Addressing transit mode location bias in built environment-transit mode use research

https://doi.org/10.1016/j.jtrangeo.2020.102786Get rights and content

Highlights

  • This study compares built environment predictors of train, tram and bus use.

  • Two sample matching approaches are applied to mitigate mode location bias.

  • Results indicate BE impacts vary by mode, irrespective of location types serviced.

  • Differentiating between modes could improve the accuracy of demand forecasting.

  • Stratified sampling is recommended for mitigating mode location bias.

Abstract

Many studies have identified links between the built environment (BE) and transit use. However, little is known about whether the BE predictors of bus, train, tram and other transit modes are different. Studies to date typically analyze modes in combination; or analyze one mode at a time. A major barrier to comparing BE impacts on modes is the difference in the types of locations that tend to be serviced by each mode. A method is needed to account for this ‘mode location bias’ in order to draw robust comparison of the predictors of each mode.

This study addresses this gap using data from Melbourne, Australia where three types of public transport modes (train, tram, bus) operate in tandem. Two approaches are applied to mitigate mode location bias: a) Co-located sampling – estimating ridership of different modes that are located in the same place; and b) Stratified BE sampling – observations are sampled from subcategories with similar BE characteristics.

Regression analyses using both methods show that the BE variables impacting ridership vary by mode. Results from both samples suggest there are two common BE factors between tram and train, and between tram and bus; and three common BE factors between train and bus. The remaining BE predictors – three for train and tram and one for bus - are unique to each mode. The study's design makes it possible to confirm this finding is valid irrespective of the type of locations serviced by modes. This suggests planning and forecasting should consider the specific associations of different modes to their surrounding land use to accurately predict and match transit supply and demand. The Stratified sampling approach is recommended for treating location bias in future mode comparison, because it explains more ridership variability and offers a transferrable approach to generating representative samples.

Introduction

The built environment (BE) influences transit demand by determining the locations to which individuals travel and the ease with which they can be accessed by transit (Cervero et al., 2002; Litman and Steele, 2017; Mitchell and Rapkin, 1954). However little research explores how the BE affects ridership for different transit modes (e.g. heavy/light rail and bus transit). Research in this area faces a major methodological problem: ‘mode location bias’. Transit modes have cost and capacity characteristics and a development history which means some modes always tend to operate in areas of cities with specific BE characteristics.This represents a sample bias when trying to understand how ridership of each transit modes relates to all BE features. A new methodology is needed to objectively explore how each BE feature might affect ridership for each mode, whatever spatial areas they service.

This study develops and applies two new methods to determine how BE attributes link to ridership of individual transit modes accounting for mode location bias. In doing this, the research seeks to establish if different BE factors are associated with ridership for train, tram and bus after mode location bias is addressed.

The remainder of this paper is structured as follows; previous research literature on BE and transit use is outlined including research on mode location bias and how bias is generallyaddressed in the literature. Methodology is then described including the two proposed methods addressing bias. Results are described and then discussed. Conclusions outline findings and implications for research and practice.

Section snippets

Built environment, transit use and transit modes

Empirical research finds an important role for the BE in predicting transit usage (Boulange et al., 2017; Ewing and Cervero, 2010). However almost all studies in the field explore how BE influences transit as a whole. Yet transit modes are distinctive with respect to speed, reliability and space. The priority afforded to different transit modes is a defining characteristic (Kittelson and Associates Inc. et al., 2013). Buses typically have no fixed right of way infrastructure and flexible

Research setting

This study aims to determine whether the BE attributes that predict ridership differ by mode after adjusting for mode location bias. The sample for this study was derived from metropolitan Melbourne's network of almost 20,000 transit stops (Fig. 1) (Public Transport Victoria, 2018b). Melbourne, Australia, has been the subject of studies that explore the relationship between the BE and travel behaviour (Boulange et al., 2017; Jeffrey et al., 2019). Melbourne is also a city with a diverse modal

Results

Descriptive statistics for all facilities in Melbourne are shown in Fig. 1. Descriptives for each sample are available in the online data associated with this paper (3).

Discussion

Table 7 provides an overview of the key results by bias mitigation method and also compares the results with the analysis undertaken without the use of any bias mitigation. Results with alpha levels meeting a critical threshold of 0.05 are considered to have statistical significance and are highlighted bold. Results which do not meet the significance threshold but which have a standardized effects size exceeding 0.1 are included in the table, to show effects that may have practical significance.

Conclusions

This is the first study that compared BE associations with demand for different transit modes while accounting for systematic bias in the location characteristics of modes. Testing unbiased samples makes the main finding of this study generalizable: the BE affects ridership of modes differently, irrespective of the type of location each mode services.

Linear regression of transit ridership for train, tram and bus in Melbourne produce different results when modes are modelled individually, rather

Notes

1 - Results pertaining to the ‘Co-located’ sample were presented as a poster at the 2020 Transportation Research Board Annual Meetingtion Aston et al., 2020.

2 - The data and data aggregation procedures pertaining to this paper are available on Bridges at https://doi.org/10.26180/5d9994f4704ea.

3 - Supplementary results and tables are available on GitHub at https://github.com/Laura-k-a/BE-TU_Melb_Clusters and indexed in supplementary Table 8 associated with this paper.

Acknowledgments

We would like to acknowledge funding provided by the Victorian Department of Transport and the Commonwealth Government's Research and Training Program to support the research activities of Laura Aston, the lead author of this paper.

References (74)

  • A. Monzón et al.

    Efficient urban interchanges: the City-HUB model

    Transportation Research Procedia

    (2016)
  • A. Nasri et al.

    Multi-level urban form and commuting mode share in rail station areas across the United States; a seemingly unrelated regression approach

    Transp. Policy

    (2019)
  • B.D. Taylor et al.

    Nature and/or nurture? Analyzing the determinants of transit ridership across US urbanized areas

    Transp. Res. A Policy Pract.

    (2009)
  • W. Tu et al.

    Spatial variations in urban public ridership derived from GPS trajectories and smart card data

    J. Transp. Geogr.

    (2018)
  • C.L. Aberson

    Applied Power Analysis for the Behavioral Sciences

    (2010)
  • L. Aston

    Exploring Variation in Built Environment Predictors of Ridership by Transit Mode

    In Transportation Research Board Annual Meeting

    (2020)
  • Australian Bureau of Statistics

    Victoria (STE) (Statistical Area Level 1) 2016 General Community Profile: G09H Table Country of Birth of Person by Age by Sex, 2016 Census of Population and Housing

  • Australian Bureau of Statistics

    Victoria (STE) (Statistical Area Level 1) 2016 General Community Profile: Table G01: Selected Person Characteristics by Sex, 2016 Census of Population and Housing

  • Australian Bureau of Statistics

    Victoria (STE) (Statistical Area Level 1) 2016 General Community Profile: Table G32: Dwelling Structure, 2016 Census of Population and Housing

  • Australian Bureau of Statistics

    Victoria (STE) (Statistical Area Level 1) 2016 General Community Profile: Table G38: Dwelling Structure by Household Composition and Family Composition, 2016 Census of Population and Housing

  • Australian Bureau of Statistics

    Victoria (STE) (Statistical Area Level 1) 2016 General Community Profile: Table G43B: Dwelling Structure by Household Composition and Family Composition, 2016 Census of Population and Housing

  • Australian Bureau of Statistics

    Victoria (STE) (Statistical Area Level 1) 2016 General Community Profile: Table G46B Non-School Qualification: Level of Education by Age by Sex, 2016 Census of Population and Housing

  • Australian Bureau of Statistics

    Victoria (STE) (Statistical Area Level 2) 2016 Working Person Profile: Table W01 Labour Force Status by Age by Sex, 2016 Census of Population and Housing

  • Australian Bureau of Statistics

    Victoria (STE) (Statistical Area Level 2) 2016 Working Person Profile: Table W09 Industry of Employment by Sex, 2016 Census of Population and Housing

  • M. Batty

    The New Science of Cities

    (2013)
  • Bicycle Network Victoria

    Parkiteer Locations

  • B. Boehmke

    K-means Cluster Analysis

  • P. Bruce et al.

    Practical Statistics for Data Scientists

    (2017)
  • X. Cao et al.

    Exploring the influences of density on travel behavior using propensity score matching

    Environ. Plan. Plan. Design

    (2012)
  • X. Cao et al.

    Examining the impacts of residential self-selection on travel behaviour: a focus on empirical findings

    Transp. Rev.

    (2009)
  • R. Cervero et al.

    Transit-Oriented Development and Joint Development in the United States: A Literature Review

    (2002)
  • Cervero, R, Murphy, S, Ferrell, C, Goguts, N, Tsai, YH, Arrington, GB, Boroski, J, Smith-Heimer, J, Golem, R, Peninger,...
  • W.G. Cochran

    The effectiveness of adjustment by subclassification in removing bias in observational studies

    Biometrics

    (1968)
  • J. Cohen

    Statistical Power Analysis for the Behavior Sciences

    (1988)
  • C. De Gruyter

    Travel Plans for New Residential Developments : Insights from Theory and practice’, Thesis

    (2017)
  • Y. Deng et al.

    Propensity score weighting with generalized boosted models to explore the effects of the built environment and residential self-selection on travel behavior

    Transpor. Res. Rec.

    (2019)
  • Department of Environment Land Water & Planning

    Planning scheme zones - Vicmap Planning

  • Cited by (0)

    View full text