Errors in variables in multinomial choice modeling: A simulation study applied to a multinomial logit model of travel mode choice

Modeling travel demand is a vital part of transportation planning and management. Level of service (LOS) attributes representing the performance of transportation system and characteristics of travelers including their households are major factors determining the travel demand. Information on actual choice and characteristics of travelers is obtained from a travel survey at an individual level. Since accurate measurement of LOS attributes such as travel time and cost components for different travel modes at an individual level is critical, they are normally obtained from network models. The network-based LOS attributes introduce measurement errors to individual trips thereby causing errors in variables problem in a disaggregate model of travel demand. This paper investigates the possible structure and magnitude of biases introduced to the coefficients of a multinomial logit model of travel mode choice due to random measurement errors in two variables, namely, access/egress time for public transport and walking and cycling distance to work. A model was set up that satisfies the standard assumptions of a multinomial logit model. This model was estimated on a data set from a travel survey on the assumption of correctly measured variables. Subsequently random measurement errors were introduced and the mean values of the parameters from 200 estimations were presented and compared with the original estimates. The key finding in this paper is that errors in variables result in biased parameter estimates of a multinomial logit model and consequently leading to poor policy decisions if the models having biased parameters are applied in policy and planning purposes. In addition, the paper discusses some potential remedial measures and identifies research topics that deserve a detailed investigation to overcome the problem. The paper therefore significantly contributes to bridge the gap between theory and practice in transport.     

Exploring the link between built environment and walking choice in São Paulo city,Brazil

This study assesses the connection between the walking choice and built environment characteristics in São Paulo city, Brazil. Trips to work and school were considered because they are the most representative motives to travel in the city, corresponding to more than 70% of the trips. The traditional logit model is applied, having as the dependent variable the choice for walking and as explicative variables, the built environment and socioeconomic variables. The results indicate that, for the city of Sao Paulo, built environment variables are more relevant at the origin and the dimension most related to walking choice is diversity, probably due to socioeconomic reason. The characteristics of individuals also have significant effects, as well as age, gender and income, and this study concludes that they must be considered in the development of local public policies to encourage walking. Also, in developing countries, the D's theories might have some application limits. Apparently, the theory is applicable to a greater social equality context, and in contexts of grater social inequalities there is no influence of built environment on walking choices.     

Associations of built environments with spatiotemporal patterns of public bicycle use

This study examines the association of built environment attributes with spatiotemporal patterns of public bicycle use. The study observations are based on the rental records of YouBike, which is a public bicycle system (PBS) in Taipei Metropolitan Area, Taiwan, from July to December 2015. A hierarchical clustering method is applied to identify the spatiotemporal patterns of YouBike use, and multinomial logit regression is used to clarify the associations. Empirical results indicate that the spatiotemporal patterns of PBS utilization differ in weekdays and holidays and are associated with locations, land use and facilities in a city. The empirical evidence fills the knowledge gap on the factors associated with the spatiotemporal patterns of PBS use and provides system operators with a valuable basis for the integrated planning and management of PBSs and built environments.     

Constructing the cycling citizen: A critical analysis of policy imagery in Brisbane,Australia

Cycling is often promoted as a low cost, accessible and virtuous strategy for solving many urban problems, including air pollution, congestion, climate change and the ‘obesity epidemic’. Yet the status of cycling as a quick and easy transport solution available to all is rarely problematised in policy documents. Focussing on cycling policy documents in sub-tropical Brisbane (Australia) we apply interpretive policy analysis to identify the ways policy representations of cycling and cyclists may work to exacerbate the marginality of certain groups by excluding them from representation. Through analysis of these policy documents, and reference to international research on cycling and the right to the city, this article sketches out the figure of the ‘Cycling Citizen’ constructed within them. The Cycling Citizen is characterised in these policy documents by a combination of actions (such as demonstrations of speed and skill), personal attributes (such as body-type, clothes and gender) and attitudes (particularly around virtue). We argue that the dominance of representations of MAMIL (middle-aged men in lycra) cyclists in the policy documents analysed may work to make cycling less accessible to those less likely to identify as MAMIL such as women, people of colour, people with lower incomes, and fat people, and this may effectively exclude them from cycling policy decisions, and negatively shape personal choices about cycling. We further argue that policy representations of cycling and cyclists matter because they have the potential to influence infrastructure and funding decisions which may have material consequences with respect to cycling mode share, equity and safety.     

Taking another look at cycling demand modeling: A comparison between two cities in Canada and New Zealand

Cities around the world are investing in their cycle network as a way to reduce traffic congestions and improve population health. For effective infrastructure investment decision-making, it is critical to understand the factors affecting cycling demand. This study investigates the combined effects of spatial and temporal factors on the demand of cycling in Auckland, New Zealand and Kelowna, Canada. A latent segmentation-based negative binomial (LSNB) model is developed utilizing daily cycling counts over a year from Auckland and Kelowna. The LSNB model captures unobserved heterogeneity based on the temporal attributes, and land use and neighborhood characteristics of the cycling facilities. The models confirm the influence of weather, bicycle facility type, built environment, traffic, land use, and accessibility attributes. The model results suggest that higher temperature, lower rainfall, increased length of shared paths, lower AADT, and closer distance to bus stop are likely to increase cycling demand in Auckland. The model also confirms heterogeneity. For example, road-connectivity index and bike index vary across the urban and suburban areas of Auckland. For Kelowna, higher temperature, lower rainfall, lower snowfall, and closer distance to water bodies are likely to increase cycling demand. In the case of heterogeneity, the effects of road-connectivity index, bike index, and AADT varies over the urban and suburban areas of Kelowna. Analysis suggests that increasing bike index is one of the critical factors to increase cycling demand in the suburban areas of Auckland and Kelowna.     

Exploring the effects of the rural built environment on household car ownership after controlling for preference and attitude: Evidence from Sichuan,China

With the rapid infrastructure development and economic growth in China, household car ownership in the country's rural areas has changed dramatically in the past 16 years. The total number of cars owned by households in rural areas is currently 12 times higher than it was 16 years ago. The exploration of the effects of the built environment on household car ownership in China's rural areas is worthwhile. However, few studies have investigated this topic. To fill in the research gap, this work collected 374 household data from rural areas in China to examine the effects of the built environment in Sichuan's rural areas on the number of cars in a household. It considered family structure, socioeconomic characteristics, and individual's perceptions of the built environment, preferences towards the built environment and attitudes towards car ownership (shortened to perceptions, preferences and attitudes from now on). Geographic information system (GIS) technology, combined with on-site measurement, was used for data collection. The multinomial logit model was applied for estimation. Household structure and the built environment (including the perceived built environment and the objective built environment) significantly influence the number of cars in a household. By contrast, preference and attitude attributes have less influence on car ownership. Most of the findings are in line with the literature in the context of Chinese cities. Nevertheless, new results are also found. For example, rural hukou, and building density have significant positive impacts on household car ownership in China's rural areas, which is in contrast with their effects on cities. As the first study on rural areas in China, this research provides some insights for rural planners and policymakers to understand better the relationship between built environment and household car ownership.     

Analysis of the impact of street-scale built environment design near metro stations on pedestrian and cyclist road segment choice: A stated choice experiment

The mismatch between the design of the micro-scale built environment around metro stations and pedestrian/cyclist preferences causes inconvenience and dissatisfaction. How to design streets near metro stations to provide a walking/biking friendly built environment is still a key question in promoting the use of metro systems. To identify which general attributes of the street-scale built environment are relevant for pedestrians/cyclists and increase walkability/cycle-ability, this paper reports the results of a stated choice experiment in which eight built environment attributes were systematically varied: street segment length, average number of building floors on both sides of the street, retail shops in frontage of streets, street crossing facilities for pedestrians/cyclists, width of sidewalks/bicycle paths, greenery, density of street lamps and crowdedness of pedestrian/cyclists to understand their influence on a road segment choice and preferences. A total of 803 respondents were recruited from Tianjin, China to complete the stated choice experiment through on-street face-to-face interviews. A multinomial logit model was estimated to unravel pedestrian/cyclist preferences using the stated choice data. The results indicate that pedestrians and cyclists have similar preferences for road segments with building lower than 6 floors, 50% retail shops in frontage, more greenery, lamps between 15 m and 30 m, more crossing facilities, wider sidewalk/bike lane and not crowded. These significant built environment attributes can be used in urban design projects with a walking/biking friendly built environment around a metro station.     

Public transit user satisfaction: Variability and policy implications

This research focuses on passenger's perception of transit performance with an emphasis on the variability between operators and the policy implications of such differences. Two statistical methods (factor analysis and ordered logit modeling) have been used to assess the quality implications of the variability of the users’ perceived satisfaction across operators. A market segmentation analysis (between male and female respondents) provides further insight into the differences among groups of the population. Five transit systems in the two major conurbations in Greece, Athens and Thessaloniki, have been examined. The analysis demonstrated that a well-coordinated transportation environment should be the primary aim of the policy makers in Athens, followed by other quality attributes such as service frequency and accessibility. In Thessaloniki, the sole transit operator should include in its policy plans immediate corrective measures addressing the service frequency, waiting time and vehicle cleanliness attributes.     

Gentrification and the evolution of commuting behavior within America's urban cores, 2000–2015

This study examines recent changes (2000 - ~2015) in the socio-economic/demographic make-up of U.S. urban cores (UCs), and potential associations with commuting behavior including mode and time. Based on a sample of 101 UCs and their encompassing urbanized areas (UAs), the data suggest that UCs have undergone substantial demographic change during the first fifteen years of the 21st century, with key attributes of gentrification on the rise. At the same time, commuting via transit has declined faster within UCs than within their encompassing UAs, while the proportion of workers working from home and cycling to work grew faster than any other mode. Using data from the U.S. Census Bureau's decennial census and American Community Survey (ACS), a series of longitudinal regression models indicated several significant associations between common indicators of gentrification, including proportion of white/non-minority residents, home owners, young adults (18–39), household income, and proportion of college graduates, and changes in non-automotive commute modes (i.e. walking, cycling, transit). This may have implications for transportation infrastructure and policy as urban neighborhoods in the U.S. continue to undergo socio-economic/demographic change.     

Where are the electric vehicles? A spatial model for vehicle-choice count data

Electric vehicles (EVs) are predicted to increase in market share as auto manufacturers introduce more fuel efficient vehicles to meet stricter fuel economy mandates and fossil fuel costs remain unpredictable. Reflecting spatial autocorrelation while controlling for a variety of demographic and locational (e.g., built environment) attributes, the zone-level spatial count model in this paper offers valuable information for power providers and charging station location decisions. By anticipating over 745,000 personal-vehicle registrations across a sample of 1000 census block groups in the Philadelphia region, a trivariate Poisson-lognormal conditional autoregressive (CAR) model anticipates Prius hybrid EV, other EV, and conventional vehicle ownership levels. Initial results signal higher EV ownership rates in more central zones with higher household incomes, along with significant residual spatial autocorrelation, suggesting that spatially-correlated latent variables and/or peer (neighbor) effects on purchase decisions are present. Such data sets will become more comprehensive and informative as EV market shares rise. This work’s multivariate Poisson-lognormal CAR modeling approach offers a rigorous, behaviorally-defensible framework for spatial patterns in choice behavior.     

A latent class choice based model system for railway optimal pricing and seat allocation

In this paper, discrete choice methods in the form of multinomial logit and latent class models are proposed to explain ticket purchase timing of passenger railway. The choice model and demand functions are incorporated into a revenue optimization problem which jointly considers pricing and seat allocation. The framework provides insightful policy implications in term of fare and capacity distribution derived from actual passenger behavior. It shows that accepting short-haul demand provides greater revenue than long-haul demand using the same capacity. Revenue improvement ranges from 16.24% to 24.96% in multinomial logit models and from 13.82% to 21.39% in latent class models respectively.     

Latent class nested logit model for analyzing high-speed rail access mode choice

This paper explores access mode choice behavior, using a survey data collected in Taiwan. The latent class nested logit model is used to capture flexible substitution patterns among alternatives and preference heterogeneity across individuals while simultaneously identifying the number, sizes, and characteristics of market segments. The results indicate that a four-segment latent class nested logit model with individual characteristics in segment membership functions is the most preferred specification. Most high-speed rail travelers were cost-sensitive to access modes, and thus strategies that reduce the access costs can be more effective than reducing the access times.     

Assessing the role of the built environment and sociodemographic characteristics on walking travel distances in Bogotá

Planning for more sustainable mobility is an important goal for public authorities worldwide. Setting better conditions for pedestrians and cyclists has been related to many social benefits, such as less pollution and congestion. Local authorities, aware of the relationship between transport and land-use, are adopting planning tools to analyze growth in cities and their impacts using multiple indicators. This paper develops an econometric trip length model that correlates socioeconomic characteristics and built environment conditions with walking travel distances in the city of Bogotá, Colombia. A log-linear regression model was estimated using information for 2015 of Bogotá's household travel survey, along with data of employment and population by block obtained from official sources. The built environment component of the model serves to evaluate the effect of land-use patterns. A proof of concept was developed using land-use information extracted from a Cellular Automata Land-Use model calibrated for the city. From its results, some factors of the built environment were estimated and two scenarios were tested in this study. One depicts the average walking distance for households surveyed in 2015, and the other takes a simulation of land-use policy in developing a new urban area as an input. The results indicate that people living in areas were the land-use mix and job/population ratio are low, tend to have longer walking distances, suggesting the need to implement special policies that reallocate some activities and encourage higher land-use mix in the urban periphery where low-income households tend to be located. The proposed method can identify the impact of land-use policies in the distances traveled by walking, facilitating the impact assessment of land-use regulations.     

The influence of stratification by motor-vehicle ownership on the impact of built environment factors in Indian cities

Development of compact cities, many contend, aids in the promotion of sustainable modes (public transit, walking and cycling). But the studies done until now have inadequately represented the effect of socio-economic stratification on the relationship between built environment factors and non-motorized transport (NMT) mode choice, which is important in context of a developing country like India. The present study, done in the city of Bangalore, analyzes the influence of built environment factors –density and diversity - on the mode choice and trip distance for the two segments: respondents owning at least one personal vehicle and respondents not owning any personal vehicle. The built environment factors are analyzed for their marginal effects in the presence of various socio-demographic and alternate specific attributes. The results of the built environment factors for the vehicle non-owning group highlighted the requirement of a policy framework to reduce their trip distance by controlling their employment and housing location. The gender of a commuter had a significant effect on the choice of modes, and the results that females had a higher likelihood of using NMT compared with males contradicted the results in other cities. Also, the trip distance model determined that females preferred a shorter walking distance compared with males. Further, the study determined the need for a well-planned, inclusive and coordinated land-use and transport control strategies in the future.     

Regional airport choice: Consumer behaviour and policy implications

The analysis of origin airports in multi-airport regions has a well established tradition in transportation and regional economics. The main goal of the paper is to estimate the importance of the different attributes that determine origin airport choice. In our case we adopt a stated preference approach to study this problem and evaluate the effects of possible policy interventions. We also perform a detailed segmentation of the sample studied according to the socio-economic variables that prove statistically relevant when interacted with the attributes used to characterise airport choice. Moreover, in order to test for the presence of heterogeneity in agents’ preferences we estimate several mixed logit models with different specifications, including heteroscedasticity and error component. With respect to previous studies we develop and extend the traditional SP approach by also analysing the role and relevance of attribute cut-offs in this research field. The policy simulations produced are based on the estimation of airport-specific attributes. The study concentrates on a multi-airport region in central Italy where four competing airports are located.     

Millennials,built form,and travel insights from a nationwide typology of U.S. neighborhoods

We examine the relationship between the built environment and the travel of Millennials in the United States. We develop a neighborhood typology to characterize the built environment and transportation networks in almost every U.S. census tract, allowing us to identify possible synergistic and/or threshold effects on travel. We measure travel behavior in two ways: (1) using a multi-faceted traveler typology created using latent class analysis, and (2) by measuring the vehicle miles of travel among people in each of these traveler types. This dual approach allows us to distinguish between the built environment changes needed to encourage travel by modes other than driving, and those needed to reduce vehicle miles traveled among drivers. Using a multinomial logistic regression, we find that travel patterns are relatively stable along much of the urban-rural continuum, everything else equal. Driving was substantially lower only in “Old Urban” neighborhoods, where densities, job access, and transit service are dramatically higher than in all other neighborhood types. This finding implies that dramatic changes in the built environment—doubling or even tripling development density or transit service—may do little to get young people out of their cars when initial densities or transit services are low, as they are in most of the U.S. Conversely, reducing vehicle miles traveled among drivers appears to require more modest built form changes, a finding that offers some room for optimism among those concerned with auto dependence.     

Does cycling infrastructure reduce stress biomarkers in commuting cyclists? A comparison of five European cities

There is substantial evidence that the environment has an important impact on the use of bicycles. Changes in the built environment, such as cycling infrastructure provision, usually aim at improving the efficiency, enjoyability and safety of cycling. They can also shape affective responses, for instance by triggering or preventing stress situations during cycling. The repeated occurrence of intensely stressful events may make actual cyclists more likely to abandon cycling and deter prospective cyclists from actually taking up this form of mobility. Therefore, using a novel approach, based on stress biomarker measurements obtained directly from cyclists, the objective of this study is to investigate the relationship between urban environment and cyclists' stress. It also investigates if different types of cycling infrastructures in the contexts of two different countries and in five different cities have different relationships with stress. Using a stress sensor, 70 young adults were invited to cycle along a standard route in Oxford, London (the United Kingdom), Amsterdam, Houten and Groningen (the Netherlands). These routes were around 6 km long and had a wide range of characteristics. Multilevel logistic regression analysis indicates that the probability of stressful events occurring is significantly lower on physically segregated cycle paths than on cycle paths on streets, with cycling on general use streets falling in-between these extremes. We also find higher probabilities of stress for primary roads compared to tertiary roads, at intersections than on straight roads, on cobbled and off-road surfaces compared to asphalt, and in noisier places. Models for the individual cities suggested that the relationship between cycling infrastructure and the likelihood of stressful events occurring may depend on the local context. Only for noise conditions, intersection types and cycling infrastructures were the effects consistent across the cities. These findings may be useful for urban infrastructure planning and management, indicating specific attributes that should be adjusted to make cycling less stressful.     

Examining the influence of stop level infrastructure and built environment on bus ridership in Montreal

We studied transit ridership from the perspective of the transit provider, with the objective of quantifying the influence of transit system operational attributes, transportation system infrastructure attributes and built environment attributes on the disaggregate stop level boardings and alightings by time of day for the bus transit system in the Montreal region. A Composite Marginal Likelihood (CML) based ordered response probit (ORP) model, that simultaneously allows us to incorporate the influence of exogenous variables and potential correlations between boardings and alightings across multiple time periods of the day is employed. Our results indicate that headway affects ridership negatively, while the presence of public transportation around the stop has a positive and significant effect. Moreover, parks, commercial enterprises, and residential area, amongst others, have various effects across the day on boardings and alightings at bus stops. An elasticity analysis provides useful insights. Specifically, we observe that the most effective way to increase ridership is to increase public transport service and accessibility, whereas enhancements to land use have a smaller effect on ridership. The framework from our analysis provides transit agencies a mechanism to study the influence of transit accessibility, transit connectivity, transit schedule alterations (to increase/reduce headway), and land-use pattern changes on ridership.     

A spatial analysis of demand patterns on a bicycle sharing scheme: Evidence from London

This paper investigates the spatial demand for bikesharing through the application of a series of trip generation models for the London Bicycle Sharing Scheme (LBSS). The production of trips from and the arrival of trips at scheme stations are evaluated in reference to how they connect with features of the built environment, demographics of the resident and workplace populations, and attributes of the scheme's structure. A spatial econometrics approach is taken to specify the models, with four different time windows considered throughout the day for all trips taken during 2016. The built environment features show a consistent pattern of results in the model, indicating that proximity to cycling infrastructure, rail stations, parks, university facilities, as well as the density of shops and conventional roads in the vicinity of stations is linked with trip generation rates. The presence of males and Caucasians are associated with higher station demand, aligning with other work on the introduction of new mobility solutions elsewhere, though we do find that greater distances to work tend to depress use. Trip generation is also reduced at the minority of stations located south of the River Thames, indicating that the presence of natural barriers can affect the operation of schemes. The results carry implications for scheme integration in other cities.