Investments in rail infrastructure are often motivated by the need to reduce travel time and to reduce travel time unreliability. Knowing travellers' valuation of travel time unreliability relative to in-vehicle time and travel cost is hence important for cost-benefit analysis of rail investments. For scheduled services with relatively high reliability and long headways, the most common approach seems to be the "expected delay" approach, where passengers are assumed to value a possible delay proportional to its expected value. Using three different data sets, it was investigated whether the expected delay approach holds empirically. How the valuation of a possible delay with probability p and length L depends on p and L was studied. The main result is that this valuation is not proportional to the expected delay, but increases slower than linear in the delay probability. This is particularly pronounced for small risks. This means that estimated "values of delay time" will depend on the delay risk level p: "delay time values" will be higher the lower the risk levels in the study are. This means that estimated values of delay times which do not take the non-linearity in delay risk into account will result in valuations that cannot be transferred between contexts with different delay risks. A theoretical reasonis provided why this should be an expected phenomenon, as long as headways are large. Regarding the dependence on delay length L, the valuation increases slower than linearly in L in about half of the cases, whereas in the remaining cases, the linearity hypothesis (the valuation is proportional to delay length) cannot be rejected. Contrary to what might be expected,the valuation never increases faster than linearly in delay length. Accounting for random population heterogeneity is shown to be important. In particular, the valuation of delays is shown to have a much larger variationacross the population than the valuation of travel time or travel cost. For the covering abstract see ITRD E145999
Abstract