This research focuses on travel time, commuting patterns and socio demographic factors, by leveraging the US Census LODES data. The approach is twofold, with a trip-focused and time-focused conceptualizations of commuting data: categorizing the ODs by the built environment characteristics; and implementing Generalized Additive Models fitting a curve simulating the number of trips based on travel time and the indicators. Results suggest a significant role played by the BE and sociodemographic characteristics, with demand relating to the impedance and the walkability delta between the OD tracts, supporting the hypothesis of site and trip-specific time values.
Introduction
The aim of this article is to contribute to the 50 years long line of research around travel time (Hägerstrand, 1970), conceptually starting from the foundational work of Marchetti (1994) and Zahavi (1974). In particular, this research is a preliminary study on data acquired through US Census quantifying work-commuting flows, specifically with the objectives of (i) exploring a spatial disaggregation of average travel time, (ii) introducing additional indicators which may influence the average travel time, and (iii) formalize the relation between the trip demand, the travel time, and additional indicators. The main hypothesis is that the average travel time is both site-specific and influenced by built environment and sociodemographic characteristics. These factors have always been part of the traditional Gravity Model formulation (Equation 1), with Kij being a factor to anticipate variables other than trip impedance on trip distribution.
The operationalization of Kij opens to appraisal frameworks aimed at evaluating factors beyond the time saving. (Cervero, 2011), for example by considering accessibility as an externality, which is a concept heavily studied by accessibility research (Geurs & van Wee, 2004), and consequently in relation with walkability (Bielik et al., 2018), 15-Minute City (Moreno et al., 2022), and equity (Pereira & Karner, 2021).
Enabling Data and Methodology
This research focuses on the relation between travel time and commuting patterns, investigating how the time allocated to home-to-work trips varies widely depending on demographic and spatial characteristics. Two main sets of data were considered: one quantifying the commuting flows, and one mapping the social characteristics. The former is the Origin-Destination Employment Statistics (US Census Bureau, ND), which is a fine-grained OD matrix for work commute, which was enriched with a realistic driving time estimate (Figure 1). The latter includes three additional indicators to evaluate the commuting structure, namely the National Walkability Index (NWI) (Thomas & Zeller, 2021), the population density (Pop) and the median household income (Inc). The OD data accounts for ~35 million relations, starting from or ending in 86,842 census tracts, later this was filtered to 30,457,793 OD pairs, to include just trips shorter than 3 hours. All the additional indicators were calculated at the census track level. Figure 1 shows the distribution of the cumulated number of trips plotted aggregated in 2-minute buckets.
The additional data can be evaluated in terms of indicator deltas between the destination and origin tracts. The three indicators, expressed as the average delta variation at each time interval (i.e., ΔNWI, Δpop, and ΔInc), can help to identify statewide trends in the OD pairs and are shown in Figure 2.
Furthermore, the analytical process can be conceptualized as trip-focused and time-focused. The trip-focused approach is implemented to evaluate the individual mean travel time against the variation of the indicators describing the starting and the ending point of each trip. This is formalized in the comparison of average travel time among 16 combinations of NWI intersections, 9 combination of Pop intersections, and 9 combinations of Inc intersections.
The time-focused approach conceptualizes the time as the relevant factor, aggregating the datasets based on time intervals. This approach is used to build a series Generalized Additive Models (GAMs) fitting a curve able to simulate the distribution of trips in relation to the travel time and the indicators deltas. This approach evaluates the role that built-environment and socio-demographic factors play in the definition of commuting patterns. GAMs are derived from Generalized Linear Models (GLMs), by introducing non-linear relationships between predictors and the dependent variable, replacing the linear predictor in a GLM with a sum of smooth functions of the predictors. Equation 2 shows the formulation of the model, where E[Y] is the expected value of X, β0 is the intercept, and fi(xi) are the smooth functions of the predictor variables, in this case a series of splines.
Results
Figure 3 shows the categories’ intersections in an heatmap where the color of the cell depends from the average individual travel time. The first group on the left shows how trips coming from Least Walkable (LW) tracts are in general longer in almost every state, while relations between Above Average Walkable (AAW) or Most Walkable (MW) are usually shorter, except when directed to Least Walkable (LW) areas. Similarly, the second group, categorized by population density, shows a similar pattern, with Rural (R) tracts having a longer driving time than Suburb (S) and Urban (U) ones. No clear pattern emerges from the income categorization.
The results from the GAM curves fitting tend to confirm the outcomes of the categorization analysis, showing variation among states and among the indicators. Figure 4 shows the results of the model fit for each state. Most of the models appear reliable, with a maximum R2 of .998, a minimum of .702, and an average .975, and with a low Root Mean Square Error (RMSE), not reported. Specifically, it appears how the time (CT) is the higher coefficient for 30 over 49 models, followed by ΔPop (13 models), ΔInc (5 models), and ΔNWI (1 model). However, ΔNWI is significant in 37 models, often following CT (which is always significant) in terms of relevance, suggesting that the walkability may play a role in the definition of travel patterns.
It is important to mention that the β coefficients of the GAMs are a simplification of the overall variation. Being a LinearGAM implementation (Servén & Brummitt, 2018), these parameters can be interpreted as the linear effect of each feature in the case all the others are held constant, however the model is fitted using a series of spline, and the complexity of the spline is captured by the partial dependency plots, not reported here.
Conclusions and Future Works
The results are discussed intersecting the outcomes of the two main analyses performed. The left columns of Figure 3 shows that trips originating from Least Walkable (LW) tracts are in average longer than the other categories. The average across states is 58.461min compared to the total average 42.145min. On the contrary, trips starting from Most Walkable (MW) tracts have an average individual travel time of 34.960min, suggesting that walkability plays a role in the definition of mobility patterns. While this is in line with recent works (Alessandretti et al., 2020) and with the outcomes of the GAMs curves (Figure 4), where ΔNWI is often relevant and coupled with CT, additional research will be needed to validate this relation.
The conversation about travel time and travel time budget is going to be a prominent line of research in the following years as it was in the last fifty. This research is an exploratory work aimed at contributing to the conversation, evaluating time-demand relation with the lens of built environment and socio demographic indicators. Moreover, the selection of the three indicators used in this research is arbitrary, a wider array of factors should be explored, building on the concept of site-specificity outlined in this research.
The results of this research have been presented at the 2207 Poster Session “Road Scholars: New Research in Travel Time, Speed, and Reliability Data” of the 104th Transportation Research Board Annual Meeting in Washington DC, January 5-9 2025.
Acknowledgments
The analyzed data were treated according to the GDPR-General Data Protection Regulation (EU, 2016/679). This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. During the preparation of this work the authors used scite.ai tool in order to support a preliminary literature review. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.
References
Alessandretti, L., Aslak, U., & Lehmann, S. (2020). The scales of human mobility. Nature. 587(7834):402-407. https://doi.org/10.1038/s41586-020-2909-1
Bielik, M., König, R., Schneider, S., & Varoudis, T. (2018). Measuring the Impact of Street Network Configuration on the Accessibility to People and Walking Attractors. Networks and Spatial Economics, 18:657–676. https://doi.org/10.1007/s11067-018-9426-x
Cervero, R. (2011). Going beyond travel-time savings: an expanded framework for evaluating urban transport projects. World Bank Group, Washington DC. https://documents1.worldbank.org/curated/pt/466801468178764085/pdf/702060ESW0P1200s0in0Urban0Transport.pdf
Geurs, K., & van Wee, B. (2004). Accessibility evaluation of land-use and transport strategies: review and research directions. Journal of Transport Geography, 12:127–140. https://doi.org/10.1016/j.jtrangeo.2003.10.005
Hägerstrand T (1970). What about people in regional science. Regional Science Association. 24:6-21. https://doi.org/10.1007/BF01936872
Marchetti, C. (1994). Anthropological invariants in travel behavior. Technological forecasting and social change. 47(1):75-88. https://doi.org/10.1016/0040-1625(94)90041-8
Moreno, C., Allam, Z., Chabaud, D., & Pratlong, F. (2022). Proximity-Based Planning and the “15-Minute City”: A Sustainable Model for the City of the Future. In: The Palgrave Handbook of Global Sustainability. Palgrave Macmillan, Cham, pp. 1523-1542. https://doi.org/10.1007/978-3-031-01949-4_178
Pereira, R. H. M., & Karner, A. (2021). Transportation Equity. In: International Encyclopedia of Transportation Vol 1. Vickerman (Eds), Elsevier, pp. 271-277. https://doi.org/10.1016/B978-0-08-102671-7.10053-3
Servén, D., & Brummitt, C. (2018). pyGAM: Generalized Additive Models in Python. Zenodo. https://doi.org/10.5281/zenodo.1476122
Thomas. J., & Zeller, L, (2021). National Walkability Index Methodology and User Guide. U.S. Environmental Protection Agency (EPA). https://www.epa.gov/sites/default/files/2021-06/documents/national_walkability_index_methodology_and_user_guide_june2021.pdf
US Census Bureau (ND). Longitudinal Employer-Household Dynamics. Accessed July 29, 2024
Available at: https://lehd.ces.census.gov/data/ Accessed July 29, 2024.
Zahavi, Y. (1974). Traveltime budgets and mobility in urban areas. U.S. Federal Highway Administration. FHWA-PL-8183. https://rosap.ntl.bts.gov/view/dot/12144