Travel substitution after the World Trade Center attack: A time series analysis

Background: On 9/11/2001 about 2,750 persons perished in the WTC terror attack. Domestic air travel precipitously dropped in the wake of the attack and did not recover for several years. Prior research estimated that 1,500-2,170 additional highway fatalities resulted from travelers substituting land for air travel in the first two years after the attack. The travel substitution effect may have lasted longer and may have cost more lives. Data and methods: Unobserved component models (UCM) are fitted to U.S. monthly motor vehicle traffic death counts reported by FARS, 1/1994-12/2009 (n=192). The 2010 data are used for model validation. Four monthly predictor series are added to a basic structural model: (1) the civilian unemployment rate, (2) real average retail gasoline prices, (3) population weighted average precipitation amounts, and (4) population weighted average surface temperatures. The intervention variable is coded 0 prior to 10/01 and 1 for October. Subsequent months follow varying distributed lag functions (1/(1-d)) for which the decay parameter is set in the range {.7<=d <=1.0}. Final models meet white noise criteria for residuals and minimize AIC (within sample) and MAPE (out-of-sample) statistics, respectively. Models are estimated with SAS 9.3 software. Results: The travel substitution effect is estimated at ω=244 (t=3.05) with a decay rate of d=.94. The estimate implies 4,066 additional highway fatalities most of them occurring within 4 years after the attack. The model forecast error is 3.2%. Conclusion: Misperception of travel injury risks remains a public health concern.