some idea of what it means for the model to do a 'good' job, or for one approach to work 'better' than another. Typically, the developers of prediction systems define quality in terms of statistical measures that are generic, in the sense that they have no necessary connection to the uses for which the system is being developed. For consumers of prediction systems, however, a 'good' prediction system is one that helps them to make better decisions, e.g., to avoid costly errors. A central goal of this project is to develop measures of quality for prediction systems that incorporate the needs and goals of users. These user-sensitive measures of quality provide a means for communicating priorities back to technical specialists, to help guide system development in genuinely useful directions. The project is led by a diverse team of economists, meteorologists (hurricanes, air pollution), geoscientists (climate thresholds), and statisticians. Theoretical work will complement applications in four areas of vital economic and social importance: the generation of electric power, hurricane evacuation decisions, protocols for issuing air-quality alerts, and thresholds in the climate system. Through interactions with stakeholders, the research will feed directly into decisions on how to improve weather and climate predictions, with potentially large economic benefits. The project involves graduate and undergraduate students at both Penn State and Howard University. Howard University is the leading HBCU (Historically Black Colleges and Universities) in the atmospheric sciences and has a nationally ranked graduate school. Collaboration between these two programs is providing a range of research experiences and advisor expertise to students from both schools.