Travel time variation and its impact on the elasticity of route choice model parameters

Abstract

Travel time is an uncertain variable with inherited uncertainty for many reasons, such as measurement error, perception error, and temporal aggregation variation impacts. Many researchers have considered travel time as an exact value for different limitations, including models’ restrictions and being unaware of the travel time distribution. However, previous studies have shown that standard logit and probit models can result in biased estimations with uncertain variables or variables with errors (EIV). Nevertheless, limited evidence is available on the magnitude of this bias and the impact of the uncertain variable distribution on the estimated parameters. In this paper, we use a simulation-based approach to explore the uncertainty of variables in the models’ estimation. We check the difference between standard and simulation-assisted models for four previously observed distributions with uncertainty values in the range of common coefficient of variation (Cov) of travel time. Results show that the logit's coefficients and elasticity values with EIV consideration can be 63% and 70% different from those of a standard logit model. The probit coefficients and elasticity differ up to 75% and 124%, respectively, in a binary choice situation. Changing the data and models for a multinomial situation shows that logit has 60% biasedness again while probit undergoes up to 800%. A sensitivity analysis on elasticity values of CoV is done, which unveils that both models’ elasticities change according to a similar pattern but with different intensities. The results suggest that the standard logit handles uncertainty better than the probit model. Yet we cannot present any intuitive reason to support this observation or against it. Procedures like those implemented in this research are suggested to address the biasedness of the results.