Content uploaded by Yujie Hu
Author content
All content in this area was uploaded by Yujie Hu on Nov 11, 2022
Content may be subject to copyright.
1
Neighborhood divides: where you live matters for commuting and its
efficiency
Yue Jing1, Yujie Hu1,*, Michał A. Niedzielski2
1GeoNAVI Lab, Department of Geography, University of Florida, Gainesville, FL
32611, USA
2Institute of Geography and Spatial Organization, Polish Academy of Sciences,
Warsaw 00-818, Poland
*Corresponding author: Yujie Hu (yujiehu@ufl.edu)
Abstract
Excess commuting reflects a city’s overall commuting efficiency by quantifying the
proportion of non-optimal commute that would be avoided if resident workers could
freely swap houses or jobs in a given urban form. This framework has been widely
used to evaluate urban land use and transportation policy decision-makings. One
major methodological limitation in the excess commuting literature is that most
existing studies establish on an oversimplified assumption of homogeneous resident
workers/jobs, which neglects the complexity of residential (and employment) location
choices. To fill this gap, this research develops a methodology to measure excess
commuting across worker subgroups differentiated by residential neighborhood type,
which captures the comprehensive socioeconomic profiles of workers defined based
on multiple attributes. Based on a cross-sectional study of Portland, Boston, New
Orleans, and Detroit, it is found that the traditional method overestimates the
commuting benchmarks and causes biased evaluation of commuting efficiency
performance. Moreover, this research reveals significant disparities in excess
commuting across different neighborhood types and identifies three representative
subgroups of distinctive excess commuting patterns, which are the traditional
suburban subgroup, new starts subgroup, and socioeconomically disadvantaged
subgroup. With this proposed methodology, more effective and heterogeneous
policies targeting different population subgroups could be developed.
Keywords: Excess commuting; urban form; residential neighborhood; disparities;
CTPP
1. Introduction
Commuting, a major component of daily urban travel behavior, has been widely
studied to understand the interactions of land use, transportation systems, and human
behaviors. One stream of urban commuting research focuses on the so-called excess
commuting or wasteful commuting (Hamilton and Röell, 1982). Excess commuting
measures the degree of non-optimal and surplus work travel when commuters travel
longer than the theoretical minimum commute suggested by a given urban form
2
(Horner, 2002). Over the past decades, excess commuting has evolved into a
systematic research framework for examining urban form and commuting efficiency
(Murphy and Killen, 2011; Chowdhury et al., 2013; Kanaroglou et al., 2015; Schleith
et al., 2019; Ha et al., 2021), with important implications in land use and
transportation policymaking. Relevant policy debates revolve around a range of social
and environmental issues, such as land use mix, urban dispersion, housing, job
accessibility, automobile emissions, energy consumption, traffic congestion, and road
pricing (Scott et al., 1997; Ma and Banister, 2006a; Yang, 2008; Yang and Zhang,
2019; Zhou et al., 2020). One major focus of related research is about jobs-housing
balance, which has been viewed as an essential planning tool for reducing traffic
congestion and environmental cost, and hence to promote urban sustainability by
encouraging workers to live closer to their workplaces (Korsu and Le Néchet, 2017).
In the excess commuting framework, jobs-housing balance is captured by the
theoretical minimum commute, as a lower value reflects a more balanced spatial
relationship between housing and jobs (Ma and Banister, 2006b; Ma and Banister,
2007; Suzuki and Lee, 2012; Chowdhury et al., 2013; Ha et al., 2021).
At the heart of excess commuting is the comparison between the actual commute and
theoretical minimum commute. The actual commute refers to the average commuting
cost consumed by the whole urban system. The theoretical minimum commute
describes the optimal scenario where the workers are, on average, assigned to the
nearest jobs thus reducing the total commuting cost to the minimum level. To achieve
this optimal commute scenario, most existing studies have adopted a flow re-
distribution mechanism where resident workers could freely swap residences or jobs
under the fixed spatial arrangement of jobs and housing. In this sense, this system-
wide minimum commute is solely determined by the physical land use layout of the
city and captures the degree of the spatial separation between jobs and housing.
However, the underlying assumption of the aforementioned mechanism to measure
the minimum commute is often unrealistic as it lacks behavioral context and instead
depicts an oversimplified process without the consideration of housing and
employment heterogeneity. For example, it would be unmeaningful to exchange
residences between high-income and low-income resident workers due to their
differences in residential location choices and preferences. This stringent assumption
of unlimited mobility embedded in most existing studies masks the intrinsic
complexity of residential location choices and commuting behaviors and is likely to
result in an overestimation of excess commuting, yielding erroneous urban policies
(Ma and Banister, 2006a). Some studies have been conducted to mitigate this issue by
relaxing the homogeneity assumption of resident workers and adding constraints to
the residence swapping process. Most of these efforts include the disaggregation of
resident workers into different worker subgroups by household structure and
characteristics, socioeconomic factors (e.g., occupation type, income, age), and travel
behaviors (e.g., transportation mode and trip-chaining behaviors) (Cropper and
Gordon, 1991; Kim, 1995; Horner, 2002; O’Kelly and Lee, 2005; Ma and Banister,
3
2006a; Murphy, 2009; Zhou et al., 2014; Bwire and Zengo, 2020; Niedzielski et al.,
2020; Antipova, 2020; Hu and Li, 2021). By disaggregating the commuter group into
distinctive subgroups and applying the micro-analysis, these studies generated more
meaningful and accurate estimates of excess commuting.
Nevertheless, most of these existing disaggregation analyses only focused on one
single attribute (Jing and Hu, 2022). Since commuting behavior is shaped by multiple
demographic and socioeconomic characteristics (Green et al., 1999; Hu and Wang,
2016; Hincks et al., 2017; Zheng et al., 2019), disaggregation on single dimensions
may lead to worker groupings that are not as heterogeneous as expected due to
possible interactions with other socioeconomic attributes, such as income and
race/ethnicity (Hu, 2021). As excess commuting is determined based on residence
swapping between workers, a more meaningful disaggregation strategy would be to
account for multiple demographic and socioeconomic attributes that are related to
residential location choices, such as the residential neighborhood type identified by,
among others, housing condition, income level, and race (Delmelle, 2017; Hincks et
al., 2017). In neighborhood studies, compared to the grouping by a single variable,
this disaggregation by residential neighborhood type can help derive worker
subgroups with heterogeneous residential location patterns and accordingly
commuting behaviors.
To fill the above research gap, this research develops a methodology to measure
excess commuting across worker subgroups disaggregated by residential
neighborhood type, which is defined based on multiple socioeconomic attributes. The
methodology is applied to four U.S. cities to reveal inter-neighborhood type variations
in excess commuting using the 2010 census socioeconomic variables and the 2006-
2010 Census Transportation Planning Products (CTPP). Three representative
neighborhood types of distinctive excess commuting patterns across the four cities are
identified. The proposed methodology to disaggregate workers is novel in (excess)
commuting studies. It can identify worker subgroups that are most different from one
another in terms of residential location patterns, which make them ideal subgroups to
study commuting disparities. The methodology can be easily replicated in other study
areas. Besides, this research examines multiple cities of different geographic contexts,
which additionally makes it unique from most existing studies that only look at a
single city.
2. The excess commuting framework
2.1 Concepts and metrics
Excess commuting (Cex) was originally defined to test the predictivity of the classic
monocentric city model in urban economics (Hamilton and Röell, 1982) by
comparing the average observed commute (Cobs) to the average minimum commute
4
(Cmin). See Equation 1 for its formulation. A larger value for Cex indicates a greater
commuting surplus or less efficient commute.
Drawing on the hypothetical exponential density gradient functions of residences and
jobs, Hamilton and Röell (1982) measured the actual commute for 14 U.S.
metropolitan areas and 27 Japanese cities, which, on average, was found to be about
eight times greater than the minimum commute under the monocentric city structure
assumption. They thus concluded that the classical monocentric city model did a poor
job in predicting commuting patterns due to the existence of an enormous amount of
“wasteful” commuting. Following this strand, this modeling approach was later
applied in the Los Angeles and Tokyo metropolitan areas (Small and Song, 1992;
Merriman et al., 1995), where the high extent of excess commuting (70%-90%) was
also reported. White (1988) criticized the modeling approach proposed by Hamilton
and Röell (1982) because it neglected the actual land use geography and
transportation network configuration. Instead, she proposed the linear programming
(LP) method, which had since become the mainstream approach, to calculate the
minimum commute given the existing spatial arrangement of residences and jobs. The
core idea of the LP method is to reallocate the commuting flow in a way that
minimizes the system-wide total commute. Using this method, White (1988) found a
significantly reduced amount of excess commuting (11%) within 25 U.S. cities. The
formulation of the LP approach is given in Equations (2)-(3):
where Xij is the number of resident workers commuting from zone i to zone j, Dij is
the travel distance or time between zones i and j, Wi denotes the total number of
resident workers living in zone i, Ej refers to the total number of jobs in zone j, and N
is the total number of resident workers in the city.
The excess commuting framework was then enriched as other commuting metrics
were developed. In essence, Cmin captures the lower bound of the commute for a given
urban form (Charron, 2007). Alternatively, one may examine the upper bound of
commute, which, combined with Cmin, defines a city’s theoretical commute range.
Horner (2002) developed the theoretical maximum commute (Cmax), which is exactly
the opposite case of Cmin and thus can also be measured using the LP method (see
Equations (4)-(5)). While Cmax might be counterintuitive from the behavioral
perspective (individual workers attempting to maximize their commutes), it provides
supplementary information to Cmin for understanding the urban form as it indicates the
degree of jobs-housing decentralization. A larger Cmax is generally associated with a
more dispersed and sprawling land use pattern. Using this alternative commuting
5
benchmark Cmax, Horner (2002) additionally proposed a new excess commuting
metric Cu that measures the amount of commuting potential (Cmax - Cmin) utilized. A
higher Cu suggests a less efficient commuting system as more commuting potential is
consumed (see Equation (6)).
Cmax essentially represents a most extreme (longest) commuting scenario that may not
be observed in reality. Therefore, Charron (2007) further proposed the concept of
random commute (Crand) to replace Cmax as the upper bound of commute in the
calculation of excess commuting. The random commute refers to the average value of
all possible commutes within the fixed urban form under the assumption that
commuting cost has no impact on residential location decisions. There are two ways
to estimate Crand. The first is to simulate a large number of commuting flows under
the current spatial constraints and calculate the average commuting distance. Another
method to derive Crand is to use a gravity-type model listed in Equation (7) (Charron,
2007; Yang and Ferreira, 2008), which is adopted in this study. Using the random
commute as the new baseline, Murphy and Killen (2011) developed two new excess
commuting metrics—the commuting economy (Ce) and normalized commuting
economy (NCe)—which are formulated in Equations (8)-(9).
The three commuting benchmarks Cmin , Cmax , and Crand and their corresponding
excess commuting metrics have been widely examined in a range of empirical studies,
especially in policy contexts (e.g., Frost et al., 1998; Chowdhury et al., 2013;
Kanaroglou et al., 2015; Bwire and Zengo, 2020; Kim and Horner, 2021). However,
there still exist some methodological issues, such as the modifiable areal unit problem
(MAUP) that is well-known in geographic studies. The MAUP, describing the
uncertainty of area-based analysis caused by scale and zoning effects, is unavoidable
due to the aggregation bias embedded in the zonal datasets with arbitrary boundaries
(Horner and Murray, 2002; Ma and Banister, 2006a; Niedzielski et al., 2013; Hu and
Wang, 2015). Generally, a more aggregated spatial scale leads to less wasteful
6
commuting. The most extreme case occurs when aggregating the whole research area
into one single unit, and the excess commuting would be zero as it would be a definite
jobs-housing balanced system without any inter-zonal commuting. One potential
solution to mitigate this issue is to use commuting datasets as disaggregated as
possible (Horner and Murray, 2002). For example, Hu and Wang (2015, 2016, 2018)
developed a Monte Carlo approach to simulate disaggregated individuals and jobs for
measuring excess commuting, which could generate more accurate estimates.
2.2 Excess commuting across worker subgroups
Another methodological issue in the excess commuting literature is that most existing
studies have made a simple yet unrealistic assumption of homogeneous jobs,
households, and travel behaviors across subgroups of workers. When calculating Cmin
(and Cmax), these studies assumed that any pair of resident workers in a city,
regardless of their socioeconomic traits and job characteristics, could freely swap
residences or jobs (Ma and Banister, 2007; Chowdhury et al., 2013; Hu and Wang,
2016; Zhou and Murphy, 2019). Clearly, this assumption could result in biases in the
estimation of excess commuting.
To mitigate this issue, some studies have proposed methods to disaggregate commuter
groups into subgroups and then measure excess commuting across subgroups. For
example, Kim (1995) disaggregated the resident workers by household structure
(single worker vs. two-worker households) and measured excess commuting for each
household type using the LP method. Buliung and Kanaroglou (2002) grouped
commuters into subgroups based on their mobility restrictions. Some studies have
further explored the disaggregation of workers by occupation type (Horner, 2002;
O’Kelly and Lee, 2005; Ma and Banister, 2006a; Kim and Horner, 2021), income
(Horner and Schleith, 2012), and age (Horner et al., 2015). In general, these
commuter disaggregation approaches reported less excess commuting for the overall
resident workers, compared to existing studies treating all commuters as a whole
group. In addition to these above socioeconomic characteristics of resident workers,
some studies attempted to disaggregate commuters by travel behaviors. For instance,
some research has examined excess commuting disparities across different
transportation modes, such as by private and public modes, across countries including
Ireland, China, Poland, and Tanzania (Murphy, 2009; Zhou et al., 2014; Niedzielski,
2006; Bwire and Zengo, 2020). Most recently, Hu and Li (2021) reformulated the
traditional LP model by integrating disparities in trip-chaining behaviors in the
measurement of excess commuting and revealed the impacts of neglecting the
complexity of travel behaviors in excess commuting modeling.
Despite a range of disaggregation approaches having been pursued, a majority of
these studies essentially only examined one single socioeconomic dimension of
workers. However, residential (and employment) locational choices can vary among
7
other dimensions within a single worker subgroup, such as across racial/ethnic
subgroups for high-income workers. This requires the disaggregation of workers by
multiple socioeconomic characteristics simultaneously into distinctively
homogeneous subgroups, which has been overlooked in the literature.
3. Methodology
3.1 Study area and data
Four U.S. metropolitan statistical areas (MSA)—Portland, Boston, New Orleans, and
Detroit—are chosen as the study area. For simplicity, these four metros will be
referred to as cities hereafter. Compared to single-city case studies, our multi-city
research design can better reveal disparities in commuting and excess commuting
patterns across heterogeneous worker subgroups from a comparative perspective.
These cities are determined based on two criteria—level of residential segregation and
geographic diversity—which both could play a role in commuting patterns. According
to the dissimilarity results in Cortright (2020), two least segregated cities—Portland
and Boston—which are located in the west and east coast of the U.S., respectively—
and two most segregated cities—Detroit and New Orleans—which, respectively, are
located in the north and south—are selected. According to the U.S. Census Bureau, all
these four cities have over 90% of resident workers who both live and work in
corresponding cities, indicating high level of employment self-containment and hence
far fewer inward commuters (Frost et al., 1998). To better reveal urban structure, we
also calculate the average distance to CBD for all workers (DWorkers to CBD) and all jobs
(DJobs to CBD). See Table 1 for more details about the four cities. Notably, the area and
distance to CBD indicators are not positively correlated, which, essentially
exemplifies the distinctive spatial development patterns of the four cities. According
to Table 1, Portland is the most compact city with the largest area but the most
centralized distribution of jobs and housing, while Detroit, conversely, manifests a
much more sprawling pattern.
Table 1. Spatial development comparisons across the four cities
Portland
Boston
Detroit
New Orleans
Area (km2)
36,086.5
21,282.4
20,333.9
26,208.1
Area (with commuting flow, km2)
36,086.5
18,621.1
19,611.6
18,447.8
Employment self-containment index
96%
93%
94%
95%
Size effect
DWorkers to CBD (km)
21.8
29.9
39.8
26.4
DJobs to CBD (km)
17.2
24.7
34.4
21.1
The commuting flow data used in this research are obtained from the 2006-2010
CTPP. The CTPP data include three parts—part 1 on residential places, part 2 on
workplaces, and part 3 on the journey-to-work flow (Hu and Wang, 2016). The CTPP
data are aggregated at multiple spatial scales. In this research, we use CTPP part 3
8
data at the census tract level for measuring excess commuting. The city boundary and
road network of 2010 for the four cities are downloaded from the U.S. Census
Bureau.
3.2 Define neighborhood type
The data of neighborhood types used in this analysis are from Delmelle (2017), which
considered eighteen variables (see Table 2) describing housing (e.g., housing age and
home value), socioeconomic (e.g., education and income), and demographic (e.g., age
and race) profiles using the census tract data obtained from the Longitudinal Tract
Database (Logan et al., 2014). Using the self-organizing map and k-means clustering
methods, census tracts in the 50 largest MSAs in the U.S. were classified into nine
distinctive neighborhood types (see Table 3) for each decade from 1980 to 2010. This
research adopts the most recent result of neighborhood classification for 2010. One is
referred to Delmelle (2017) for more technical details. Figure 1 presents the spatial
distribution and a brief description of each neighborhood type for the study area.
Table 2. The census variables used to define distinctive neighborhood types
Categories
Variables
Housing
Home Value
% Homes > 30 Years Old
% Owner Occupied
% Multiunit Structures
% Vacant Housing
% Households Moved < 10 Years
Demographic
% Black
% White
% Hispanic
% Asian
% Age 60+
% Age <18
Socioeconomic
% College Degree
% Below Poverty
% Unemployed
% Manufacturing
% Service industry
% Foreign Born
Table 3. Descriptions of the nine distinctive neighborhood types
ID
Neighborhood type
1
Wealthy, white, educated
2
Newer single-family homes, white
9
3
White & Asian, multiunit housing, educated, recent in-movers, few kids
4
Older homes, white, some Hispanic, blue collar
5
Hispanic & Black, higher poverty, aging homes
6
Black, high poverty, aging homes
7
Hispanic, high poverty, single-family homes, foreign born
8
Mixed race, average poverty, renters
9
Asians, foreign born, multi-unit, high poverty, recent in-movers
The frequency distributions of residences and jobs across neighborhood types across
the four cities are shown in Figure 2. Some consistent spatial land use patterns are
noticed across the four cities. Urban sprawl can be witnessed across the four cities as
more than fifty percent of households belong to the neighborhoods concentrated in
suburban areas (Type 2 & Type 4; see Figure 1). Comparatively, job suburbanization
is less obvious across the four cities. In terms of job distribution across neighborhood
types, jobs associated with Type 3 neighborhoods dominate the job market in
Portland, Detroit, and New Orleans, whereas jobs associated with Type 2
neighborhoods dominate the market in Boston.
3.3 Measure excess commuting
The network distance is used as the measure of travel cost Dij in this research. The
distance measurement includes two components—interzonal distance (between
10
Figure 1. The spatial distribution of neighborhood types across the four cities
11
Figure 2. Distributions of housing (top) and jobs (bottom) across neighborhood
types across the four cities
different tracts) and intrazonal distance (within the same tract). Interzonal distance is
measured by finding the shortest path along the road network between two tract
centroids (population-weighted), while intrazonal distance is estimated by the radius
of a circle that approximates the area of the tract (Frost et al., 1988).
Next, both Xij (commuter flow between tracts obtained from CTPP part 3) and Dij are
fed into the models in section 2.1 to measure commuting benchmarks (Cmin , Cmax ,
Crand) and excess commuting metrics (Cex , Cu , Ce, NCe) for each city. We first follow
the existing literature to measure excess commuting metrics for the general resident
workers. We then disaggregate Xij by neighborhood type defined in section 3.2 and
calculate excess commuting metrics for each neighborhood type.
4. Results and discussion
4.1 Excess commuting for overall workers
Following most existing research, we first examine the results for the overall workers.
As listed in Table 4, across the four cities, Cmin ranges roughly between 5 and 7 km,
which are significantly lower than Cobs ranging between 15.4 and 20.3 km. In
contrast, Cmax and Crand, which are both highly correlated with the city size,
demonstrate larger variations across the cities (roughly 40-60 km for Cmax and 30-44
km for Crand).
12
The commuting benchmarks provide a lens through which the urban form could be
examined. Boston has the lowest Cmin (4.9 km), indicating the greatest degree of jobs-
housing intermixing among the four cities. Portland and Detroit demonstrate similar
level of jobs-housing proximity as their Cmin are quite close (6 km). In contrast,
suggested by a relatively larger Cmin (7 km), jobs and housing in New Orleans are
more spatially separated, exhibiting the traits of a more monocentric city structure
inducing more inward commutes. Turning to Cmax, Detroit ranks the highest (61.5 km)
followed by Boston (55.5 km), which indicates a greater jobs-housing imbalance than
other cities as it is more likely to generate longer commutes. Cmax of New Orleans
(44.6 km) and Portland (39.1 km) are comparatively smaller, reflecting less dispersed
urban forms. As an alternative upper bound of urban commute, the pattern of Crand is
similar to that of Cmax, but with relatively smaller variance. A more complete picture
of the urban commuting system could be obtained by looking at these commuting
benchmarks together. For example, while the lower bound of commute of Portland
and Detroit are quite similar, the disparity in the potential commuting capacity is
remarkable due to the large difference between the upper bound of the commute
spectrum. Specifically, Portand has the smallest commuting capacity among the four
cities, which might be due to its compact spatial development patterns formed under
the strict urban containment policies. By contrast, the spatial configuration of Detroit
bears much larger commuting capacity, as both jobs and housing are more dispersed
throughout the area caused by urban decay and sprawl.
In terms of excess commuting, Detroit and Boston have less efficient commuting
systems than other cities exemplified by a remarkably high Cex (about 70%).
Specifically, workers of Detroit travel the longest average distance to jobs (20.3 km),
which is much higher than the required minimum commute, leading to a large portion
of excess commuting (70.3%). This means more than 70% of the observed commute
in Detroit could be unnecessary under its current jobs-housing distribution. In
contrast, although the observed commuting distance of Boston is the lowest (15.4
km), it is far from its required minimum commute, which is the smallest among the
four cities, suggesting there is still much potential for Boston to reduce commute cost
given the current urban form. Comparatively, the commuting systems of New Orleans
and Portland are more efficient given their lower Cex, which are 57.3% and 60.4%,
respectively.
The results for commuting efficiency could be ambiguous when looking at the other
three metrics, i.e., Cu, Ce, and NCe, which integrate the upper bound of commute
range. In fact, excess commuting metrics depend on the relative distance between the
actual commute and the theoretical commute bounds, implying different commuting
behavioral preferences. For example, contrary to the conclusion drawn from Cex, the
commuting system of Boston appears to be the most efficient among the four cities in
terms of Cu, Ce, and NCe. This makes sense because on a relative basis, the actual
commute of Boston, which has the second largest commuting capacity, simply locates
closer to the lower bound of its commute range. On the contrary, Portland becomes
13
the most inefficient city with regard to the highest Cu as well as the lowest Ce and NCe
among the four cities, as the commuters are seemingly less sensitive to the commute
distance given a relatively limited commute range.
14
Table 4. Comparisons of commuting and excess commuting metrics across four cities
Cmin
Cmind
D
(%)
Cobs
Crand
Crandd
D
(%)
Cmax
Cmaxd
D
(%)
Cex
Cexd
D
(%)
Cu
Cu d
D
(%)
Ce
Ced
D
(%)
NCe
NCe
d
D
(%)
Portland
6.1
8.5
39.3
15.4
31.0
30.0
-3.2
39.1
38.8
-0.8
60.4
45.0
-25.5
28.1
22.8
-18.9
50.4
48.
6
-
3.6
62.7
67.
8
8.1
Boston
4.9
7.6
55.1
15.4
41.9
39.1
-6.7
55.5
53.6
-3.4
68.0
50.5
-25.7
20.7
16.9
-18.4
63.2
60.
6
-
4.1
71.7
75.
3
5.0
New
Orleans
7.1
9.7
36.6
16.7
35.3
31.0
-
12.2
44.6
43.0
-3.6
57.3
41.8
-27.1
25.6
20.9
-18.4
52.6
46.
0
-
12.
5
66.0
67.
1
1.7
Detroit
6
8.5
41.7
20.3
44.4
41.7
-6.1
61.5
58.0
-5.7
70.3
52.2
-25.7
25.7
21.9
-14.8
54.3
51.
3
-
5.5
62.9
66.
9
6.4
Note: The superscript d refers to the mean value of commuting and excess commuting metrics among neighborhood types and D refers to the percentage deviation in metric
values from the overall workers scenario.
15
Next, we turn to examine to what extent the proposed disaggregation analysis by
neighborhood type could affect the estimates of excess commuting for the overall
workers. Specifically, we measure excess commuting for each neighborhood type (see
section 4.2 for more details) and calculate the weighted mean values across
neighborhood types as a proxy for the overall workers’ (excess) commuting
performance. See Table 4 for more details.
Clearly, the procedure of restricting residence (or job) exchange within the same
neighborhood type in the LP model adds further spatial constraints to the optimization
process, resulting in less significant commuting thresholds, i.e., greater Cmin and lower
Cmax or Crand, and, as such, smaller commuting capacity. This makes sense as the
general workers would be deprived of some commuting possibilities due to more
restricted residential (or employment) location choices. However, the magnitude of
change in these commuting metrics varies widely. The difference, for example, is
significantly higher for Cmin (37% to 55%) than for Cmax (1% to 6%) and Crand (3% to
12%) across cities. This suggests that Cmin perhaps is more sensitive to the
disaggregation analysis compared to commute upper bound Cmax or Crand. In terms of
excess commuting, the disaggregation yields a sharp decrease in Cex for the general
workers, which ranges between 25.5% and 27.1% across cities due to the significant
increase in Cmin. In contrast, the changes in Cu, Ce, and NCe resulting from the
disaggregation are much smaller due to the relatively modest changes in Cmax or Crand.
4.2 Excess commuting among worker subgroups by neighborhood type
In this section, we take a closer look at the excess commuting variation across
neighborhood types across cities. Figure 3 illustrates the values of commuting metrics
across neighborhood types across cities. In addition, the heatmaps of excess
commuting metrics are displayed to further illustrate the inter-neighborhood disparity
in commuting efficiency performances (Figure 4). Overall, it is found that the inter-
neighborhood type variation of commuting benchmarks and excess commuting
metrics is remarkable, indicating significant disparities in both actual and theoretical
commuting behaviors across neighborhood types. This signifies the importance of
commuter group disaggregation by neighborhood type in (excess) commuting studies.
Figure 5 demonstrates the deviations from the baseline scenario (without
disaggregation) values for each neighborhood type’s (excess) commuting metrics
across cities. Obviously, for any given (excess) commuting metric, the deviations vary
across neighborhood types across cities. To better understand the inter-neighborhood
type variation in (excess) commuting metrics across cities, we calculate and plot two
types of distance metrics associated with each neighborhood type in Figure 6—
distance to CBD from each tract, which reflects the overall spatial separation from a
given neighborhood type to CBD, and distance between every two tracts of the same
neighborhood type, which reveals the overall spatial separation between
16
neighborhoods of the same type. The distance metrics can help understand the
characteristics of city structure in terms of neighborhood spatial configurations.
Figure 3. Values of commuting benchmarks across neighborhood types across cities
Figure 4. Heatmaps of the excess commuting metrics across neighborhood types
across cities
17
Figure 5. Deviations of (excess) commuting metric values from the baseline scenario
(no disaggregation) across neighborhood types across cities
18
Figure 6. Density curves of distance distributions across neighborhood types across
cities: (a) distance to CBD and (b) intra-neighborhood type distance. [Density is
calculated using the Gaussian kernel, and the bandwidth is calculated using
19
“Silverman’s thumb” method (Silverman, 1986).]
The different skewness of the kernel density curves in Figure 6, especially for
Portland, Boston, and New Orleans, indicates a clear discrepancy in neighborhood
typologies between central city and suburbs. The central city witnesses a greater
diversity and mixing of different neighborhood types, characterized by the hybrid of
high poverty minority clusters (type 6, 7, and 9) and highly educated, White, and
Asian populations (type 3; also termed as the “New Starts” in Mikelbank (2011)). By
contrast, the suburbs are more homogeneous, predominantly inhabited by the White
residing in newer single-family homes (type 2) and a few scattering neighborhoods
featuring a mixing of the White and Hispanic, blue-collar family (type 4). The spatial
distributions of neighborhood types in these cities conform to the traditional
metropolitan model contending that the metropolitan area consists of a diversified
urban core surrounded by homogeneous suburbs (Finer, 2003; Hanlon et al., 2006).
As the most acclaimed segregated city, Detroit exhibits a rather distinctive spatial
pattern of neighborhood types. As shown in Figure 1 and Figure 6, neighborhoods in
the central city of Detroit are more spatially isolated by both race and income, while a
more diversified and intermixing neighborhood composition is found in the inner
suburban areas.
In terms of excess commuting variation across neighborhood types, some general
trends and patterns are observed across the four cities, bringing three most
representative worker subgroups which share common patterns across cities into our
focus: the traditional suburban subgroup, new starts subgroup, and socioeconomically
disadvantaged subgroup.
The traditional suburban subgroup is the neighborhood type featuring White, newer
single-family homes, and high socioeconomic status (type 2), which has experienced
substantial growth in suburban areas. This neighborhood type, where the dominant
workforce resides, is the most spatially dispersed over the suburban areas as
suggested by its rather flattened density curves in Figure 6. As demonstrated by the
inter-neighborhood type variation of Cobs in Figure 5, workers of this subgroup have
the longest average actual commuting distance across the four cities due to their
overall low job proximity. This finding is consistent with previous studies examining
commuting patterns of this worker subgroup (Shen, 2000; Morris and Zhou, 2018).
Being socioeconomically advantaged, this subgroup exhibits higher tolerance of
longer commute and is willing to endure high travel costs in terms of the tradeoff
between commute and housing.
Looking at other commuting benchmarks (Cmin, Crand, Cmax), as expected, this
subgroup also ranks the highest for all these commuting benchmarks across all four
cities. The commuting benchmarks of this subgroup are significantly higher than
those of the baseline scenario (no disaggregation), with a percentage difference
ranging from 112%-152% for Cmin, 24%-72% for Cmax, and 27%-49% for Cmax, which
20
marks a high degree of spatial separation between residences and jobs. Note that for
Cmax and Crand, this subgroup is the only subgroup exceeding the baseline scenario
across the four cities, indicating the most imbalanced jobs-housing relationship. In
terms of excess commuting metrics, Cex of this subgroup varies between 37.3% and
47.1%, which is much lower than the baseline scenario (57.3%-70.3%), implying
higher commuting efficiency for this subgroup relative to the overall workers. Similar
findings are also found for Cu, Ce, and NCe. The greatest commuting efficiency in this
subgroup is largely ascribed to its widest commuting capacity associated with its
sprawling spatial pattern.
The new starts subgroup is the neighborhood type dominated by the White and
Asian, high-educated recent residents with few kids residing in multiunit housing
(type 3). This neighborhood type, with a high concentration of employment
opportunities, is largely the outcome of gentrification and neighborhood revitalization
(Delmelle, 2017). The spatial patterns of this subgroup are slightly different across the
cities. As displayed in Figure 6, in Portland, Boston, and New Orleans, this subgroup
tends to concentrate in the central city integrated with other neighborhood types,
while in the case of Detroit, it is mainly scattered throughout the inner ring area.
Although this subgroup is quite close to the traditional suburban subgroup at the
socioeconomic spectrum, it manifests a different commuting pattern compared to type
2 neighborhoods.
As the most job-rich neighborhood type, owing to a high intermixing of jobs and
housing distributions, the new starts subgroup tends to have a lower Cmin than other
neighborhood types do. Most notably, for Portland, Boston, and New Orleans, the
value of Cmin is even lower than that of the baseline scenario with a deviation ranging
between -35% and -18%. As for Cmax and Crand, the results vary across cities. In
Portland, Boston, and New Orleans, the new starts subgroup ranks the top 2 subgroup
with the lowest value of Cmax and Crand among neighborhood types due to a more
concentrated spatial pattern of jobs and housing. In Detroit, however, this subgroup is
ranked in the middle in terms of Cmax and Crand due to a more dispersed spatial
distribution across the inner ring area. This pattern also holds for the actual commute
Cobs among cities. That said, despite being a socioeconomically advantaged subgroup
that can afford a wider range of residential options like the traditional suburban
subgroup, the new starts subgroup shows different residential preferences and tend to
reside in areas of more urban characteristics (Delmelle, 2015).
In terms of excess commuting metrics, the trends are generally consistent with these
commuting benchmarks. In Portland, Boston, and New Orleans, the new starts
subgroup demonstrates a relatively lower commuting efficiency than other subgroups
and the baseline scenario, implied by a higher Cex (48.3%-63.6%) and Cu (25.5%-
39.2%) as well as a lower Ce (25.2%-54.5%) and NCe (41.2%-65.2%). That said,
given the existing spatial distributions of jobs and housing, the new starts subgroup
across the cities, except for Detroit, is less sensitive to long commuting distance.
21
The socioeconomically disadvantaged subgroup is the neighborhood type featuring a
high concentration of high poverty, racial minority populations. While this subgroup
consists of several racial minority categories, our discussion is focused on the Black
and the Asian dominated neighborhoods (type 6 and type 9). In Portland, Boston, and
New Orleans, the high poverty Black neighborhoods (type 6) are highly clustered in
the central city close to the CBD, which is reflected by the right-skewed density
curves in Figure 6 across the four cities. This is consistent with existing segregation
research examining North American cities (Hartshorn, 1971; Hyra and Rugh, 2016).
The high poverty Asian dominated neighborhoods (type 9) are also confined to the
inner city, although not as obvious as type 6. In Detroit, type 6 is not only highly
clustered in the inner city but also more spatially isolated with little exposure to other
neighborhood types. Type 9 is mostly distributed in inner suburbs, in proximity to
their Asian counterpart with higher socioeconomic status (type 3). The spatial
distribution patterns of this socioeconomically disadvantaged subgroup largely
explain their commuting patterns. For example, values of both Cmin and Cmax of type 9
are smaller, indicating a greater degree of jobs-housing intermixing distributed within
a confined spatial extent. In comparison, type 6, though also highly clustered in inner
city, experiences greater jobs-housing separation, which is suggested by a moderate
Cmin and a much smaller Cmax, and a lower commuting capacity. This pattern is more
prominent for Boston and Portland, whose Cmin ranks the third highest among all
neighborhood types.
Another notable point associated with this subgroup relates to the actual commute
Cobs. As shown in Figure 3, both type 6 and type 9 commute shorter distances than
other subgroups across cities, which does not conform to the well-known spatial
mismatch hypothesis that minorities, particularly Blacks, are confronted with a higher
level of unemployment and longer commutes due to suburbanization and residential
segregation (Kain, 1968). Additionally, for type 6, its relatively shorter actual
commute and longer theoretical minimum commute compared to other neighborhood
types result in less excess commuting or greater commuting efficiency.
In addition to the three most representative neighborhood types across cities discussed
above, there are some exceptional observations about certain subgroups in certain
cities. For example, in Boston, neighborhood type 5 featuring high poverty, Hispanic
and Black populations have the largest Cmin (11 km), which is almost as twice as the
baseline scenario, indicating the poorest jobs-housing proximity. It also has a high
Cobs (17 km) second only to the traditional suburban subgroup. The greatest Cmin for
type 5 among neighborhood types yields a relatively efficient commute for this
neighborhood type (Cex: 32.1%); however, its relative efficiency is an outcome of
higher actual travel cost and, at the same time, greater minimum travel cost. The
double disadvantages make type 5 the most disadvantaged worker subgroup in terms
of commuting in Boston, which would have been overlooked by existing methods
treating all neighborhoods as a homogeneous whole group.
22
5. Conclusions
The existing methodology for measuring excess commuting has been criticized for the
rather strict assumption of homogenous housing and/or employment characteristics in
calculating the theoretical minimum (or maximum) commute, which may result in
biased estimates and erroneous policy implications. An emerging body of research has
attempted to tackle this issue by disaggregating commuter groups into rather
heterogeneous subgroups along certain socioeconomic dimensions, such as
occupation type, household structure, income, and age. However, most of these
disaggregation approaches only examined one single attribute. This may lead to
worker groupings that are not as heterogeneous as expected due to possible
interactions with other socioeconomic attributes, such as income and race/ethnicity
(Hu, 2021). This research contributes to the excess commuting literature by proposing
a methodology of measuring excess commuting across distinctive worker subgroups
disaggregated by residential neighborhood type, a more comprehensive factor
reflecting multiple socioeconomic profiles including housing condition, income level,
and race among others. This multi-variable-based disaggregation method is expected
to provide more accurate results because it allows for a more realistic residence-
swapping process by controlling for individuals’ residential location choices. Using
the 2006-2010 CTPP data and nine identified different neighborhood types in four
cities, including Portland, Boston, New Orleans, and Detroit, we reveal the inter-
neighborhood type variations of (excess) commuting patterns across cities and also
compare the results between the existing and our proposed methodology. Major
conclusions are discussed as follows.
It is found that, overall, the grouping of commuters into distinctive subgroups by
residential neighborhood type leads to greater minimum commute and less excess
commute for the general workers across the four cities. We additionally observe
remarkable inter-neighborhood type variations in (excess) commuting across cities,
which emphasizes the importance of promoting heterogeneous policy options across
different population subgroups. Based on the discovered inter-neighborhood type
variations, we further identified three representative worker subgroups with
distinctive commuting spatial patterns across cities, and they are the traditional
suburban subgroup (neighborhood type 2), new starts subgroup (type 3), and
socioeconomically disadvantaged subgroup (type 6 and type 9). While being similar
in socioeconomic characteristics, the traditional suburban subgroup and new starts
subgroup demonstrate different residential preferences and, as such, disparate
commuting patterns. Specifically, the traditional suburban subgroup has a longer
commute yet higher commuting efficiency, due to their largest commuting capacity.
The new starts subgroup, by contrast, commutes shorter distances but has lower
commuting efficiency due to their limited range of commute capacity, showing a
residential locational preference for the job-rich areas. In terms of the
socioeconomically disadvantaged subgroup highly concentrated in the inner city, the
high poverty Black neighborhood type experiences a greater level of jobs-housing
23
imbalance while at the same time has shorter actual commute, leading to less excess
commuting. This does not conform to the spatial mismatch hypothesis, and instead
suggests that perhaps other nonspatial factors may play a role in this process. Besides
these general patterns across the four cities, we also identified place-specific trends.
For example, the neighborhood type 5 in Boston serves as a special case with high
minimum commute as well as longer actual commute, which deserves more attention
in terms of policymaking. In addition, as the most acclaimed residential segregated
city, Detroit experiences much larger commuting length compared to other three
cities, especially for the socioeconomically disadvantaged neighborhoods in the inner
city (i.e., type 6 and type 9). The sharp contrast suggests that a higher level of
segregation might cause larger commuting burden as well as employment disparity.
This article also sheds lights on the comparisons among excess commuting metrics.
There is, perhaps, no single best metric for measuring commuting efficiency, and an
overreliance on a single metric, thus, may yield misleading results (Kanaroglou et al.,
2015). To better understand commuting efficiency, especially for comparative studies,
we find it critical to not only examine excess commuting metrics but also commuting
benchmarks. Since all excess commuting metrics only evaluate the difference
between the actual commute and a theoretical optimal (e.g., minimum) scenario, the
actual values of both actual and optimal commute are overlooked. In other words, a
seemingly efficient commuting pattern for a particular worker subgroup only suggests
this subgroup’s actual commute is close to the optimal value. But whether this
subgroup enjoys great job accessibility or suffers from poor job accessibility is
unclear from looking at excess commuting metrics alone. The high poverty Hispanic
& Black neighborhood type (type 5) in Boston, for instance, has the second lowest
excess commute across neighborhood types. But in fact, it has both high actual and
theoretical minimum commute, which essentially indicates large jobs-housing
separation and poor job accessibility for this subgroup. Taken together, it is
suggestive of a truly disadvantaged group in terms of job accessibility constrained by
both spatial and nonspatial factors. However, we would have failed to identify this
worker subgroup if only the excess commuting metrics were examined.
This research, however, has some limitations. First, as this study is based on census
data from 2010, the identified patterns may not be true to commuting patterns today
due to changes in urban form and neighborhood typologies over the last decade.
Future studies will apply the methods to most recent census data to better understand
the dynamics. Second, we only utilize network distances to measure commuting cost,
a metric often used to reflect the city spatial structure. Future studies may consider
actual travel times associated with different travel modes (Hu et al., 2020) to also
reveal the modal disparity across neighborhood types. Third, our analyses are subject
to the modifiable areal unit problem (Hu and Wang, 2018) due to the use of
aggregated data. It will be interesting to apply our methods on individual-level
datasets, if/when they are available, which allow for a more refined segmentation of
labor market. Lastly, while we find some significant discrepancies of commuting
24
outcomes across the four selected cities, further studies based on a larger sample of
cities are warranted to help us more fully disentangle the potential links between
residential segregation and excess commuting.
Acknowledgements
We thank Dr. Elizabeth Delmelle for sharing the neighborhood classification data
with us. We also thank the editor and the three anonymous referees for their valuable
comments that greatly improved the paper.
References
Antipova, A. (2020). Analysis of commuting distances of low-income workers in
Memphis metropolitan area, TN. Sustainability, 12(3), 1209.
Buliung, R. N., & Kanaroglou, P. S. (2002). Commute minimization in the Greater
Toronto Area: applying a modified excess commute. Journal of Transport
Geography, 10(3), 177-186.
Bwire, H., & Zengo, E. (2020). Comparison of efficiency between public and private
transport modes using excess commuting: An experience in Dar es Salaam. Journal of
Transport Geography, 82, 102616.
Charron, M. (2007). From excess commuting to commuting possibilities: more
extension to the concept of excess commuting. Environment and Planning A, 39(5),
1238-1254.
Chowdhury, T. A., Scott, D. M., & Kanaroglou, P. S. (2013). Urban form and
commuting efficiency: A comparative analysis across time and space. Urban
Studies, 50(1), 191-207.
Cropper, M. L., & Gordon, P. L. (1991). Wasteful commuting: a re-
examination. Journal of Uban Economics, 29(1), 2-13.
Delmelle, E. C. (2015). Five decades of neighborhood classifications and their
transitions: A comparison of four US cities, 1970–2010. Applied Geography, 57, 1-
11.
Delmelle, E. C. (2016). Mapping the DNA of urban neighborhoods: Clustering
longitudinal sequences of neighborhood socioeconomic change. Annals of the
American Association of Geographers, 106(1), 36-56.
Delmelle, E. C. (2017). Differentiating pathways of neighborhood change in 50 US
metropolitan areas. Environment and Planning A: Economy and Space, 49(10), 2402-
2424.
Finer, J. (2003). Boston’s racial barriers slow to fall, Washington Post, 16 September,
p. A2.
Frost, M., Linneker, B., & Spence, N. (1998). Excess or wasteful commuting in a
selection of British cities. Transportation Research Part A, 32(7), 529-538.
Green, A. E., Hogarth, T., & Shackleton, R. E. (1999). Longer distance commuting as
a substitute for migration in Britain: a review of trends, issues and
implications. International Journal of Population Geography, 5(1), 49-67.
25
Ha, J., Lee, S., & Kwon, S. M. (2021). Revisiting the relationship between urban form
and excess commuting in US metropolitan areas. Journal of Planning Education and
Research, 41(3), 294-311.
Hamilton, B. W., & Röell, A. (1982). Wasteful commuting. Journal of Political
Economy, 90(5), 1035-1053.
Hanlon, B., Vicino, T., & Short, J. R. (2006). The new metropolitan reality in the US:
Rethinking the traditional model. Urban Studies, 43(12), 2129-2143.
Hincks, S., Kingston, R., Webb, B., & Wong, C. (2018). A new geodemographic
classification of commuting flows for England and Wales. International Journal of
Geographical Information Science, 32(4), 663-684.
Horner, M. W. (2002). Extensions to the concept of excess commuting. Environment
and Planning A, 34(3), 543-566.
Horner, M. W., & Murray, A. T. (2002). Excess commuting and the modifiable areal
unit problem. Urban Studies, 39(1), 131-139.
Horner, M. W. (2008). ‘Optimal’ Accessibility Landscapes? Development of a New
Methodology for Simulating and Assessing Jobs—Housing Relationships in Urban
Regions. Urban Studies, 45(8), 1583-1602.
Horner, M. W., Schleith, D. K., & Widener, M. J. (2015). An analysis of the
commuting and jobs–housing patterns of older adult workers. The Professional
Geographer, 67(4), 575-585.
Hu, Y. (2021). The unequal commute: Comparing commuting patterns across income
and racial worker subgroups. Environment and Planning A: Economy and Space,
0308518X211068852.
Hu, Y., & Li, X. (2021). Modeling and analysis of excess commuting with trip
chains. Annals of the American Association of Geographers, 111(6), 1851-1867.
Hu, Y., & Wang, F. (2015). Decomposing excess commuting: A Monte Carlo
simulation approach. Journal of Transport Geography, 44, 43-52.
Hu, Y., & Wang, F. (2016). Temporal trends of intraurban commuting in Baton
Rouge, 1990–2010. Annals of the American Association of Geographers, 106(2), 470-
479.
Hu, Y., & Wang, F. (2018). GIS-based Simulation and Analysis of Intra-urban
Commuting. CRC Press.
Hu, Y., Wang, C., Li, R., & Wang, F. (2020). Estimating a large drive time matrix
between ZIP codes in the United States: A differential sampling approach. Journal of
Transport Geography, 86, 102770.
Hyra, D., & Rugh, J. S. (2016). The US great recession: Exploring its association with
black neighborhood rise, decline and recovery. Urban Geography, 37(5), 700-726.
Jing, Y., & Hu, Y. (2022). The unequal commuting efficiency: A visual analytics
approach. Journal of Transport Geography, 100, 103328.
Kain, J. F. (1968). Housing segregation, negro employment, and metropolitan
decentralization. The Quarterly Journal of Economics, 82(2), 175-197.
Kanaroglou, P. S., Higgins, C. D., & Chowdhury, T. A. (2015). Excess commuting: a
critical review and comparative analysis of concepts, indices, and policy
implications. Journal of Transport Geography, 44, 13-23.
26
Kim, K., & Horner, M. W. (2021). Examining the impacts of the Great Recession on
the commuting dynamics and jobs-housing balance of public and private sector
workers. Journal of Transport Geography, 90, 102933.
Kim, S. (1995). Excess commuting for two-worker households in the Los Angeles
metropolitan area. Journal of Urban Economics, 38(2), 166-182.
Korsu, E., & Le Néchet, F. (2017). Would fewer people drive to work in a city
without excess commuting? Explorations in the Paris metropolitan
area. Transportation Research Part A: Policy and Practice, 95, 259-274.
Logan, J. R., Xu, Z., & Stults, B. J. (2014). Interpolating US decennial census tract
data from as early as 1970 to 2010: A longitudinal tract database. The Professional
Geographer, 66(3), 412-420.
Ma, K. R., & Banister, D. (2006a). Excess commuting: a critical review. Transport
Reviews, 26(6), 749-767.
Ma, K. R., & Banister, D. (2006b). Extended excess commuting: a measure of the
jobs-housing imbalance in Seoul. Urban Studies, 43(11), 2099-2113.
Ma, K. R., & Banister, D. (2007). Urban spatial change and excess
commuting. Environment and Planning A, 39(3), 630-646.
Merriman, D., Ohkawara, T., & Suzuki, T. (1995). Excess commuting in the Tokyo
metropolitan area: measurement and policy simulations. Urban Studies, 32(1), 69-85.
Mikelbank, B. A. (2011). Neighborhood déjà vu: Classification in metropolitan
Cleveland, 1970-2000. Urban Geography, 32(3), 317-333.
Murphy, E. (2009). Excess commuting and modal choice. Transportation Research
Part A: Policy and Practice, 43(8), 735-743.
Morris, E. A., & Zhou, Y. (2018). Are long commutes short on benefits? Commute
duration and various manifestations of well-being. Travel Behaviour and Society, 11,
101-110.
Murphy, E., & Killen, J. E. (2011). Commuting economy: An alternative approach for
assessing regional commuting efficiency. Urban Studies, 48(6), 1255-1272.
Niedzielski, M. A. (2006). A spatially disaggregated approach to commuting
efficiency. Urban Studies, 43(13), 2485-2502.
Niedzielski, M. A., Horner, M. W., & Xiao, N. (2013). Analyzing scale independence
in jobs-housing and commute efficiency metrics. Transportation Research Part A:
Policy and Practice, 58, 129-143.
Niedzielski, M. A., Hu, Y., & Stępniak, M. (2020). Temporal dynamics of the impact
of land use on modal disparity in commuting efficiency. Computers, Environment and
Urban Systems, 83, 101523.
O'Kelly, M. E., & Lee, W. (2005). Disaggregate journey-to-work data: implications
for excess commuting and jobs–housing balance. Environment and Planning
A, 37(12), 2233-2252.
Schleith, D., Widener, M. J., Kim, C., & Horner, M. W. (2019). Categorizing urban
form for the largest metro regions in the US using the excessive commuting
framework. Built Environment, 45(4), 450-461.
27
Scott, D. M., Kanaroglou, P. S., & Anderson, W. P. (1997). Impacts of commuting
efficiency on congestion and emissions: case of the Hamilton CMA,
Canada. Transportation Research Part D, 2(4), 245-257.
Shen, Q. (2000). Spatial and social dimensions of commuting. Journal of the
American Planning Association, 66(1), 68-82.
Silverman, B.W. (1986) Density Estimation for Statistics and Data Analysis.
Chapman & Hall, London.
Small, K. A., & Song, S. (1992). " Wasteful" commuting: a resolution. Journal of
Political Economy, 100(4), 888-898.
Suzuki, T., & Lee, S. (2012). Jobs–housing imbalance, spatial correlation, and excess
commuting. Transportation Research Part A: Policy and Practice, 46(2), 322-336.
White, M. J. (1988). Urban commuting journeys are not" wasteful". Journal of
Political economy, 96(5), 1097-1110.
Yang, J. (2008). Policy implications of excess commuting: Examining the impacts of
changes in US metropolitan spatial structure. Urban Studies, 45(2), 391-405.
Yang, J., & Ferreira Jr, J. (2008). Choices versus choice sets: A commuting spectrum
method for representing job—housing possibilities. Environment and Planning B:
Planning and Design, 35(2), 364-378.
Yang, L., & Zhang, W. (2019). Evaluation of transport policy packages in the excess
commuting framework: The case of Xiamen, China. Cities, 87, 39-47.
Zheng, W., Wong, C., & Qiao, M. (2019). Socio-spatial variations in commuting
patterns in suburban Beijing. Built Environment, 45(4), 523-543.
Zhou, J., Murphy, E., & Long, Y. (2014). Commuting efficiency in the Beijing
metropolitan area: An exploration combining smartcard and travel survey
data. Journal of Transport Geography, 41, 175-183.
Zhou, J., & Murphy, E. (2019). Day-to-day variation in excess commuting: An
exploratory study of Brisbane, Australia. Journal of Transport Geography, 74, 223-
232.
Zhou, J., Murphy, E., & Corcoran, J. (2020). Integrating road carrying capacity and
traffic congestion into the excess commuting framework: the case of Los
Angeles. Environment and Planning B: Urban Analytics and City Science, 47(1), 119-
137.