Shortest path or anchor-based route choice: a large-scale empirical analysis of minicab routing in London

Understanding and modelling route choice behaviour is central to predicting the formation and propagation of urban road congestion. Yet within conventional literature disagreements persist around the nature of route choice behaviour, and how it should be modelled. In this paper, both the shortest path and anchor-based perspectives on route choice behaviour are explored through an empirical analysis of nearly 700,000 minicab routes across London, United Kingdom. In the first set of analyses, the degree of similarity between observed routes and possible shortest paths is established. Shortest paths demonstrate poor performance in predicting both observed route choice and characteristics. The second stage of analysis explores the influence of specific urban features, named anchors, in route choice. These analyses show that certain features attract more route choices than would be expected were individuals choosing route based on cost minimisation alone. Instead, the results indicate that major urban features form the basis of route choice planning – being selected disproportionately more often, and causing asymmetry in route choice volumes by direction of travel. At a finer scale, decisions made at minor road features are furthermore demonstrated to influence routing patterns. The results indicate a need to revisit the basis of how routes are modelled, shifting from the shortest path perspective to a mechanism structured around urban features. In concluding, the main trends are synthesised within an initial framework for route choice modelling, and presents potential extensions of this research.     

Comparing distance,time, and metabolic energy cost functions for walking accessibility in infrastructure-poor regions

Accessibility is a widely used concept in transportation planning and research. However a majority of the literature is concerned with accessibility in infrastructure-rich regions where it is used to assess the output of infrastructure. Relatively scant attention in contrast has been paid to the topic of accessibility in infrastructure-poor regions. These are regions characterized by non-homogeneous landscapes with limited or no transportation infrastructure. Even studies that deal with infrastructure-poor regions tend to transpose the methods used elsewhere. This practice seems inappropriate when mobility happens by active rather than motorized modes since the effort required for movement is likely different. The objective of this paper is to compare distance, time, and metabolic energy cost functions in walking accessibility. To this end, we present a case study of accessibility to water in central Kenya. The results indicate that Euclidean distance, surface distance, and travel time correlate better between them than any of them does with metabolic energy. Furthermore, while shortest paths tend to be symmetric for distance and time criteria, under consideration of metabolic energy expenditure pathways change significantly depending on the direction of movement. This has implications for measuring accessibility and equity. By providing alternate mechanisms for valuing the cost of movement, this research suggests avenues to consider vulnerable populations, such as pregnant women who require greater nutritional intake and expend more energy per unit activity. Directions for further research include certain trade-offs between route choice variables across various applications, for example, walking and cycling route choice algorithms.     

Route choice behavior in a radial structured urban network: Do people choose the orbital or the route through the city center?

We use a license plate survey to study route choice through the city center of a medium-sized Dutch city, in which car drivers can basically choose between the orbital and center ring. For a sample of 1397 trips, we fitted a multinomial logit regression model. According to this model, route choice is relatively little influenced by actual travel time. This corresponds with the fact that many drivers did not choose the shortest time route. Travel distance in combination with one “route type velocity” for all orbital routes, and one route type velocity for all center routes is the most decisive factor. The route type velocity indicates how fast and attractive routes of that type are (being perceived). The results support the hypothesis that orbital routes are more attractive as these routes avoid the busy city center. This effect is however partly offset by the fact that drivers also prefer routes in the direction of their destination. These direct routes are mainly center routes. The results show the importance of considering complete routes in relation to their location and not only as a set of links and nodes when modeling route choice.     

A GIS-based toolkit for route choice analysis

This paper develops and applies the route choice analysis (RCA) toolkit. This GIS-based toolkit generates a suite of over 40 variables describing route characteristics such as distance, travel time, speed statistics, number of intersections, number of turns, number of stop signs/stop lights, and a measure of route circuity, to name a few. The input to the toolkit is one or more routes, which can be obtained from global positioning system (GPS) data or some other means (e.g., shortest path). While the toolkit is designed to support route choice modeling by generating variables that have been tested in previous modeling efforts, we demonstrate its utility by testing the hypothesis that workers choose routes to minimize either travel time or distance between home and work. A GPS-enhanced data set of 237 observed routes for home-to-work trips collected for auto drivers in Halifax, Nova Scotia, Canada is used in our analysis. We find that the null hypothesis is refuted - that is, a comparison of observed routes to their shortest-path alternatives based on time and distance via inferential statistics indicates that observed routes are significantly longer compared to their alternatives. This finding suggests that workers may choose routes based on other route attributes. The attributes generated by the RCA toolkit for observed, shortest time, and shortest distance routes are compared and significant differences are noted.     

Methods for determining route distances in active commuting - Their validity and reproducibility

Distance is a variable of pivotal importance in transport studies. Therefore, after checking the validity of a potential criterion method for measuring active commuting route distances, this method was used to assess the validity and reproducibility of four methods of approximating the commuting route distances covered by pedestrians and bicyclists. The methods assessed were: self-estimated distance, straight-line distance, GIS shortest-route distance, and GPS-measured distance. For this purpose, participants were recruited when walking or bicycling in Stockholm, Sweden. Questionnaires and individually-adjusted maps were sent twice to 133 participants. The distances of map-drawn commuting routes functioned as criterion distances. The participants were also asked to estimate their distances. The straight-line distance between origin and destination was measured using map-drawn routes. The shortest route between home addresses and workplace addresses was calculated with three GIS algorithms. Eighty-six trips were measured with GPS. The main results were that test-retest intraclass correlation coefficients (ICC) exceeded 0.99 for all methods, except for self-estimated distance (ICC = 0.76). No order effects existed between test and retest. Significant differences were, however, noted between criterion distance and self-estimated distance (114 ± 63%), straight-line distance (79.1 ± 10.5%), GIS shortest route (112 ± 18% to 121 ± 22%) and GPS distance (105 ± 4%). We conclude that commonly-used distance estimation methods produce systematic errors of differing magnitudes when used in a context of active commuting in suburban and urban environments. These errors can at average level be corrected for, whereas individual relative errors will remain.     

Route choice of bike share users: Leveraging GPS data to derive choice sets

To identify the determinants of bike share users' route choices, this research collects 132,397 hub-to-hub global positioning system (GPS) trajectories over a 12-month period between April 1, 2015 and March 31, 2016 from 750 bicycles provided by Hamilton Bike Share (HBS). A GIS-based map-matching algorithm is used to derive users' routes along the cycling network within Hamilton, Ontario and generate multiple attributes for each route, such as route distance, route directness, average distance between intersections, and the number of turns, intersections, and unique road segments. Concerning route choice analysis, the origin and destination pair should be the same for all routes within a choice set, thus HBS users' trips are grouped by origin-destination hub pairs. Since trips taken by different users between a hub pair can follow the same route, unique routes are extracted using a link signature extraction tool. Following this, a normalized Gini (G_n) coefficient is calculated for each hub pair to evaluate users' preferences among all the unique hub-to-hub route choices. A G_n closer to 0 indicates that routes between a hub pair are more evenly used, while a value closer to 1 implies a higher preference toward one dominant route. Three route choice models, a global model, a medium G_n model, and a high G_n model, are estimated using Path-Size Logit to determine how route choice is affected by the presence of dominant routes. These models suggest that HBS users are willing to detour for some attributes, such as bicycle facilities, but tend to avoid circuitous routes, turns, steep slopes, and roads with high traffic volume.     

The impact of topological properties of built environment on children independent mobility: A comparative study between discretionary vs. nondiscretionary trips in Dhaka

Researchers broadly represented the built environment (BE) using geographic and topological indicators. Despite studies have shown that the geographic BE affects children independent mobility (CIM), little is known about the effects of topological BE on CIM. Less so, how the effects vary between discretionary and nondiscretionary CIM trips. The study addresses these gaps using self-reported two-day mobility data of 151 children aged 10–14 years from Dhaka, Bangladesh. Geographic BE data (e.g. land uses, street width, building height) were collected through a virtual BE audit following each route. Topological BE data (e.g. step-depth, integration, choice) were derived in Depthmap X. CIM was measured in a binary scale by checking whether the reported trips were taken independently or not. Three binary logistic regression models (an overall model, a discretionary trip model, and a nondiscretionary trip model) were estimated to determine the effects of geographic and topological BE on CIM, controlling for other confounding effects. The findings demonstrate that both geographic and topological BE affect CIM. However, they affect discretionary and non-discretionary CIM differently – e.g. step-depth, angular connectivity and presence of institutional land use affect only non-discretionary CIM, whereas integration, recreational land use and traffic composition affect only discretionary CIM. The findings highlight that geographical features need to be considered in tandem with topological features of the BE, stratified by destination types, to maximise CIM.     

Does bicycle network level of traffic stress (LTS) explain bicycle travel behavior? Mixed results from an Oregon case study

This paper draws on census data, mode choice, and regional household travel survey data to investigate the relationship between the varying levels of traffic stress (LTS) routes and bicycle travel behavior. Specifically, does bicycle level of traffic stress explain bicycle behavior enough to warrant its use given the limitations? Bicycle level of traffic stress is a new system of bicycle infrastructure classification for cities that cannot afford existing alternatives such as bicycle level of service which is more data-intensive. The LTS criteria lacks many of the input variables found significant in predicting both route choice and mode choice, including several that are required for its more expensive alternatives. The authors select the Salem-Keizer metropolitan area, a case study representative of the kinds of small and medium-sized communities that may prefer to use the LTS to evaluate cycling infrastructure given its lower input costs. The results validate LTS on travel behavior data from the Oregon Household Activity Survey (OHAS), but not mode choice data from the American Community Survey (ACS). These results suggest the LTS criteria may not be useful for cities looking to prioritize infrastructure improvements for specifically increasing commuter cycling. That said, results suggest the system provides a valid measure of a household's propensity to cycle. Further research on a broader cross-section of communities can clarify these mixed results.     

Route churn: an analysis of low-cost carrier route continuity in Europe

Discontinuity of air routes is a subject that has been analysed in various ways. For example, the complex network approach focuses on network robustness and resilience due to route interruptions during a relatively short period. Also seasonal interruptions of air routes are a well-documented phenomenon in the context of demand variability. However, only recently discussions emerge on the more permanent cessation of air routes, the route churn. Today, European low-cost carriers frequently apply route churn in their networks. To enable early route churn detection in these networks, a regression model is developed in which route characteristics explain the churning likelihood of individual routes. The results of the econometric analysis show that churn rates are higher during an economic downturn, within the Eastern European market, between the Eastern and the Mediterranean market and between primary airports. In addition, we find that distance and the number of seats offered have a significant negative effect on the churn likelihood. The results also show significant effects of market share, seasonality and route age on the chance of cessation. To conclude, the paper demonstrates substantial differences in churn behaviour amongst specific low-cost carriers.     

Decision Support for an Optimal Choice of Subsidised Routes in Air Transportation

The aim of the work presented here is to develop a mathematical model that can assist decision-makers in selecting the optimal network of subsidised routes in air transportation. Sweden is used as a case study. The results show that even if most of the Swedish population already has a good accessibility, the model can suggest new routes that can further improve accessibility without increasing the subsidisation cost. The closure of an airport reduces the available commercial routes, which would impair the accessibility to a given destination; a subsidised route would hence be required as a replacement. The ability of the model to consider several accessibility criteria makes it useful for managers at transportation authorities when making airport location decisions.     

Route choice and residential environment: introducing liveability requirements in navigation systems in Flanders

Vehicle route planning and navigation systems aim to provide the most beneficial routes to their users while disregarding the impact on the liveability of the surrounding residential areas. Therefore, future integration of route choice behaviour by route planners and measures to improve liveability and safety standards should be pursued. The Spatial Plan for Flanders, which is the overarching spatial policy plan in the northern part of Belgium, determines a system of road categories aimed at optimising the liveability of sensitive areas, such as residential neighbourhoods or school precincts, without jeopardizing accessibility. This paper examines to what extent routes proposed by commercial route planners differ from more socially desirable routes that are guided by the policy principles of road categorisation in Flanders as proposed by the plan. Results show that commercial route-planners’ routes choose more often roads of the lowest category than socially acceptable. However, for some of the assessed connections, the socially desired alternative is a feasible route as well, which is not excessively increasing time consumption or distance travelled. It is concluded that the implementation of the prevailing road categorisation system in Flanders in routing algorithms has the potential to promote more sustainable route choices, while infrastructural measures that discourage cut-through traffic may help materialising the categorisation system.     

The Northern Sea Route versus the Suez Canal: cases from bulk shipping

The navigation distance via the Northern Sea Route (NSR) from a Northwest-European port to the Far East is approximately 40% shorter compared to the route via the Suez Canal. The shorter distance may facilitate more than a doubling of vessels’ operational energy efficiency performance. There is at present substantial uncertainty in schedule reliability via the NSR. Unless the schedule reliability is improved, the NSR should primarily be explored for bulk rather than for liner shipping. A major disadvantage with the NSR is its seasonality. Shipping operations in the summer time via the NSR may already today be profitable for minor bulk trades. Additional shipping routes may give more flexibility, and the NSR route choice option may facilitate supply chain agility and adaptability.     

Optimal route decision with a geometric ground-airborne hybrid model under weather uncertainty

Adverse weather is the dominant cause of delays in the National Airspace System (NAS). Since the future weather condition is only predictable with a certain degree of accuracy, managing traffic in the weather-affected airspace is a challenging task. In this paper, we propose a geometric model to generate an optimal combination of ground delay and route choice to hedge against weather risk. The geometric recourse model (GRM) is a strategic Probabilistic Air Traffic Management (PATM) model that generates optimal route choice, incorporating route hedging and en-route recourse to respond to weather change: hedged routes are routes other than the nominal or the detour one, and recourse occurs when the weather restricted airspace becomes flyable and aircraft are re-routed to fly direct to the destination. Among several variations of the GRM, we focus on the hybrid Dual Recourse Model (DRM), which allows ground delay as well as route hedging and recourses, when the weather clearance time follows a uniform distribution. The formulation of the hybrid DRM involves two decision variables - ground delay and route choice - and four parameters: storm location, storm size, maximum storm duration time, and ground-airborne cost ratio. The objective function has two components: expected total ground delay cost and expected total airborne cost. We propose a solution algorithm that guarantees to find the global optimum of the hybrid-DRM. Based on the numerical analysis, we find that ground-holding is effective only when combined with the nominal route. Otherwise, it is optimal to fly on the route determined by the DRM without ground delay. We also find the formula of the threshold ground-airborne cost ratio, which we call the Critical Cost Ratio (CCR), that determines the efficacy of ground delay: the higher the CCR, the more effective the strategies involving ground delay. We conclude that both ground delay and route hedging should be considered together to produce the best ATM decisions.     

Dependence between travel distance,individual socioeconomic and health-related characteristics,and the choice of the travel mode: a cross-sectional study for Kaunas,Lithuania

The aim of this study was to analyse the associations between individual socioeconomic and health-related characteristics, travel distance, and the choice of different travel modes in urban population. A cross-sectional study included 932 adults of Kaunas city, Lithuania. The choice of the travel mode and individual characteristics were self-reported by the participants, and their travel routes were calculated using the geographic information system. Multivariate logistic regression was used to assess the most significant factors determining the choice of a car, cycling, walking, or public transport. In total, 529 participants reported using a car, of whom 65.8% had medium or high education levels. These participants were more likely to be younger, male, married, and employed. Among bicycle users, statistically significant differences between the employment status, body mass index, and travel distance were observed. Walkers were significantly more likely to be older, those with lower incomes, unemployed, and travelling the shortest distances. The analysis of the travel distance on the choice of the travel mode revealed that men travelled longer distances with a car compared to women. The employment status was significantly associated with travel distance by car or public transport. Employed individuals travelled longer distances by public transport or by car, compared to unemployed individuals. Among bicycle users, we found that people with higher levels of education and overweight individuals cycled the longest distances. Our study emphasizes the importance of considering different individual characteristics when analysing the choice of transport modes. It provides evidence that is relevant for all urban populations on the choice of the transport mode, particularly considering active versus passive transport.     

Using spatial network analysis to model pedal cycle flows,risk and mode choice

Spatial network analysis (SpNA) provides a promising alternative to traditional transport models for the modelling of active travel, because walking and cycling behaviour is influenced by features smaller than the scale of zones in a traditional model. There is currently a need for link-level, city wide modelling of cycling, both to ensure the needs of existing cyclists are catered for in planning, and to model the effects of changing infrastructure in shaping cyclist behaviour. Existing SpNA models treat cyclists and car drivers as if they make navigational decisions in a similar way, which in reality is not the case.This paper presents an SpNA model using hybrid betweenness, which fits cyclist flows in Cardiff, Wales using distance, angular distance, motor vehicle traffic and slope as predictors of route choice. SpNA betweenness is also shown to implicitly capture the effect of urban density on mode choice. As it handles route finding decisions of drivers and cyclists separately, the model presented is also applicable to road safety models examining the interaction between the two classes of road user. The model has low cost of data collection and is reproducible using publicly available network analysis software and open mapping data. Further avenues for modelling the effect of infrastructure on cycling are discussed.     

Why do bicyclists take detours? A multilevel regression model using smartphone GPS data

Bicyclists often deviate from the shortest possible routes and take detours in search of more pleasant riding conditions. The extent of detours and the factors affecting bicyclists to ride excess distances have not yet been fully explored. This study aims to measure and analyze the detour extent of utilitarian bicycle trips and their relationships with the route-level environmental components using data collected from individual bicyclists' smartphone GPS in Columbus, Ohio. Comparing the chosen routes with their shortest counterparts, we calculate two detour indices (a distanced-based index and an area-based index) and provide a comparative analysis of built environment attributes for low, moderate, and high levels of detours. We then estimate multilevel mixed-effect generalized linear regression models to identify the contribution of built-environment characteristics to such detours while accounting for individual heterogeneity. We find that most bicycle trips (91.1%) include a detour and are 13.5% longer on average than their shortest alternatives with large variations. Detour degrees are higher for long-distance trips and for peak-period trips. We find that bicyclists choose routes with smaller shares of commercial and single-family land-uses and low levels of land-use diversity. Longer detours are positively associated with street greenery. We find that sparse bicycle facilities and high-speed limits are strong contributors to bicyclists' detour decisions, while multilevel mixed-effect linear regression models further present significant heterogeneity in bicyclists' responses to some environmental attributes. The area-based detour index performs better in explaining the relationships between land-use features and detour degrees.     

Developing advanced route choice models for heavy goods vehicles using GPS data

This paper presents a novel application in route choice modelling using Global Positioning System (GPS) data, focussing on heavy goods vehicles which typically make longer journeys with decisions potentially underpinned by different priorities from those used by car drivers. The scope of the study is larger than many previous ones, using the entire road network of England. Making use of the error components model put forward for route choice by Frejinger and Bierlaire (2007), the work reveals low elasticities in response to changes in travel time, reflecting the limited opportunity for avoiding specific roads on long distance journeys by heavy goods vehicles.     

The maxisum and maximin-maxisum HAZMAT routing problems

We design routes for transportation of hazardous materials (HAZMAT) in urban areas, with multiple origin-destination pairs. First, we introduce the maxisum HAZMAT routing problem, which maximizes the sum of the population-weighted distances from vulnerable centers to their closest point on the routes. Secondly, the maximin-maxisum HAZMAT routing problem trades-off maxisum versus the population-weighted distance from the route to its closest center. We propose efficient IP formulations for both NP-Hard problems, as well as a polynomial heuristic that reaches gaps below 0.54% in a few seconds on the real case in the city of Santiago, Chile.     

A GPS data-based analysis of built environment influences on bicyclist route preferences

This study examines the effects of built environment features, including factors of land use and road network, on bicyclists' route preferences using the data from the city of Seattle. The bicycle routes are identified using a GPS dataset collected from a smartphone application named “CycleTracks.” The route choice set is generated using the labeling route approach, and the cost functions of route alternatives are based on principal component analyses. Then, two mixed logit models, focusing on random parameters and alternative-specific coefficients, respectively, are estimated to examine bicyclists' route choice. The major findings of this study are as follows: (1) the bicycle route choice involves the joint consideration of convenience, safety, and leisure; (2) most bicyclists prefer to cycle on shorter, flat, and well-planned bicycle facilities with slow road traffic; (3) some bicyclists prefer routes surrounded by mixed land use; (4) some bicyclists favor routes which are planted with street trees or installed with street lights; and (5) some bicyclists prefer routes along with city features. This analysis provides valuable insights into how well-planned land use and road network can facilitate efficient, safe, and enjoyable bicycling.     

An analysis of drivers route choice behaviour using GPS data and optimal alternatives

This work aims to study drivers' route choices using a dataset of low frequency GPS coordinates to identify travels' trajectories. The sample consists of 89 drivers who performed 42 thousand paths in the province of Reggio Emilia, in Italy, during the seventeen considered months. Four attributes that may be important for the driver are identified and four optimal alternative paths are created based on the selected objectives to evaluate route choice behaviour. The comparison between the characteristics of the paths allows to conclude that drivers select routes that are overall longer than their optimal alternatives but that allow for higher speeds. Furthermore the statistical analysis of drivers' route choices in macroareas evidences that drivers have different behaviours depending on the geography of the territory. Specifically, there is higher heterogeneity of route choices in the plain areas compared to the mountains. In the second part of this work, clusters of repetitive travels are identified and a Geographical Route Directness Index is proposed to identify the areas of the province where the deviation from the shortest alternative path is higher. The analysis shows that, among groups of repetitive travels, the value of the index is higher along the ring road of the city of Reggio Emilia and there is a strong negative correlation between the frequency the driver selects the longer alternative that allow for higher speed, and the number of additional kilometres the same driver has to travel.