Analysing Heterogeneity in the Value of Travel Time and Reliability: A Semiparametric Estimation Approach

Travel congestion is considered as the most important negative externality of urban agglomerations. Despite the recommendations of urban and transport researchers, congestion tolls are rarely applied, mainly because these policies are not generally accepted by a broad public. Recently, peak-avoidance rewarding experiments have shown to be a successful strategy to alleviate congestion. Because travellers are rewarded if they travel outside the peak, the policy is more accepted. This paper studies the preferences for reliable and speedy travel of participants based on such a rewarding experiment. More specifically, we explicitly analyse heterogeneity in the value of time and reliability using a novel semi-econometric procedure, based on a large-scale Dutch stated choice experiment. The contribution of this paper is then twofold. First, we derive the value of reliability in such a way that it takes into account the dynamic nature of travellers' choices. We derive an easy to apply closed-form cost function assuming that travellers choose their departure time optimally and travel times follow a log-normal distribution. The expected scheduling costs due to travel time variation can then easily be calculated if the preferences of arriving at the preferred arrival time at work are estimated. Second, it is well known that there is substantial heterogeneity in the value of time and reliability. In order to avoid misspecification due to restrictive functional form assumptions, but more importantly, to analyse the importance of observed and unobserved heterogeneity in the value of time and arriving at the preferred arrival time at work, we use a new flexible semiparametric estimation approach. We propose a so-called semiparametric local latent class model, which enables us to distinguish between observed and unobserved heterogeneity. The econometric model results in a unique nonparametric distribution of preferences for each unique combination of observables. We find that there is substantial heterogeneity in the willingness to pay parameters. It is shown that only about 12.5 percent of the heterogeneity is attributable to observed characteristics of individuals, while the remainder is attributed to unobserved characteristics. This stresses the importance of controlling for unobserved heterogeneity. In line with expectations, it is also shown that income and educational level are strongly positively correlated to the value of time and reliability.