Risk Perception and the Credibility of Hazard Warnings: Evidence from Tornado Warnings


Published on

Risk Perception and the Credibility of Hazard Warnings: Evidence from Tornado Warnings

Published in: Technology, Business
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Risk Perception and the Credibility of Hazard Warnings: Evidence from Tornado Warnings

  1. 1. Credibility of Hazard Warnings: Evidence from Tornadoes Kevin M. Simmons Austin College Fulbright Research Scholar International Centre for GeoHazards Oslo, Norway
  2. 2. Economic and Societal Impacts of Tornadoes By Kevin M. Simmons and Daniel Sutter © 2010, 296 pages in paperback ISBN: 978-1-878220-99-8 AVAILABLE NOVEMBER 2010 from the American Meteorological Society and the University of Chicago Press
  3. 9. Warnings Save Lives <ul><li>Fatality Rate in the 1920’s was 1.8 per million </li></ul><ul><li>Today it is estimated at .11 per million </li></ul>
  4. 10. Warning Errors <ul><li>The value of a weather forecast or warning in an expected value framework depends on the reliability of the information signal. </li></ul><ul><li>Two error probabilities are relevant, the probability a tornado occurs unwarned, and the probability a warning is issued when no tornado occurs. </li></ul><ul><li>The second probability corresponds to the False Alarm Ratio (FAR) that the NWS reports for tornadoes and other types of warnings. </li></ul>
  5. 11. NWS Metrics for Performance <ul><li>The National Weather Service has 3 metrics to evaluate the performance of tornado warnings. </li></ul><ul><ul><li>Lead Time </li></ul></ul><ul><ul><li>Probability of Detection </li></ul></ul><ul><ul><li>False Alarm Rate </li></ul></ul>
  6. 12. NWS Metrics for Performance <ul><li>Leadtime </li></ul><ul><ul><li>Time from warning to first touchdown </li></ul></ul><ul><li>Probability of Detection </li></ul><ul><ul><li>Warned Tornadoes / Verified Tornadoes </li></ul></ul><ul><li>False Alarm Rate </li></ul><ul><ul><li>1 – (Verified Tornadoes / Warnings) </li></ul></ul>
  7. 13. The False Alarm Effect <ul><li>A higher FAR should reduce the value of a tornado warning and might induce people to ignore the warning. </li></ul><ul><li>Yet evidence of a false alarm effect in the field has been elusive. </li></ul><ul><li>A recent study notes “Evidence for the cry-wolf effect in natural hazards research, however, has not been forthcoming.” (Barnes et al. 2007) </li></ul>
  8. 14. Is there a False Alarm Problem? <ul><li>A false alarm problem exists if a resident in the path of a tornado chooses not to respond to the warning. </li></ul>
  9. 15. In Econo Lingo: <ul><li>Utility of taking cover given an approaching tornado is less than the Utility of not taking cover, given an approaching tornado. </li></ul>
  10. 16. How Can This Happen?
  11. 17. Empirical Test <ul><li>Can we find a way to test whether or not increases in false alarms have an impact on casualties from tornadoes? </li></ul>
  12. 18. False Alarms and Casualties <ul><li>We calculate a recent, local FAR, use this as a control variable in our casualty regressions. If false alarms reduce warning response, this should translate into more casualties, everything else equal. </li></ul><ul><li>We have tried six different definitions of the FAR in our research. </li></ul>
  13. 19. Econometric Models <ul><li>The number of persons killed or injured in a tornado takes on nonnegative integer values with a large portion of zeros, and thus is count data. </li></ul><ul><li>We estimate Poisson models of casualties and test for overdispersion. Injuries are generally overdispersed, but not fatalities, so we estimate Negative Binomial models for injuries. </li></ul>
  14. 20. Data Set for Tornadoes <ul><li>Storm Prediction Center national tornado archive </li></ul><ul><li>NWS Tornado Warning Verification records </li></ul><ul><li>County warning areas of NWS forecast offices, Doppler radar installation dates </li></ul><ul><li>Census data at county level </li></ul>
  15. 21. Study Parameters <ul><li>Database includes all tornadoes in the U.S. between 1986 and 2004. </li></ul><ul><li>Approximately 21,000 tornadoes. </li></ul><ul><li>County level census data from the 1980, 1990 and 2000 census. </li></ul>
  16. 22. Control Variables in Tornado Casualty Analysis <ul><li>Tornado characteristics: Rating on the Fujita(F) scale, path length, time of day, month of year, day of week </li></ul><ul><li>Path characteristics: Economic and demographic variables for the counties struck by each tornado </li></ul><ul><li>Warning variables: Warning in effect, lead time, false alarm ratio </li></ul>
  17. 23. Results Injury Model <ul><li>Injuries </li></ul><ul><ul><li>A 1 standard deviation increase in the one year FAR increases injuries from 12.4 to 14.7% </li></ul></ul>
  18. 24. Results Fatality Model <ul><li>Fatalities </li></ul><ul><ul><li>A 1 standard deviation increase in the one year FAR increases fatalities from 12.7 to 18.4% </li></ul></ul>