A Statistical Estimate of Total Annual Hazardous Material Incident Costs

1
A Statistical Estimate of Total Annual Hazardous
Material Incident Costs

• Steven J. Naber, Ph.D.

2
Acknowledgements and Disclaimer

• Co-authors
• Ron DiGregorio, PHMSA
• Mark Lepofsky, Battelle
• Bennett Pierce, Battelle
• The presentation is a Battelle presentation of a
  peer-reviewed paper authored by Steven Naber, et
  al., from Battelle. The paper was based on
  research conducted under contract for PHMSA to
  develop a statistical estimate of the total
  annual hazardous material incidents costs, based
  on updated cost information on reported incidents
  obtained from the reporting carriers and other
  third parties.

3
Problem Statement

• HMIRS database consists of information regarding
  incidents involving hazardous material
• Data are reported by carriers involved in
  incidents using a standardized form
• Questions have arisen about the potential for
  incomplete and inconsistent reporting
• Some incident-related costs are not borne by
  carriers and, therefore, are not reported
• Regulated community suspects under-reporting of
  incident costs
• Consequently, the credibility of the cost
  information may be compromised

4
Proposed Solution

• Statistical approach
• Random selection of incidents for data validation
  and revision
• Obtain revised cost information from carriers and
  other parties (shippers, responders, clean-up
  contractors)
• Statistical modeling to estimate incident costs

5
Methodology

• Stratified sample selection
• Stratified by mode of transportation
• Separate stratum for high-cost incidents
• Total of 500 incidents sampled between 11/1/02
  and 10/31/03
• Telephone interviews with carriers
• Obtain costs for 5 categories
• Contact third-party sources, where necessary
• Statistical modeling
• Stratified regression estimation
• Model costs as function of incident
  characteristics from HMIRS
• Examined single model and separate models for
  each cost component
• Final model used different predictors for each
  stratum and each cost component

6
Summary of Data
7
Significant Predictors of Cost
8
Results Cost (in Dollars) by Component and
Stratum
9
Comparison to HMIRS Database
10
Conclusions

• Final statistical estimate was 77.7 M 11.1
  M, with reported costs of 49.8 M, indicating
  that HMIRS does under-report the true costs (by
  an average of 36)
• Stratified regression using separate models for
  each stratum and cost component performed well
• Most explanatory variables make sense

11
Current Research

• Verify incident characteristic data in addition
  to cost data
• Applicability of these models beyond the data
  used in the study
• Addition of predictor variables that were left
  out of the modeling.