DRMS: Improving Public Response to Weather Warnings

University Of Washington, Seattle WA

Investigators

Abstract

Despite improvements in weather forecasts both in terms of timeliness and accuracy, weather-related injury and death remain a serious problem. There is growing consensus that at least part of the problem is public distrust in the warnings themselves. This project investigates three psychological issues related to public distrust in weather warnings. The first has to do with inconsistency. Forecasts for high-impact weather events are often made days in advance to allow residents time to prepare. Subsequent forecasts for the same event may differ from earlier forecasts, giving the impression of inconsistency. Forecasters tend to assume that people distrust inconsistent forecasts and are reluctant to change forecasts even when better information becomes available, preferring to sacrifice accuracy for consistency. This project tests whether inconsistency or inaccuracy is more injurious to trust. Distrust may also arise from the fact that severe weather events are usually presented as certain because forecasters worry that admitting uncertainty signals incompetence. Most people, however, understand that weather prediction involves some level of uncertainty. Therefore too much certainty may seem implausible. This project tests whether adding an uncertainty estimate (e.g. the probability of a tornedo at your location) increases or decreases trust and compliance with warnings. Finally, distrust in warning forecasts may lead to delaying precautionary action in order to gather more information. If people wait too long, they may not have enough time to adequately protect themselves before severe weather hits. This project will determine the appropriate information to include in weather warnings to inspire trust and allow people to make timely decisions. Thus, the results of this project will influence forecast communication practices to provide people with better and more trustworthy information upon which to base critical weather related decisions, ultimately saving lives. The value of accurate weather warnings with generous lead times will be realized only if the public trusts them and acts accordingly. This research will inform "best practices" in warning communication procedures to promote both trust in the forecast and timely responses. As such, results of this work may well improve compliance with warnings and save lives. The research team investigates these issues in experimental studies, using realistic weather-related decision tasks. Establishing the hypothesized effects on trust and decisions in a controlled laboratory environment will permit testing forecast communication methods to address them. Of particular interest is the inclusion of specific uncertainty estimates. Conveying the notion that the forecast was intended as probabilistic may reduce effects of both inconsistency and inaccuracy. Uncertainty estimates may also attenuate delay beyond optimal stopping by satisfying the need for additional information. Likelihood communication, however, may need to be simplified in dynamic, time pressured situations so color coded risk scales will also be tested. Thus, these experiments will compare identical situations in which people receive either probabilistic forecasts, color-codes or conventional warnings, to determine which method leads to greater trust and better decisions. In sum, the goal is to carve out answers to a set of specific but critical questions, using careful experimental procedures that make direct comparisons between one situation and another and one form of communication and another, holding all other factors constant to yield firm conclusions about their effects. It is a cognitive-experimental approach to what has heretofore been a problem tackled primarily with other social science tools. This work will benefit the scientific community at large by providing a unique theoretical understanding of the cognitive processes involved in interpreting, trusting and acting on complex and dynamic predictions, with implications in diverse domains.