Weihao Yin

Changes in Evacuation Decisions Between Hurricanes Ivan and Katrina

Hurricanes cause some of the worst traffic conditions, affecting evacuees’ ability to reach safety before they are subjected to high winds, heavy rain, and flooding. This paper is one of the few to use a panel survey to examine similar household decisions over consecutive hurricanes. The study focuses on Hurricanes Ivan and Katrina, which were of similar strength and followed similar paths, and is fairly comprehensive in the number of traffic-related decisions considered. Contingency tables, binary logit models, and Goodman and Kruskals gamma measure were used to examine the effects of previous decisions on (a) whether to evacuate, (b) day of departure, (c) destination type and location, (d) number of household vehicles taken, and (e) reason for route selection. Through the statistical analyses, it was discovered that (a) to a great extent, citizens made the same decision to evacuate or stay for Katrina as they did for Ivan, and higher incomes were not significant in changing that decision; (b) some evacuees departed earlier, but most evacuees departed on the last day possible; (c) most evacuees selected the same type of accommodations and made the same inside-the-county-or-parish or out-of-the-county-or-parish decisions in consecutive evacuations; (d) the number of household vehicles used in the evacuation did not decrease; and (e) route guidance as a selection criterion did not depend on previous evacuation experience.

Imputing Erroneous Data of Single-Station Loop Detectors for Nonincident Conditions: Comparison Between Temporal and Spatial Methods

Effective use of loop detector data for traffic management requires that errors be efficiently detected, diagnosed, and corrected. We present two new spatial approaches and compare them to state-of-the-art correction procedures for station flow estimation when detectors within that station malfunction in nonincident conditions. One new method exploits the relationship between individual detector flow and station flow using linear regression. The second incorporates lane use percentages through kernel regression. To comprehensively compare the procedures, systematic and random-error evaluations are conducted for two detector stations with distinct lane configurations. Lane configuration is important for spatial correction methods, which perform well under certain detector failure combinations. The random-error evaluation indicates that temporal correction performs better at all error levels and spatial approaches are inaccurate under light traffic conditions, especially when estimates are based on zero flow readings. When choosing a correction procedure, one should consider facility configurations, error types and magnitudes, and traffic conditions, and calibrate the method for location-specific characteristics.

Statistical Analysis of the Number of Household Vehicles Used for Hurricane Ivan Evacuation

The objectives of this paper are to identify the contributing factors to households’ choice of the number of vehicles used for evacuation and to develop predictive models of this choice that explicitly consider the constraint imposed by the number of vehicles owned by the household. This constraint is not accommodated by regular ordered response logit models. Data comes from a poststorm survey for Hurricane Ivan. Two models that are variants of the regular Poisson regression model are developed: a Poisson model with exposure and right-censored Poisson regression. The right-censored Poisson model is preferred due to its inherent capabilities, better fit to the data, and superior predictive power. The model and individual variable analyses indicate that households traveling longer distances or evacuating later are more likely to use fewer vehicles. Households with prior hurricane experience and pet owners are more likely to use a greater number of vehicles. Income and distance from the coast are insignificant in the multivariable models, although the individual effect of distance from the coast has a statistically significant bivariate relationship with vehicle usage choice based on the Pearson correlation measure. A method for using the right-censored Poisson model to produce the desired share of vehicle usage is also discussed for generating individual predictions for hurricane evacuation demand simulation.