Urban Spatial Structure and Vehicle Miles Traveled in 461 U.S. Cities

Abstract

This study investigates the effects of socioeconomic characteristics, density, built environment, and urban spatial structure on vehicle miles traveled (VMT) across 461 urbanized areas in the United States. Using multiple regression models, we compare the explanatory power of conventional variables with those representing the spatial distribution of major urban elements, such as population, employment, land use, and travel demand to city centers. The results show that population-weighted distance to the city center and the contiguity of high-intensity land use significantly influence vehicle travel, with greater distances from city centers and more fragmented land use leading to higher VMT. Models incorporating urban spatial structure variables exhibit improved explanatory power and better fit than those including only socioeconomic, density, and built environment variables. These findings demonstrate that urban spatial structure captures key aspects of vehicle travel behavior that are overlooked by traditional measures. Policy implications include the promotion of compact, mixed-use development, infill and brownfield redevelopment, and urban growth boundaries to reduce vehicle dependence. The study highlights the importance of spatial planning in managing urban travel demand and offers a refined analytical framework for examining the interplay between urban form and mobility.

1. Introduction

In the United States, the transportation sector is the largest contributor of Green House Gas (GHG) emissions. It accounts for 28.4% of total U.S. GHG emissions by economic sector in 2022, and vehicles contribute to 80% of emissions from the transportation sector [1]. Indeed, it is estimated that in the absence of substantial reductions in vehicle travel, technological improvements such as increases in fuel efficiency and low-carbon fuels will only slow the rise in GHG emissions from the transportation sector [2,3]. There are significant variations in vehicle travel among urbanized areas (UAZs) in the United States, with per capita daily vehicle miles traveled (DVMT/C) varying by up to five times. For instance, the DVMT/C of Brunswick, California, was 47.8 miles, while Porterville, California, was 9.0 miles in 2020. There is also a large variation in vehicle travel between UAZs with similar population sizes. In the case of UAZs with a population of around 100,000, DVMT/C varies by up to three times. For instance, the DVMT/C of Lawton, Oklahoma, was 14.0 miles, while Portsmouth, New Hampshire, and Maine were 40.3 miles in 2020. These variations in vehicle travel between UAZs are not only due to socioeconomic characteristics but also to urban spatial structure.

Many research studies have been conducted at varying degrees of depth and sophistication on how urban spatial structure influences vehicle travel. Most of these studies have used density and the built environment as measures for describing urban spatial structure. Some studies emphasized the influences of population density [4,5,6,7] and residential density [8,9,10] on vehicle travel. Other studies suggested that mixed land use [11], road networks [3,12,13], transit accessibility [14,15], and walkability [11] have greater influences on vehicle travel than density. These studies have contributed to understanding the relationship between urban spatial structure and vehicle travel and have demonstrated the importance of spatial planning for managing vehicle travel.

However, since urban spatial structure refers to the spatial distribution of various urban elements and interaction among them in the urban system [16,17,18,19,20,21,22,23,24,25,26], it is insufficient to describe urban spatial structure solely by overall density and built environment measures. Overall density is too coarse a measure that cannot describe internal spatial variation, and built environment measures focused on physical features of the urban space are too microscale to describe fundamental determinants of urban spatial structure. Therefore, it is necessary to examine more refined measures that can bridge the gap between overall density and built environment measures to improve our understanding of the relationship between urban spatial structure and vehicle travel.

The objective of this study is to examine the influences of the spatial distribution of major urban elements on vehicle travel of 461 UAZs in the United States. First, we set variables that can describe the spatial distribution of major urban elements such as population, employment, land use, and travel demand to the city center, which were overlooked by previous studies. Then we set three regression models that include these variables, and variables on socioeconomic characteristics, density, and built environment, which were suggested by previous studies to examine their influences on vehicle travel. Moreover, we applied two different assumptions to these regression models and derived some of the spatial distribution variables in two ways in order to examine the influences of centrality on vehicle travel. Next, we examine the significance of variables and their impact on vehicle travel. Then we compare explanatory power and quality of the models to examine whether the model including spatial distribution variables better explains variation in vehicle travel than other models.

We hypothesized that adding spatial distribution variables in the model would result in a decrease in the impact of other variables on socioeconomic characteristics, overall density, and built environment. In other words, spatial distribution variables can explain variances that could not be explained by socioeconomic characteristics, density, and built environment variables suggested by previous studies. Additionally, we hypothesized that the model including spatial distribution variables has better explanatory power and model fit than other models that do not include distribution variables.

Section 2 presents a literature review on studies on density, the built environment, and their influence on vehicle travel. Then, we discuss the analysis framework, data sources, and variables. Section 4 and Section 5 will discuss the results of the empirical analysis of 461 UAZs in the United States and summarize the major findings of the paper.

2. The Literature Review

2.1. Overall Density and Vehicle Travel

Until the early 1980s, studies on vehicle travel had rarely focused on the influences of spatial differences in cities on vehicle travel and insisted that vehicle travel can be accounted for by economic factors such as gasoline price and income [27,28,29]. However, in the late 1980s, Newman and Kenworthy [4] emphasized the role of land use and urban planning to reduce greenhouse gas (GHG) emissions and energy consumption induced by vehicle travel. By examining the relationship between urban land use intensity and vehicle travel in 42 worldwide cities, they insisted that gasoline consumption per capita varies primarily due to land use and transportation planning factors of cities rather than price or income variations and concluded that fuel savings are possible by increasing urban density [4].

Since their study, density measures have been used as key parameters to measure the spatial structure of cities. These studies focused on density measured by population density or job density, and its influence on vehicle travel [8,9], GHG emissions of the transportation sector [5,30], and fuel consumption of vehicles [4,6,7]. Most of these studies concluded that population or residential density has a negative influence on vehicle travel or GHG emissions (Table 1).

However, many studies [13,31,32,33,34,35] have criticized that overall density is insufficient to measure the spatial structure of cities, such as population and job distribution over space. There have also been criticisms [11,13,14,35] that the influence of overall density on vehicle travel is not clear. Some studies [13,36] argue that a more refined multivariate measure of density or parameter referring to the location of urban elements is more significant than overall density.

Table 1. Summary of research studies on density and vehicle travel.

Researcher	Study Cases	Dependent Variable	Explanatory Variables	Major Findings
Newman and Kenworthy [4]	42 Cities (World)	Gasoline use per capita	Population density	There is an exponential negative correlation between gasoline use and population density.
			Job density
			Jobs in the city center
			Population in the inner city
			Average to-work trip length, etc.
Levinson and Kumar [8]	38 cities (U.S.)	Commute distance, time, and speed for automobile commuters	Population density	The higher the residential density, the lower the speed and the shorter the trip distance.
			Number of edge cities
			Presence of heavy rail
			Proportion of travel on freeways, etc.
Taniguchi, Matsunaka and Nakamichi [30]	38 cities (Japan)	Automobile CO2 emissions per capita	Population density	Residents in low-density cities are consuming a lot of automobile fuel.
			Automobiles per capita
			Existence of a tram or a new traffic system, etc.
Glaeser and Kahn [5]	66 metropolitan areas (U.S.)	Emissions from driving and transportation	Population density	Population density is an important determinant of GHG emissions.
			Family income
			Share of employment within 5 miles of the city center, etc.
Brownstone and Golob [9]	2583 households (U.S.)	Annual vehicle miles traveled and fuel consumption	Residential density	A lower density of housing units implies an increase in vehicle miles driven per year and more fuel used per household.
			Household income
			Educational attainment
			Race, etc.
Zhao [37]	Beijing (China)	Commuting by car	Population density	People in areas with a higher population density tend to choose public transport rather than a car for commuting.
			Job-housing balance
			Public transport accessibility
			Household income, etc.
Hong and Shen [10]	Seattle (U.S.)	Road-based transportation emissions per household	Net housing unit density	Increasing residential density leads to a significant reduction in transportation emissions.
			Household income
			Number of workers
			Distance to bus stop, etc.
Creutzig et al. [6]	274 cities (World)	Transport energy use	Population density	Gasoline price and population density correlate most strongly with transport energy use and GHG emissions.
			GDP per capita
			Gasoline price
			Household size, etc.

Table 1. Summary of research studies on density and vehicle travel.

Researcher	Study Cases	Dependent Variable	Explanatory Variables	Major Findings
Newman and Kenworthy [4]	42 Cities (World)	Gasoline use per capita	Population density	There is an exponential negative correlation between gasoline use and population density.
			Job density
			Jobs in the city center
			Population in the inner city
			Average to-work trip length, etc.
Levinson and Kumar [8]	38 cities (U.S.)	Commute distance, time, and speed for automobile commuters	Population density	The higher the residential density, the lower the speed and the shorter the trip distance.
			Number of edge cities
			Presence of heavy rail
			Proportion of travel on freeways, etc.
Taniguchi, Matsunaka and Nakamichi [30]	38 cities (Japan)	Automobile CO2 emissions per capita	Population density	Residents in low-density cities are consuming a lot of automobile fuel.
			Automobiles per capita
			Existence of a tram or a new traffic system, etc.
Glaeser and Kahn [5]	66 metropolitan areas (U.S.)	Emissions from driving and transportation	Population density	Population density is an important determinant of GHG emissions.
			Family income
			Share of employment within 5 miles of the city center, etc.
Brownstone and Golob [9]	2583 households (U.S.)	Annual vehicle miles traveled and fuel consumption	Residential density	A lower density of housing units implies an increase in vehicle miles driven per year and more fuel used per household.
			Household income
			Educational attainment
			Race, etc.
Zhao [37]	Beijing (China)	Commuting by car	Population density	People in areas with a higher population density tend to choose public transport rather than a car for commuting.
			Job-housing balance
			Public transport accessibility
			Household income, etc.
Hong and Shen [10]	Seattle (U.S.)	Road-based transportation emissions per household	Net housing unit density	Increasing residential density leads to a significant reduction in transportation emissions.
			Household income
			Number of workers
			Distance to bus stop, etc.
Creutzig et al. [6]	274 cities (World)	Transport energy use	Population density	Gasoline price and population density correlate most strongly with transport energy use and GHG emissions.
			GDP per capita
			Gasoline price
			Household size, etc.

2.2. Built Environment and Vehicle Travel

Built environment measures encompass various measures related to physical features of the urban landscape [11] in addition to density. Studies on the built environment [3,14,34,35] have argued that density is a proxy for other difficult-to-measure variables that are often expressed by built environment variables, and argued that population density is weakly associated with vehicle travel than with the built environment. These studies [3,11,12,13,15,32,35,38,39,40,41] found positive associations with built environment, such as mixed land use and street connectivity with both walking and transit use, and negative associations with vehicle travel.

These built environment measures were derived from a more micro-scale than overall density and expanded by various studies. For instance, Cervero and Kockelman [11] defined built environment variables as 3Ds in their study: density, diversity, and design. Following studies [32,35,39,41] expanded them to 7 Ds: density, diversity, design, destination accessibility, distance to transit, demand management, and demographics (

Table 2. Built environment variables in previous studies.

	Definitions
Density	population, housing units, job or employment per unit of area, etc.
Diversity	land use composition (e.g., the number of different land uses, jobs-population ratio, jobs-housing ratio, etc.
Design	street network characteristics of an area (e.g., block size, number of intersections, ROAD, and pedestrian-oriented environments, etc.)
Destination accessibility	ease of access to trip attractions (e.g., central business districts, jobs, or other attractions reachable within a given time or distance, etc.)
Distance to transit	the nearest transit stops (e.g., distance to rail station or bus stop, transit route density, transit stop density, distance between transit stops, etc.)
Demand management	parking supply and cost, etc.
Demographics	personal and household characteristics (e.g., income, age, household composition, education, etc.)

).

These studies concluded that the built environment has a stronger relationship to vehicle travel than overall density alone. They emphasized influences of job-housing balance, road density [12], freeway capacity [13], land use fragmentation [40], land use diversity, and urban design factors [15] on vehicle travel. Some studies [3,14] concluded that population or employment density measure is a proxy for the built environment (

Table 3. Summary of studies on the built environment and vehicle travel.

	Study Cases	Independent Variable	Dependent Variables	Major Findings
Cervero and Kockelman [11]	San Francisco Bay Area (U.S.)	Daily vehicle miles traveled per capita	Population density	Compact, mixed-use, pedestrian-friendly development can reduce vehicle miles traveled per capita.
			Employment density
			Retail store density
			Land use dissimilarity index
			Mean entropy of land use
			Percentage of 4-way intersections, etc.
Bento et al. [12]	114 urban areas (U.S.)	Annual vehicle miles traveled	Population density	Population centrality, jobs-housing balance, and ROAD have a significant effect on annual household vehicle miles traveled.
			Population centrality
			Jobs-housing centrality
			ROAD
			Supply of rail/bus transit, etc.
Handy, Cao and Mokhtarian [38]	8 neighborhoods (U.S.)	Vehicle miles driven per week	Accessibility to downtown, shopping areas, freeway, public transit, etc.	Vehicle miles driven are higher for residents of suburban than traditional neighborhoods. Increases in accessibility have a negative effect on driving.
			Physical activity facilities (bike routes, sidewalks, parks, etc.)
			Safety for walking
			Attractiveness (housing styles, appearance of neighborhood, etc.), etc.
Cervero and Murakami [3]	370 urbanized areas (U.S.)	Daily vehicle miles traveled per capita	Population density	Higher population densities are associated with reduced vehicle travel. However, this effect is moderated by the traffic-inducing effects of denser road networks and better local retail accessibility.
			Employment density
			ROAD
			Passenger rail density
			Basic-employment accessibility
			Share of commute trips by private automobile, etc.
Makido, Dhakal and Yamagata [40]	50 Cities (Japan)	Transportation sector CO2 emissions	Population	Less fragmented and compact cities emit less CO2 from passenger transportation.
			Compactness index
			Buffer compactness index
			Mean patch fractal dimension, etc.
Ewing et al. [13]	157 urbanized areas (U.S.)	Daily vehicle miles traveled per capita	Population density	Income and freeway capacity are more significant and have greater elasticities of vehicle miles traveled than density.
			Transit route density
			Freeway lane miles per population
			Transit service frequency
			Compactness index
			Income per capita, etc.
Thé, Carantino and Lafourcade [15]	Metropolitan areas (France)	Fuel consumption	Population density	Density, diversity, and design significantly affect car emissions. Densely populated metropolises with good public-transport networks have lower driving footprints.
			Density of public transit stops
			Distance from residence to CBD
			Average commuting distance
			Fractal dimension in residence
			Road potential
			Rail potential, etc.

).

These studies on the built environment and vehicle travel suggested diverse measures to complement the limitations of density measures and to explain urban spatial structure in more detail. However, there have been criticisms that built environment measures are too microscopic to explain more fundamental determinants that affect vehicle travel. Most built environment measures focused on neighborhood-scale variations in physical environment and facilities such as intersections, transit stops, street blocks, and so forth.

In order to consider the spatial distribution of urban elements from a more macroscopic perspective, some studies [15,38] examined the influence of the distance between the city center and residential areas on vehicle travel. However, these studies are insufficient to describe travel demand variation over space and the degree of centrality of urban spatial structure.

2.3. Summary and Current Research

Studies investigating the relationship between density and VMT frequently conceptualize density in overly simplistic terms, relying primarily on population or residential density while neglecting the broader array of urban elements that shape spatial structure. Such approaches cannot fully address spatial heterogeneity and the complex interactions among urban features. Thus, the conclusion that increasing density necessarily reduces VMT has limited utility for informing concrete urban and transportation policy decisions. Research examining the relationship between the built environment and VMT often employs a range of micro-level variables. However, these studies tend to focus on proximate determinants of travel behavior, thereby overlooking more fundamental factors that shape travel decisions. Moreover, they frequently fail to adequately capture the interactions among various urban elements or among built environment factors themselves. The limited availability of detailed micro-level data on the built environment further constrains such analyses. Consequently, many studies rely predominantly on a narrow set of variables—such as land use mix, road density, and transit density—or on general density measures. These data limitations also restrict the geographic scope of the analyses and limit the generalizability of the findings. Building on the insights from the reviewed studies, this study seeks to bridge the gap between macro-level density measures and micro-level built environment variables by examining whether variables capturing the spatial distribution of key urban elements—such as population, employment, and land use—serve as significant determinants of VMT.

3. Research Design

3.1. Analysis Framework

In order to examine the influences of more fundamental factors of urban spatial structure on vehicle travel than overall density and built environment, as suggested by previous studies, we develop and examine three models (

Table 4. Models containing different set of variables (○ indicates variables included in the model).

	Model 1	Model 2	Model 3
Socioeconomic Variables	○	○	○
Density Variables	○	○	○
Built Environment Variables	○	○	○
Spatial Variation Variables		○ (Monocentric)	○ (Polycentric)

). First, the Model 1 includes variables on socioeconomic, density, and built environment suggested by previous studies to estimate their influences on vehicle travel. Then, we examined two models (Model 2 and 3) that include variables on spatial variation in travel demand and land use, and compared their explanatory power and relative fit with Model 1.

In order to take into account the spatial variation in the center, monocentric and polycentric, and its influences on vehicle travel, Model 2 and Model 3 set different assumptions about the number of centers. Model 2 assumes a monocentric spatial structure in which each UAZ has one center, while model 3 assumes a polycentric spatial structure in which each UAZ has multiple centers, including centers and subcenters.

We use multiple regression analysis to estimate the models. In every set of regression, all variables are transformed into logarithmic form, making relationships among variables more linear and reducing the influence of outliers. The logarithmic transformation also allowed us to interpret regression coefficients as elasticities [13,42]. Socioeconomic variables such as median household income and gasoline price served as statistical controls in the models. In this manner, we are able to compare the influences of spatial variations in urban elements, socioeconomic characteristics, density, and built environment on vehicle travel. In order to compare the model quality of regression, we used Akaike’s information criterion (AIC). AIC considers both model complexity and goodness of fit; $\mathrm{A}\mathrm{I}\mathrm{C}=-2\mathrm{L}\mathrm{o}\mathrm{g}\left(\mathrm{L}\right)+2\mathrm{p}$ where L is maximized value of the likelihood function for the model, p is number of variables in the model [43]. The higher the model’s likelihood, the lower the AIC value. However, as the number of variables increases, the model’s complexity increases, leading to a higher AIC value. Therefore, the model with the lowest one is considered the one best fitting the data.

3.2. Study Cases

We set UAZs as our study cases because UAZs are designated based on densely populated census blocks, making them more appropriate as a spatial unit of analysis than metropolitan statistical areas (MSAs), which are designated based on county boundaries. The FHWA-adjusted UAZs in the contiguous United States (CONUS) serve the study cases because vehicle miles traveled (VMT) data is collected by FHWA-adjusted UAZs rather than census-designated UAZs. While most previous studies set their study cases to specific cities or some UAZs [3,5,8,10,11,12,13,14,44], we set all UAZs as study cases, where data are available, to make our findings more generalizable than previous studies. Of the 542 UAZs listed in the FHWA Highway Statistics 2020 (table hm-72) [45], 65 cases were excluded for data redundancy, and some were excluded for a lack of employment data, and others for a geometry problem with the UAZ boundary shapefiles. A total of 81 UAZs were ultimately dropped from 542 UAZs. Therefore, our final study cases are 461 UAZs (Figure 1).

Table 5. Descriptive statistics for the study cases (N = 461).

	Min	Max	Mean	Standard Deviation
Number of Census Tracts in UAZ	14.0	4872.0	136.1	347.4
Per Capita Daily Vehicle Miles Traveled	9.0	47.8	23.7	6.9
Population	47,409.0	18,999,363.0	549,148.6	1,403,293.0
Median Household Income (in dollars)	41,804.8	141,748.5	65,795.1	16,507.3
Employment	10,073.0	8,513,396.0	251,015.8	654,755.3

summarizes the descriptive statistics of the study cases.

3.3. Data

We obtained data from several sources and merged them with spatial data (

Table 6. Data sources and unit.

Data	Source (Year)	Spatial Unit
Daily Vehicle Miles Traveled	Federal Highway Administration (2020) [45]	Urbanized Area
FHWA-adjusted UAZs Boundaries (shapefile)	Federal Highway Administration (2020) [45]
Population and Socioeconomic Data (ACS 2016–2020 5-year estimated)	U.S. Census Bureau (2020) [46]	Census Tract
Employment (LEHD Origin-Destination Employment Statistics, format version 8.0, year 2020)	U.S. Census Bureau (2023) [47]	Census Block
Cartographic Boundaries and Road Network (TIGER/LINE shapefiles)	U.S. Census Bureau (2020) [46]	1:500,000
National Transit Map (Routes, Stops)	Bureau of Transportation Statistics (2023) [48]	Point and Line
National Land Cover Dataset	United States Geological Survey (2019) [49]	$30\mathrm{m}\times$ 30 m grid
Gasoline Price	American Automobile Association (2023) [50]	State

). While the spatial unit of VMT data is UAZ, the spatial unit of most data is a census tract or a census block. In order to take advantage of the fine resolution of the census tract or block data, and to measure the spatial variations within UAZs, we used these census tract and block-level data for various spatial analysis techniques, such as density gradient and land use patch analysis. Then we aggregated or joined the results with the UAZ boundary (shapefiles) for the consistency of the spatial unit of data. Data analyses were conducted using SPSS (version 29.0.0.0) and QGIS (version 3.40.12).

3.4. Study Variables

We set per capita daily vehicle miles traveled (DVMT/C) as a dependent variable of our regression models rather than vehicle miles traveled (VMT) because DVMT/C can control the variation in population size of study cases (Figure 1). Explanatory variables are classified into four categories: socioeconomic characteristics, density, built environment, and spatial variation in urban elements. As our key research question is examining the influences of spatial variation in population, job, travel demand, and land use on vehicle travel, we first set these variables. These variables are population density gradient (POP_GRAD), job density gradient (JOB_GRAD), population-weighted mean distance to city center (PWD_CEN), number of subcenters (SUBCEN), and contiguity of high-intensity developed area (LPI).

Prior to deriving variables such as POP_GRAD, JOB_GRAD, PWD_CEN, and SUBCEN, we identified city centers and subcenters for all 461 UAZs (Figure 1). Traditionally, city centers or central business districts (CBDs) are defined as areas with a sufficiently high concentration of employment, typically meeting both a minimum employment density and a minimum total employment threshold [19,51,52,53,54]. In this study, city centers and subcenters were defined as employment centers characterized by both large job totals and high job densities. While there are no universally accepted minimum thresholds for total employment or employment density, many previous studies have adopted a single threshold for both criteria. However, applying a uniform threshold without consideration of local context may fail to accurately capture the spatial structure of individual cities [54]. Given that this study focuses on the spatial distribution of population and employment—and that the 461 UAZs analyzed vary widely in size and local characteristics—we adopted variable criteria to identify centers and subcenters more appropriately. The basic minimum thresholds for identifying city centers in this study were set at the 95th percentile of total employment and the 95th percentile of employment density, following the approach suggested by Giuliano et al. [51]. In cases where no census tract met both 95th percentile thresholds, the tract with the highest employment density was designated as the city center. After identifying city centers for each UAZ, subcenters were then determined. Census tracts with the next highest employment densities that also met the 95th percentile thresholds for both total employment and employment density were classified as subcenters. Similarly, if no tract satisfied both criteria, the tract with the highest employment density was designated as a subcenter. To ensure a manageable number of subcenters for estimating polycentric density gradients, and to align with the definition of centers as contiguous clusters of employment, subcenters located within a 5-mile service area of another subcenter were aggregated. The aggregation followed the method proposed by Small and Song [54], in which adjacent subcenters within five miles are merged into a single subcenter with higher total employment and employment density. Applying this process, we found that among the 461 UAZs analyzed, 187 exhibited a monocentric spatial structure—having only a single city center—while 274 demonstrated a polycentric structure with both a city center and one or more subcenters.

Density gradient of population or job (POP_GRAD, JOB_GRAD) represents population/job centrality, which means the degree of concentration of population/job within UAZs. We estimated the monocentric density gradient function by Clark [55] for Model 1 and its multiplicative form [51] for the polycentric density gradient for Model 2. Out of 461 UAZs of our study cases, we found 187 UAZs to be monocentric, which have one center, and 274 UAZs to be polycentric, which have multiple centers.

The density gradient function for a monocentric spatial structure is as follows:

$${\mathrm{D}}_{\mathrm{x}}={\mathrm{D}}_0{\mathrm{e}}^{-\mathrm{g}\mathrm{x}}$$

where ${\mathrm{D}}_{\mathrm{x}}$ is the density at the location $\mathrm{x}$, ${\mathrm{D}}_0$ is the density of the center, $\mathrm{g}$ is the density gradient for the center, and $\mathrm{x}$ is Euclidean distance to the center in miles. This can be transformed into the logarithm form $\ln {\mathrm{D}}_{\mathrm{x}}=\ln {\mathrm{D}}_0-\mathrm{g}\mathrm{x}$ and be calculated by regressing density against distance by Ordinary Least-Squares (OLS) means.

The density gradient function for polycentric spatial structure is as follows:

$${\mathrm{D}}_{\mathrm{x}}=\mathrm{A}{\prod }_{\mathrm{j}}^{\mathrm{J}}\left[\exp \left({-\mathrm{g}}_{\mathrm{j}}{\mathrm{D}\mathrm{i}\mathrm{s}\mathrm{t}}_{\mathrm{x}\mathrm{j}}\right)\right]$$

where ${\mathrm{D}}_{\mathrm{x}}$ is the density at location $\mathrm{x}$, ${\mathrm{D}\mathrm{i}\mathrm{s}\mathrm{t}}_{\mathrm{x}\mathrm{j}}$ is the Euclidean distance between locations $\mathrm{x}$ and subcenter $\mathrm{j}$ in miles, $\mathrm{A}$ is the intercept, and ${\mathrm{g}}_{\mathrm{j}}$ is a density gradient. This equation can be transformed into the logarithm form as follows:

$$\ln {\mathrm{D}}_{\mathrm{x}}={\mathsf{\alpha }}_0+{\mathrm{g}}_1{\mathrm{D}\mathrm{i}\mathrm{s}\mathrm{t}}_{\mathrm{x}\mathrm{c}}+{\sum }_{\mathrm{j}}^{\mathrm{J}}{{\mathrm{g}}_{\mathrm{j}}\mathrm{D}\mathrm{i}\mathrm{s}\mathrm{t}}_{\mathrm{x}\mathrm{j}}(\mathrm{j}\ne 1)$$

where ${\mathrm{D}}_{\mathrm{x}}$ is the density at the location $\mathrm{x}$ that is outside of centers, ${\mathrm{D}\mathrm{i}\mathrm{s}\mathrm{t}}_{\mathrm{x}\mathrm{c}}$ is the Euclidean distance between locations $\mathrm{x}$ and the center in miles, ${\mathrm{D}\mathrm{i}\mathrm{s}\mathrm{t}}_{\mathrm{x}\mathrm{j}}$ is the Euclidean distance between locations $\mathrm{x}$ and subcenter $\mathrm{j}$ in miles, ${\mathrm{g}}_1$ is the density gradient for the center, and ${\mathrm{g}}_{\mathrm{j}}$ is a parameter referring to the density gradient of the subcenter $\mathrm{j}$.

While most previous studies used the mean distance to the center, which cannot take into account travel demand, we use the population-weighted distance to the city center (PWD_CEN) to take into account spatially heterogeneous travel demand within UAZ. Since population is one of the key factors of trip generation [56,57,58], the population-weighted distance to the city center accounts for the fact that the greater the population, the greater the trip generation. Therefore, PWD_CEN represents estimated vehicle miles traveled to the city center and is measured:

$$\mathrm{P}\mathrm{W}\mathrm{D}\_\mathrm{C}\mathrm{E}\mathrm{N}={\displaystyle {\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{\displaystyle \frac{{\mathrm{D}}_{\mathrm{i}}\times {\mathrm{P}}_{\mathrm{i}}}{\mathrm{P}}}$$

where ${\mathrm{D}}_{\mathrm{i}}$ is the Euclidean distance from the census tract $\mathrm{i}$ to the city center in miles, ${\mathrm{P}}_{\mathrm{i}}$ is the population of the census tract $\mathrm{i}$, and $\mathrm{P}$ is the total population of the UAZ.

SUBCEN is defined as the number of subcenters identified within each UAZ, representing the degree of polycentricity in the urban spatial structure.

Because the main characteristics of urban sprawl that led to more vehicle travel are leapfrogging or scattered development, land use fragmentation, and irregularity [42,59,60], we use the largest patch index of high-intensity developed area (LPI) to represent land use contiguity and intensity. LPI was computed by dividing the area of the largest high-intensity developed land by the total area of the UAZ:

$$\mathrm{L}\mathrm{P}\mathrm{I}={\displaystyle \frac{\mathrm{L}\mathrm{P}}{\mathrm{A}}}$$

where $\mathrm{L}\mathrm{P}$ is the area of the largest high-intensity developed land in square miles, and $\mathrm{A}$ is the net land area of the UAZ in square miles.

Afterwards, we set the socioeconomic, density, and built environment variables. Socioeconomic variables used in the models are median household income (INCOME) and gasoline price (P_GAS). These variables served as statistical controls in our analysis. We computed the average median household income of each UAZ from the American Community Survey 2016–2020 5-year estimated.

Density variables used in the models are population density (DEN_POP) and job density (DEN_JOB). While most previous studies used simple overall density measured as population or jobs per unit of area, we used weighted density because it can reflect spatial variation in density within the UAZ [61]. Population (or job)-weighted density is the sum of the densities of census tracts weighted by the sum of populations (or employments) as a percentage of the UAZ total:

$$\mathrm{D}\mathrm{E}\mathrm{N}\_\mathrm{P}\mathrm{O}\mathrm{P}={\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\displaystyle \frac{{\mathrm{P}\mathrm{D}}_{\mathrm{i}}\times {\mathrm{P}}_{\mathrm{i}}}{\mathrm{P}}}$$

where ${\mathrm{P}\mathrm{D}}_{\mathrm{i}}$ is the net population density of the census tract $\mathrm{i}$ in square miles, ${\mathrm{P}}_{\mathrm{i}}$ is the population of the tract $\mathrm{i}$, P is the total population of the UAZ.

Weighted job density was calculated in the same way:

$$\mathrm{D}\mathrm{E}\mathrm{N}\_\mathrm{J}\mathrm{O}\mathrm{B}={\sum }_{\mathrm{i}=1}^{\mathrm{n}}{\displaystyle \frac{{\mathrm{J}\mathrm{D}}_{\mathrm{i}}\times {\mathrm{J}}_{\mathrm{i}}}{\mathrm{J}}}$$

where ${\mathrm{J}\mathrm{D}}_{\mathrm{i}}$ is the net employment density of the census tract $\mathrm{i}$ in square miles, ${\mathrm{J}}_{\mathrm{i}}$ is employment of tract $\mathrm{i}$, and J is the total employment of the UAZ.

Built environment variables used in this study are job-population ratio (JP_RATIO), roadway density (ROAD), intersection density (INT), and public transit density (TRANSIT) (Figure 1). JP_RATIO represents not only the balance between jobs and population but also the degree of mixed land use [42,62]:

$$\mathrm{J}\mathrm{P}\_\mathrm{R}\mathrm{A}\mathrm{T}\mathrm{I}\mathrm{O}={\displaystyle \frac{\mathrm{J}}{\mathrm{P}}}$$

where $\mathrm{J}$ is the employment of the UAZ, and $\mathrm{P}$ is population of the UAZ.

ROAD and INT represent street connectivity, where higher road and intersection density mean more connected streets, more routing options, and a more walkable city [35,42,63]:

$$\mathrm{R}\mathrm{O}\mathrm{A}\mathrm{D}={\displaystyle \frac{\mathrm{R}}{\mathrm{A}}}$$

where $\mathrm{R}$ is the roadway length of the UAZ in miles, and A is the net land area of the UAZ in square miles.

$$\mathrm{I}\mathrm{N}\mathrm{T}={\displaystyle \frac{\mathrm{N}\_\mathrm{I}\mathrm{N}\mathrm{T}}{\mathrm{A}}}$$

where $\mathrm{N}\_\mathrm{I}\mathrm{N}\mathrm{T}$ is the number of intersections in the UAZ, and A is the net land area of the UAZ in square miles.

TRANSIT represents accessibility to public transit and options for commuting mode [64]:

$$\mathrm{T}\mathrm{R}\mathrm{A}\mathrm{N}\mathrm{S}\mathrm{I}\mathrm{T}={\displaystyle \frac{\mathrm{T}\mathrm{S}}{\mathrm{A}}}$$

where $\mathrm{T}\mathrm{S}$ is the number of public transit stops in the UAZ, and A is the net land area of the UAZ in square miles.

It is noteworthy that we tested more possible variables which were expected to have influences on vehicle travel; such as spatial clustering of population and jobs measured by local indicators of spatial association (LISA), land use intensity measured by proportion of high-intensity developed area and mean patch size of high-intensity developed area, and so forth. After several tests, we settled on the above variables that have a relatively good statistical fit and significance. Again, all variables are transformed by taking natural logarithms in the regression analysis (Appendix A.1 and Appendix A.2). Descriptions of variables, data sources, and basic statistics of the study cases are summarized in

Table 7. Variable description and basic statistics for 461 UAZs.

	Variables	Description	Data Source	N	Descriptive Statistics
					Mean	Standard Deviation
Dependent Variable	DVMT/C	Daily vehicle miles traveled per capita	FHWA Highway Statistics, 2020: Table HM72	461	23.857	7.114
Independent Variables	(Density)
	DEN_POP	Weighted net population density	ACS 2020, TIGER/LINE 2020	461	2696.668	2223.847
	DEN_JOB	Weighted net job density	LODES 2020, TIGER/LINE 2020	461	1132.754	506.179
	(Built Environment)
	JP_RATIO	Jobs per population ratio	LODES 2020, ACS 2020	461	40.794	4.397
	ROAD	Roadway density	TIGER/LINE 2020	461	12.303	3.276
	INT	Intersection density	TIGER/LINE 2020	461	215.125	66.408
	TRANSIT	Public transit density	National Transit Map 2023	461	3.960	5.728
	(Spatial Distribution)
	POP_GRAD	Population density gradient	ACS 2020, TIGER/LINE 2020	461	−0.364	0.229
	JOB_GRAD	Job density gradient	LODES 2020	461	−0.490	0.224
	PWD_CEN	Population-weighted distance to city center	LODES 2020, ACS 2020, TIGER/LINE 2020	461	8.384	4.450
	SUBCEN	Number of Subcenters	LODES 2020, TIGER/LINE 2020	461	-	-
	LPI	Largest Patch Index of Hi-Intensity Developed Area	NLCD 2019	461	0.793	0.972
	(Socioeconomic Characteristics)
	INCOME	Median household income	ACS 2020	461	63,120.025	16,463.413
	P_GAS	Regular gasoline price	AAA Gas Prices	461	3.619	0.589

.

Table 8. Variables of the models.

	Model 1	Model 2	Model 3
Dependent Variable	DVMT/C
Independent Variables	DEN_POP DEN_JOB JP_RATIO ROAD INT TRANSIT	DEN_POP DEN_JOB JP_RATIO ROAD INT TRANSIT POP_GRAD (Monocentric) JOB_GRAD (Monocentric) PWD_CEN LPI	DEN_POP DEN_JOB JP_RATIO ROAD INT TRANSIT POP_GRAD (Polycentric) JOB_GRAD (Polycentric) PWD_CEN LPI SUBCEN
	INCOME P_GAS	INCOME P_GAS	INCOME P_GAS

describes the variables included in the models. Model 1 is estimated, wherein socioeconomic, density, and built environment variables are included. Model 2 and Model 3 are estimated, wherein variables on spatial variation in urban elements are added to Model 1. While Model 2 assumes a monocentric spatial structure, Model 3 assumes a polycentric spatial structure. Therefore, Model 2 and Model 3 use different density gradient functions for POP_GRAD and JOB_GRAD, and Model 3 includes an additional variable, the number of subcenters (SUBCEN).

4. Results

4.1. Model 1: Socioeconomic, Density, and Built Environment Variables

The result shows that four variables are statistically significant at the level of 0.001: INCOME, P_GAS, DEN_POP, and DEN_JOB (

Table 9. Regression analysis result of Model 1 (Log–Log Form).

	Coefficient (Unstandardized)	t	p	VIF
(constant)	0.606	0.959	0.338
INCOME	0.457	8.104	0.000	1.569
P_GAS	−0.692	−7.504	0.000	1.840
DEN_POP	−0.324	−10.876	0.000	3.195
DEN_JOB	0.084	4.416	0.000	3.157
JP_RATIO	−0.050	−0.956	0.340	1.721
ROAD	0.155	2.714	0.007	1.623
INT	−0.044	−0.988	0.323	1.577
TRANSIT	0.001	0.320	0.749	1.403
R2	0.457
adjusted R2	0.447
F	23.129
Durbin–Watson	1.932
AIC	−1393.749

). The Model 1 explains 44.7 percent of the variance in DVMT/C. While income (INCOME) has a positive effect on DVMT/C, gasoline price (P_GAS) has a negative effect, as expected. A one percent increase in income will bring about a 0.24 percent increase in vehicle travel, and a one percent increase in gasoline price will bring about a 1.16 percent decrease in vehicle travel. The result also suggests that population density (DEN_POP) has a negative effect on vehicle travel, as expected; a one percent increase in population density will bring about a 0.32 percent decrease in vehicle travel. The result suggests that population density elasticity of vehicle travel is bigger than that of income, but smaller than gasoline price. Interestingly, the sign of job density (DEN_JOB) is positive; however, it is not surprising because there have been debates that the influence of job density on vehicle travel is a mix of positive and negative, or even has no statistically significant effect on vehicle travel [3,35,65,66,67]. On the other hand, built environment variables such as jobs to population ratio (JP_RATIO), road density (ROAD), intersection density (INT), and transit density (TRANSIT) are not significant. The reason is presumed to be that these variables are likely to depend on population density, and population density is often accompanied by a higher supply of roads, intersections, and transit services [65,67].

4.2. Model 2: Including Spatial Variation Variables (Monocentric Spatial Structure)

Model 2 is estimated by adding variables on spatial variation in population, job, travel demand, and land use to Model 1, and assumes a monocentric spatial structure of UAZs. The explanatory power of Model 2, measured by adjusted R², is 0.515, greater than 0.447 of Model 1. The Akaike information criterion (AIC), which states that the lower the value, the better the model quality, of Model 2 (−1406.183) is slightly lower than Model 1 (−1393.749), which means that Model 2 is a better-fitting model that balances goodness of fit without being overly complex and prone to overfitting than Model 1. The results show that five variables are statistically significant at the level of 0.001 (

Table 10. Result of Model 2 (Log–Log Form).

	Coefficient (Unstandardized)	t	p	VIF
(constant)	1.349	2.208	0.028
INCOME	0.323	5.742	0.000	1.730
P_GAS	−0.486	−5.231	0.000	2.066
DEN_POP	−0.252	−8.095	0.000	3.855
DEN_JOB	0.033	1.631	0.104	3.974
JP_RATIO	0.088	1.673	0.095	1.932
ROAD	0.186	3.406	0.001	1.646
INT	−0.033	−0.779	0.437	1.597
TRANSIT	−0.002	−1.028	0.304	1.471
POP_GRAD	−0.009	−0.183	0.855	3.678
JOB_GRAD	0.013	0.521	0.603	3.245
PWD_CEN	0.152	5.008	0.000	2.454
LPI	−0.040	−3.642	0.000	1.351
R2	0.528
adjusted R2	0.515
F	21.951
Durbin–Watson	1.959
AIC	−1406.183

). Two are the spatial variation variables are: PWD_CEN and LPI. However, the other two spatial variation variables are not significant: POP_GRAD and JOB_GRAD. The reason is presumed to be that the density gradient function has limitations in describing spatial variations in population and jobs over space. Socioeconomic variables such as INCOME and P_GAS, and density variables such as DEN_POP, remain significant. However, job density (DEN_JOB) is found to be no longer significant.

The effect size of population weighted distance to the center (PWD_CEN) is 0.152, which means that a one percent increase in distance to the center will bring about a 0.15 percent increase in vehicle travel. This suggests that the farther the distance between the residence and the center, the greater the vehicle travel. On the other hand, the effect size of the largest patch index (LPI) is −0.040, which means that a one percent increase in spatial contiguity of high-intensity land use will bring about a 0.04 percent decrease in vehicle travel; the more contiguous the high-intensity land use, the less the vehicle travel. The influences of density and socioeconomic variables are dropped after spatial variation variables are added to the model. The effect size of DEN_POP dropped from −0.324 in Model 1 to −0.252 in Model 2, INCOME dropped from 0.457 to 0.323, and P_GAS dropped from −0.692 to −0.486. The results prove that the spatial variation in travel demand and land use contiguity can explain the variance that could not be explained by socioeconomic and density variables suggested in previous studies.

4.3. Model 3: Including Spatial Variation Variables (Polycentric Spatial Structure)

Model 3 assumes a polycentric spatial structure in which each UAZ has multiple centers: the center and subcenters. Therefore, Model 3 used a polycentric density gradient and included the number of subcenters (SUBCEN) as an additional variable, which is not included in Model 2. The explanatory power of Model 3 (0.507) is slightly smaller than that of Model 2 (0.515) but still greater than that of Model 1 (0.447). However, the AIC of Model 3 (−848.141) is greater than that of Model 1 and Model 2. It means that the overall quality and relative fit of Model 3 is weaker than the others.

The result suggests that four variables are statistically significant at the level of 0.001 (

Table 11. Result of Model 3 (Log–Log Form).

	Coefficient (Unstandardized)	t	p	VIF
(constant)	1.293	1.612	0.108
INCOME	0.327	4.465	0.000	1.755
P_GAS	−0.498	−4.132	0.000	2.080
DEN_POP	−0.252	−6.199	0.000	3.967
DEN_JOB	0.038	1.429	0.154	4.040
JP_RATIO	0.087	1.281	0.201	1.930
ROAD	0.191	2.697	0.007	1.647
INT	−0.030	−0.548	0.584	1.605
TRANSIT	−0.002	−0.804	0.422	1.488
POP_GRAD	0.003	0.053	0.958	3.299
JOB_GRAD	0.021	0.713	0.477	3.036
PWD_CEN	0.143	3.380	0.001	2.825
LPI	−0.040	−2.838	0.005	1.351
SUBCEN	−0.011	−0.309	0.758	1.890
R2	0.530
adjusted R2	0.507
F	12.047
Durbin–Watson	1.949
AIC	−848.141

). While PWD_CEN, DEN_POP, INCOME, and P_GAS remained significant. LPI is found to be no longer significant at the 0.001 level; however, it is significant at the 0.01 level. SUBCEN is not significant. The effect sizes of INCOME, P_GAS, DEN_POP, and PWD_CEN are not significantly different from those of Model 2.

The lower explanatory power and model fit of the polycentric model (Model 3), compared with the monocentric model (Model 2), are likely due to limitations in the method used to define centers and subcenters. In this study, centers and subcenters were identified based on the number and density of employment, and the number of subcenters (SUBCEN) was employed as a key explanatory variable. Consequently, the model does not account for the internal characteristics or attributes of subcenters, such as their job–housing balance. This finding suggests that the relatively low explanatory power and model fit of Model 3 may be attributed not to the number of subcenters themselves, but rather to whether these subcenters effectively contribute to achieving a better balance between employment and residential locations. For instance, even if multiple subcenters have been developed, their influence on decreasing vehicle travel would be limited if they fail to improve the spatial balance between jobs and resident workers.

5. Discussion and Conclusions

We examined the influences of urban spatial structure, measured by spatial variation in population, job, travel demand, and land use, on vehicle travel to complement the limitations of socioeconomic, density, and built environment variables suggested by previous studies. The results confirm that socioeconomic and density variables such as INCOME, P_GAS, and DEN_POP are associated with vehicle travel as expected; the higher the gasoline price and the density, the lower the vehicle travel, while the higher the income, the greater the vehicle travel [4,5,6,7,8,9,10]. However, in contrast with previous studies [4,68] that concluded population density (DEN_POP) as a major determinant of vehicle travel, gasoline price (P_GAS) and household income (INCOME) were found to have greater effects on vehicle travel since vehicle travel elasticities of P_GAS and INCOME are greater than DEN_POP. On the other hand, variables on the built environment (JP_RATIO, ROAD, INT, TRANSIT) were found to be not significant in our models (

Table 12. Comparison of the three model results.

	Model 1	Model 2	Model 3
(constant)	0.606	1.349	1.293
INCOME	0.457 ***	0.323 ***	0.327 ***
P_GAS	−0.692 ***	−0.486 ***	−0.498 ***
DEN_POP	−0.324 ***	−0.252 ***	−0.252 ***
DEN_JOB	0.084 ***	0.033	0.038
JP_RATIO	−0.050	0.088	0.087
ROAD	0.155 **	0.186 **	0.191 **
INT	−0.044	−0.033	−0.030
TRANSIT	0.001	−0.002	−0.002
POP_GRAD		−0.009	0.003
JOB_GRAD		0.013	0.021
PWD_CEN		0.152 ***	0.143 **
LPI		−0.040 ***	−0.040 **
SUBCEN			−0.011
R2	0.457	0.528	0.530
adjusted R2	0.447	0.515	0.507
F	23.129	21.951	12.047
Durbin–Watson	1.932	1.959	1.949
AIC	−1393.749	−1406.183	−848.141

* : p < 0.05, ** : p < 0.01, *** : p < 0.001.

). This finding contrasts with previous studies [3,12,13,14,15] that concluded the built environment has a greater influence on vehicle travel than population density.

Our major findings are that first, we prove that urban spatial structure measured by spatial variation in major urban elements affects vehicle travel, PWD_CEN, and LPI. The results suggest that the farther the distance between residence and the center, and the more population living farther from the city center, the greater the vehicle travel. It implies that suburbanization and sprawl will generate more vehicle travel, as city centers are major trip destinations for diverse activities. Therefore, encouraging compact development, mixed-use development, or establishing urban growth boundaries around the city center will contribute to reducing vehicle travel by bringing origins and destinations closer together. Our results also suggest that contiguous high-intensity land use (LPI) has a negative influence on vehicle travel; the more contiguous the intensive land use, the lower the vehicle travel. Therefore, encouraging redevelopments of brownfield or underused areas, and infill development, instead of leap-frogging development or fragmented land use, to improve land use efficiency and contiguity around the city center will contribute to reducing vehicle travel. Meanwhile, the vehicle travel elasticities of PWD_CEN were found to be greater than LPI; the population of suburbs and the distance between residence and city center (PWD_CEN) have greater influences on vehicle travel than land use intensity (LPI).

Second, the results suggest that as variables on urban spatial structure (PWD_CEN, LPI) are added to the model, socioeconomic and density elasticities of vehicle travel (INCOME, P_GAS, DEN_POP) are dropped. Despite socioeconomic and density elasticities of vehicles being still greater than those of urban spatial structure, our results support that urban spatial structure variables can explain variances that could not be measured by socioeconomic, density, and built environment variables suggested by previous studies. It implies that urban spatial structure variables presented in our study can contribute to understanding the relationship between urban spatial structure and vehicle travel, which has been overlooked in the macroscopic approach emphasizing overall density and the microscopic approach emphasizing built environment variables.

Third, the results suggest that Model 2, including variables on urban spatial structure, has greater explanatory power and better model fit than the others. Meanwhile, the result that Model 2, assuming a monocentric spatial structure, has greater explanatory power and better model fit than that of Model 3, assuming a polycentric spatial structure, implies that distance and travel demand to the city center have a greater influence on vehicle travel than to subcenters.

Overall, the findings of this study provide empirical evidence for the influence of urban spatial structure on vehicle travel and contribute new insights into the usefulness of spatial structure variables that are not adequately captured by conventional macroscopic density or microscopic built environment measures. The results underscore the importance of urban spatial planning and policy interventions in managing vehicle travel. In particular, the findings highlight the need to enhance the job–housing balance and encourage high-density, mixed land use development within and around city centers and subcenters to alleviate spatial mismatches between employment and residential locations. Furthermore, promoting mixed-use development in suburban areas to increase local employment opportunities is also essential for improving the overall job–housing balance. Policy strategies such as compact city planning, mixed-use and infill development, the redevelopment of brownfield or underutilized sites, and the establishment of urban growth boundaries around city centers could be considered to achieve these goals [42,59,60].

This study has several limitations. First, the models were developed based on a selection of density and built environment variables for which data were available. Future studies could examine a broader range of density and built environment variables to enhance the generalizability of the findings. Second, more sophisticated methods could be developed and applied to measure the spatial variation in additional urban elements, thereby deepening our understanding of the relationship between urban spatial structure and vehicle travel. In particular, this study applied variable thresholds to identify centers and subcenters across diverse urban contexts. However, because the number and spatial distribution of centers and subcenters inevitably vary depending on the chosen thresholds, future research should explore more advanced threshold-setting criteria that more accurately capture urban spatial structures and assess the sensitivity of results to different threshold choices. Third, this study did not sufficiently account for the residential self-selection (RSS) effect, which reflects individuals’ preferences for travel modes and spatial characteristics in their residential choices. Finally, the analysis relied on cross-sectional data from a single year, limiting the ability to capture temporal dynamics in urban structure and travel behavior. Future research should employ more advanced methodologies to control for RSS effects and use longitudinal or time–series data to trace long-term interactions between urban structural changes and travel behavior. Such approaches would allow for a clearer identification of the causal effects of urban spatial structure on travel behavior. Meanwhile, recent research has increasingly examined how emerging work arrangements and transportation modes—such as remote work, electric vehicles and charging infrastructure, autonomous vehicles, and on-demand mobility services—affect vehicle travel. Future studies should investigate the impacts of a broader set of factors, including changes in work patterns and the use of new transportation technologies, as well as urban spatial structure, density, and built environment characteristics, on vehicle miles traveled (VMT).