Travel Time Reliability for Stockholm Roadways: Modeling the Mean Lateness Factor
J P Franklin, A Karlström, Royal Institute of Technology, SE
This paper assesses the empirical distribution of congested travel times in the Greater Stockholm Region across time of day and day of year for a variety of road segments, paying attention to implications for modeling travel choice under uncertainty.
A growing body of evidence from revealed and stated travel choices in controlled shows that travelers are substantially influenced by the uncertainty of travel times, not only their expected lengths. It thus is of great interest to transport planners to project traveler responses to policies that reduce uncertainty, and to project occasions where projected savings in average travel time might be negated by increases in travel time uncertainty.
A major challenge in doing this, however, is in measuring travelers? marginal rate of substitution between travel time and travel reliability, such that the findings are transferable to real choice situations. It is infeasible to accurately replicate the full distributions of travel times in stated choice situations such that they reflect the real travel time uncertainties that travelers will face when real policies are implemented. Yet it may not actually be necessary. A related project has shown that measured preferences are transferable to situations with different mean travel times and different standard-deviations of travel times, provided that the standardized travel time distribution remains the same. Indeed, only a certain portion of the right-hand tail of the standardized distribution need be the same for this to be true. This portion of the tail constitutes what can be referred to as the ?conditional average lateness?, or the expectation for additional delays beyond the traveler?s expected arrival time in the standardized distribution.
The question remains whether we can actually expect the conditional average lateness to be consistent in different contexts. This paper focuses particularly on that question. We start by characterizing the shape of the probability distribution for travel times and how it varies across different roadway segments, times of day, and days of the year. Using video camera observations of vehicle movements on 92 roadway segments in the Greater Stockholm region, we obtain travel time estimates for each 15-minute interval from early morning to late night. We employ non-parametric quantile regression to measure the variation in distributional shape. We then proceed to determine whether different time periods are significantly different in terms of average lateness. Finally, we classify roadway segments and major time periods (defined as those time periods that are separately represented in the regional travel modeling system) according to those with similar average lateness. The results of this classification help us determine what range of cases should be represented in stated choice experiments to prepare for most real settings, and to assess the limitations present when only a small number of state choice experiments are used to predict travel choice under uncertainty.
Association for European Transport