Sustainable Mobility Planning Accommodating Latent Characteristics and Endogeneity Bias in Travel Choice Models

Abstract

Sustainable Mobility Planning (SMP) is a core aspect of future developmental goals of an urban community that emphasises on environmental, social and economic equity. Achieving SUM requires efficient planning, execution and management of transportation systems through effective policy making. Travel behaviour models form an important component of SMP that can be used to understand, analyse and predict different travel choices and preferences among individuals in an urban community. These models include discrete choice models (DCM) that explain an individual’s decision process behind choosing an alternative from a set of mutually exclusive and collectively exhaustive set of alternatives. These alternatives can be either ordered, where there is an order of preference, or categorical, where the order of preference is non-existent. Individual level data used in estimation of these models is obtained usually through road side interviews, household interviews, online surveys etc, using questionnaires curated for the objective at hand. Among discrete choice models, Hybrid Choice Models (HCM) present a flexible architecture that incorporates latent characteristics, which mostly consists of psycho-attitudinal effects, to explain the choices of individuals. Latent characteristics cannot be obtained directly from a respondent. They can only be measured using indicators. Examples of latent variables in travel behaviour modelling include environmental consciousness, attitude towards public transport, intention of adoption etc. Public transportation (PT) is a key component of SMP. Utilization of public transportation reduces congestion and pollution on roads. Planning for public transportation comes with its own set of challenges, and if efficiently done, it can be economically beneficial for both the road user and the service provider. The impact of COVID-19 was one of the most recent challenges to the public transportation ridership. Due to the shared nature of travel, PT became a vector for the virus and aided in its transmission. The fear of the virus made people shift to personal vehicles or ride hailing services. To formulate policies that effectively raise the ridership of PT, it was necessary to analyse this change in mode preference. Trip frequency is an important component of SMP. The advent of pandemic resulted in an online work and shopping culture that reduced mandatory and recreational trips. Understanding the decision-making process behind the intention of individuals to reduce their trip frequency, for various trip types, was necessary to restore the faith of people in preventive measures and to revive the economy. Value of Time (VoT) is an important economic indicator, obtained from mode choice models, which can be used in evaluation of SMP projects. It captures the monetary value that individuals place on their time when making decisions about travel, work, or other activities. It is obtained from model parameters, and its value should be accurate enough to reflect the market scenario. However, error prone data, especially travel time and travel cost attributes collected from individuals, can lead to incorrect VoT estimates. Aptly, the models which utilise this error prone data for estimation should be robust enough to retrieve reliable parameter estimates. Mobility integration is a viable but complex solution to promote SMP. The various approaches towards promoting integrated transportation services have been conventionally local and tailor made to suit the problem. Mobility-as-a-Service (MaaS) offers a potential solution to the problems of mobility integration by providing a mobile based application to plan, book and pay for travel. It provides a seamless integration of mobility based on sustainable modes, including last mile connectivity, and can thus reduce the usage of personal vehicles. However, its implementation in a country like India, with mixed behavioural patterns and mobility, requires an understanding of the willingness of individuals to adopt a bundled mobility system. In context of the discussion presented above, the current study aims to explore the possible reasons for the shift in the mode choice towards non-sustainable modes (if any), due to the pandemic. Further, the impact of pandemic on the intentional reduction in trip frequency is investigated for different trip types. Further, the study experiments on different real-world scenarios to understand the impacts of measurement errors on parameter estimates in discrete choice models. Finally, the study explores the behavioural intentions of Indians, to adopt Mobility-as-a-Service. The first objective of the current study explores the impacts of pandemic on the short-term and long-term mode choice of individuals. A Hybrid Choice Modelling (HCM) architecture was used to unravel the impacts of pandemic oriented variables on the intended shift in mode choice of individuals in a post pandemic scenario. Further, subjective on varying execution of virus preventive measures, the study elicited the stated preference of individuals to choose public transportation in a post pandemic world. Online and offline data collection was performed for the modal shift analysis. The results indicated a significant shift in the mode of travel, where people opted personal vehicles and ride hailing services in place of sustainable modes such as Public Transportation (PT) and Non-Motorised Transport (NMT). Awareness regarding the pandemic, familial and societal conformity, and perceived strictness of lockdown, affected this shift. A significant impact was obtained for preventive measures such as sanitization, social distancing, crowd management and vaccination on the stated preference of public transportation. Further, in the long term, the fear of contracting the virus had a positive and significant effect on willingness to choose non-shared transit, with a parameter value (0.191) for mandatory trips, whereas the willingness to choose non- shared transportation modes for mandatory trips in a post vaccinated scenario reduced for respondents who were already affected by COVID-19, as their belief in preventive steps toward preventing COVID-19 increased, as observed from the negative parameter(-0.48) for mandatory trips . The second objective unravels the association between pandemic oriented variables and the willingness of individuals to reduce their trip frequency for various trip types. Ordered logit models were employed in this study for understanding both short-term and long-term implications. Similar to the first objective, the current study also followed a combination of both online and offline data collection. In short term analysis, the study elicited significant causal relationships between reduction in trip frequency, and various factors including awareness, conformity, perceived strictness of lockdown, subjective wellbeing and satisfaction from working from home. In the long term, trip frequency was affected by the fear of contracting the virus, and the belief in preventive measures. Fear had a positive and significant parameter (0.317), indicating an increased intention to reduce commute trips with an increase in fear, suggesting a need for post trauma care and awareness campaigns to reduce the fear instilled in the people. The third objective formulates a robust and reliable model to deal with measurement errors in travel time and travel cost data. In this study, initially, the data was synthetically generated to reflect real world scenarios of measurement error. Four different architectures of HCMs - additive structural and additive measurement equation model, additive structural and multiplicative measurement equation model, multiplicative structural and additive measurement equation model, and multiplicative structural and multiplicative measurement equation model - were tested on the data to retrieve the true parameter values. The results indicated that the presence of measurement error in data will lead to biased parameter estimates. Among the four HCM architectures, the novel architecture proposed in the current study, an additive structural and a multiplicative measurement equation, was found to be the best performing model. Finally, the models were estimated on a real-world data set (CarPostal dataset- a revealed preference dataset from Switzerland). The results obtained for real-word data was consistent with the observations made for synthetic data. The fourth objective models and analyses the intention of individuals in Chandigarh-Panchkula-Mohali Tricity to adopt MaaS, and further, their willingness to choose a mobility package at 3500 Indian Rupees (₹). The data was collected using face-to-face interview with respondents. In the data collected, substantial number of income values were missing. To deal with this issue, the current study formulated a HCM framework considering the income variable to be a latent characteristic. Further, extensions were provided to this latent variable approach of missing data analysis by testing log normal distributions in the structural and measurement part. Disinclination towards public transport and environmental consciousness were observed to impact the probability of the respondents choosing the curated package of ₹3500. The former demonstrated the need to make public transport more appealing before implementing long-term objectives like mobility integration.