This paper provides an overview of some alternative conceptual definitions of travel time variability, discusses their implications about behaviour, and puts them into a broader context, including deviations from the underlying assumptions regarding rational behaviour. The paper then discusses the empirical basis for assigning a value to travel time variability. This discussion leads to the conclusion that a fair amount of scepticism is appropriate regarding stated preference data and that attention should turn to the possibilities that are emerging for using large revealed preference datasets. The bottom line is that travel time variability is quantitatively important and cost-benefit analysis should account for it, using the best values we can get, in order not to imply a bias towards project that do not reduce travel time variability. Omitting the cost of travel time variability is not the neutral option.
