Urban spatial structure, employment subcenters, and freight travel

Journal of Transport Geography 60 (2017) 267–276

Contents lists available at ScienceDirect

Journal of Transport Geography journal homepage: www.elsevier.com/locate/jtrangeo

Urban spatial structure, employment subcenters, and freight travel Marlon G. Boarnet a,⁎, Andy Hong b, Raul Santiago-Bartolomei a a b

Department of Urban Planning and Spatial Analysis, Sol Price School of Public Policy, University of Southern California, Lewis Hall 312, Los Angeles, California 90089–0626 Health and Community Design Lab, School of Population and Public Health, University of British Columbia, #372-2206 East Mall, Vancouver, BC V6T 1Z3

a r t i c l e

i n f o

Article history: Received 5 March 2016 Received in revised form 15 March 2017 Accepted 22 March 2017 Available online xxxx Keywords: Freight Land use Employment subcenters

a b s t r a c t Metropolitan areas in the U.S. have become increasingly polycentric. Large employment subcenters have emerged outside of central cities, competing against the traditional city center for labor and businesses. The existing literature on land use and transportation focuses on passenger travel, providing little insight into the impact of polycentric metropolitan development patterns on freight activity. In this study, we use the Los Angeles region as a case study to examine the relationship between urban spatial development patterns and freight travel. Using the National Employment Time Series (NETS) data, we identify employment subcenters in metropolitan Los Angeles. We characterize freight activities associated with the subcenters using data from the Southern California Association of Governments (SCAG). We develop a regression model that estimates freight activity as a function of geographic characteristics, such as whether a location is in an employment subcenter, measures of nearby employment, access to the highway network, and proximity to intermodal freight facilities. The results indicate that employment is an important driver of freight activity; however, employment subcenters have an independent effect on freight activity. The results of this study suggest that further research on urban spatial structure and freight activity should assess the effects of employment subcenters and how their particular employment composition and characteristics are associated with freight activities at the metropolitan level. Such an approach would lead to more precise policy recommendations for urban goods movement. © 2017 Elsevier Ltd. All rights reserved.

1. Introduction A half-century of dispersed spatial development has intensified polycentric urban spatial patterns. In major U.S. metropolitan areas, large population and employment subcenters have emerged outside of central cities, diminishing the role of the traditional city center as a destination for businesses. While service and financial industries are more likely to locate in the central city, manufacturing and warehousing industries have decentralized to suburbs because of lower land and transport costs (Glaeser and Kahn, 2001). Moreover, employment subcenters are transforming from “business only” districts into multi-use locales that often have residential, office, retail, light industrial, and warehousing uses in close proximity, competing for space on the same road network. This changing nature and context of urban development presents challenges to many businesses trying to optimize goods and service delivery within existing transportation networks. The previous literature on land use and transportation has focused on passenger travel (Bento et al., 2005; Boarnet and Crane, 2001; Boarnet and Sarmiento, 1998), providing little insight into the impact of polycentric metropolitan development patterns on freight activity. ⁎ Corresponding author. E-mail addresses: [email protected] (M.G. Boarnet), [email protected] (A. Hong), [email protected] (R. Santiago-Bartolomei).

http://dx.doi.org/10.1016/j.jtrangeo.2017.03.007 0966-6923/© 2017 Elsevier Ltd. All rights reserved.

There is evidence that suggests that urban spatial structure at the metropolitan level has significant impacts on passenger travel behavior (Badoe and Miller, 2000; Bento et al., 2005; Naess, 2003). However, as Rodrigue (2006a) and Hesse and Rodrigue (2004) have noted, freight transport and goods movement in an urban context have been understudied despite their increasing importance on the urban economy and geography. In particular, the relationship between employment subcenters and freight travel remains largely unexplored (Hesse and Rodrigue, 2004; Woudsma, 2001). The dearth of research on urban freight transport is unfortunate given increasing policy attention to a national freight network and its significant role as a driver of regional and national economic development (Kane and Tomer, 2015). In this study, we use the Los Angeles region as the case study to explore the relationship between urban spatial development patterns and freight travel. Los Angeles is the ideal place to study the relationship between metropolitan development patterns and freight activity because of its large number of employment subcenters compared to other metropolitan areas and the region's long history of dispersed urban spatial development (Giuliano and Small, 1999; Giuliano et al., 2007; Redfearn, 2007). We first identify subcenters in metropolitan Los Angeles using the National Employment Time Series (NETS), which has the location and industry code of all business establishments in the region. We characterize freight travel associated with major subcenters using data from the Southern California Association of

268

M.G. Boarnet et al. / Journal of Transport Geography 60 (2017) 267–276

Governments (SCAG), the metropolitan planning organization for the greater Los Angeles region. This research enables us to estimate how freight travel is associated with different employment centers, providing insights into relationships between land use, industrial structure, and the use of the road and highway system by freight. 2. Literature review 2.1. Polycentric urban model and subcenter formation The traditional model of urban spatial structure is the monocentric urban model which assumes that all jobs are located in the city center (Alonso, 1964; Mills, 1967; Muth, 1969). Recent work from urban economics and regional studies suggests that major American cities have become increasingly polycentric, with multiple employment centers dispersed across a typical metropolitan area (Anas et al., 1998; McDonald and McMillen, 1990). A definition of employment subcenters tends to vary from one city to another, but urban researchers have long sought to develop a robust method to identify employment subcenters. McDonald (1987) used a simple employment density function to identify employment subcenters in the Chicago metropolitan area. He defined subcenters as a zone whose measure of employment concentration is higher than all other zones in the surrounding area. McMillen (2001) and Craig and Ng (2001) used a similar approach using a nonparametric employment density function to identify subcenters. They identified subcenters as areas with high employment concentration where the estimated density function is increasing rather than decreasing with distance from the city center. For the Los Angeles region, Giuliano and Small (1999) developed a criteria to identify employment subcenters as a cluster of contiguous zones having a minimum employment density of 10 jobs per acre and total subcenter employment of at least 10,000 jobs. A series of follow-up studies was conducted to ensure that this cut-off point is robust and consistent over time (Giuliano et al., 2007; Redfearn, 2007). Previous literature suggests that job clusters emerge where a good labor force and transportation network exist (Giuliano and Small, 1999), however, what is still in need of further understanding is how travel behavior could be affected by the resulting changes in urban form. Firms locate near available labor supply and seek to achieve economies of scale, known as “agglomeration economies.” By locating close to each other, firms benefit from externalities of agglomeration economies, e.g. access to a large labor pool, specialized and skilled labor, knowledge spillovers, and input sharing (Giuliano et al., 2007; Puga, 2010). Businesses concentrate in space because of these agglomeration benefits, and the location choice of firms among these employment subcenters is influenced by the agglomeration economies/diseconomies in each subcenter, which in turn depend on the spatial distribution of production and consumption and the existing transportation network. With the exception of one TRB report (Bassok et al., 2013), most of the theoretical and empirical work on employment subcenters has been centered on the phenomenon itself with little discussion about how the changing urban spatial pattern has influenced travel behavior. This is an especially acute gap with regard to freight demand and movement at the metropolitan scale. 2.2. Determinants of freight activity In understanding freight travel, it is important to make a distinction between freight generation and freight trip generation (Holguín-Veras et al., 2014). While goods movement and freight distribution is increasingly being understood within the context of integrated freight demand (Hesse and Rodrigue, 2004), freight demand occurs when there is an economic activity pertaining to the production and consumption of goods. Generation of freight trips is the result of meeting this integrated freight demand by transporting goods between production, distribution, and consumption locations. Therefore, freight trip generation is

not only affected by the size of an establishment (Holguín-Veras et al., 2014) but also the size and the type of shipments being delivered (Sánchez-Díaz et al., 2014) as well as the freight distribution and transportation network (Hesse and Rodrigue, 2004). Previous literature has shown that freight trip generation is generally proportional to establishment size; however, there are large variations in freight trip generation between individual firms and the types of industry. Holguín-Veras et al. (2011) developed an ordinary least squares (OLS) model to predict freight trip generation using employment size as an independent variable at the disaggregate establishment level. The authors assumed that a firm decides the optimal shipment size and frequency of delivery that minimizes the corresponding transportation and inventory costs, and these logistic decisions may differ by industry sector. Using data from New York City, Holguín-Veras et al. (2011) have shown that freight trip generation is proportional to business size for only 18% of the industry sectors. Iding et al. (2002) developed a linear regression model for various sectors of industry using a large-scale survey conducted in the Netherlands. The results indicated that while freight trip generation is generally proportional to establishment size, a large variability exists in freight trip generation between individual firms and the types of industry. Sánchez-Díaz et al. (2014) explored the relationship between freight trip attraction and key features of the urban environment. Using 343 establishments in New York, the authors found that the establishment's location has a significant effect on freight trip generation. They found a significant autocorrelation in retail establishments, suggesting that location, e.g. proximity to large employment peers or high density retail establishments, plays an important role in attracting freight trips. Furthermore, Sánchez-Díaz et al. (2014) found that freight trip attraction is better modeled as a nonlinear function of employment and other locational variables. Taken together, these studies suggest that freight trip generation could be proportional to establishment size (as measured by employment), but the types of industry and the spatial clustering of firms in certain industries also play an important role in attracting freight travel. In addition to the freight demand caused by the direct outcome of economic activities, Rodrigue (2006b) has argued that freight transport should be understood as an integrated demand, recognizing the importance of underlying economic activities (e.g. employment, population, and income). While production and consumption of goods and services play an important role in generating basic demand for goods movement, recent decentralization of warehousing and trucking activity has increasingly shaped how goods movement and distribution operate in a changing micro- and macro-economic framework (Cidell, 2010; Dablanc, 2014). Much of this changing dynamic is characterized by globalization and complex supply chain management where freight transport and distribution are interdependent within the urban and regional economy (Hesse and Rodrigue, 2004; Rodrigue, 2006b). This changing notion of freight transport also resonates with the recent development in urban economics where understanding of urban spatial structure has changed from a monocentric model to a polycentric urban model. However, little effort has been made to understand urban freight movement within the broader context of changing urban spatial structure. A review of the previous literature indicates that most of the theoretical and empirical work on employment subcenters has been centered on either describing the patterns or identifying the causes of urban spatial structure. Likewise, the freight movement literature has largely focused on factors of freight trip generation from the perspective of firm-level logistic and business decisions. The changing nature of urban spatial structure, especially with regard to subcentering patterns of employment, has broader implications for production, consumption, and distribution of goods and services. However, urban spatial patterns and the transportation network have rarely been examined in relation to goods movement within metropolitan areas. This paper, to our knowledge, is the first attempt to understand urban goods movement

M.G. Boarnet et al. / Journal of Transport Geography 60 (2017) 267–276

from the perspective of polycentric urban development and the emergence of multi-nodal and multi-functional urban regional systems. 3. Data and methods 3.1. Employment subcenter data Our study area is the Los Angeles combined statistical area (33,954 mile2), which includes the counties of Los Angeles, Ventura, Orange, San Bernadino, and Riverside (Fig. 1), Los Angeles region, hereafter. We identified employment subcenters using employment data from the 2005 National Establishment Time Series (NETS) database, and projected the subcenters using 1 mile2 hexagons as the unit of analysis. The use of hexagons instead of traffic analysis zone (TAZ) or census tract allowed us to normalize land area with a uniform geographic shape. The NETS data contain the business name, address, total employment, and North American Industry Classification (NAICS) industry code of every firm in the region. We matched firms to a square mile hexagon, based on the firm's address, and used the square mile hexagons as the building blocks to identify subcenters. Only hexagons with the actual firm data from the NETS data set were matched, which essentially represent inhabitable land areas (6491 mile2) within the Los Angeles region (Fig. 1). Following the previous method (Giuliano and Small, 1999; Giuliano et al., 2007; Redfearn, 2007), we identified employment subcenters as a cluster of contiguous zones having a minimum employment density of 10 jobs per acre and total employment (for the sum of contiguous zones in the center) of at least 10,000 jobs (Fig. 2). There are a total of 53 employment subcenters in the region using the 2005 data, slightly more than 48 obtained for the year 2000 by Giuliano et al. (2012), but otherwise the locations of the sub-centers are very similar to earlier studies. Two subcenters are much larger than the others, both in land area and employment. These two subcenters, the largest being the Downtown Los Angeles-Wilshire Avenue-Santa Monica corridor and the second largest located in central Orange County, have a large

269

concentration of employment in professional services and appear qualitatively different from the other subcenters (Table 1). We used the 2005 data instead of the more recent 2010 NETS data to define sub-centers for two reasons. First, we prefer subcenter definitions based on a non-recession year. The steep recession that began in 2008 and which continued into 2010 and beyond, by reducing employment, may cause some smaller subcenters to drop below the 10,000 jobs threshold. Believing the effect of the recession to be temporary, we prefer subcenter definitions that will show clusters of economic activity based on a pre-recession definition. Second, and relatedly, we believe it is desirable to use employment data that precedes the time period for our freight data because the formation of employment subcenters typically gives rise to economic activities and sparks movement of goods and people. Because the freight data are for the year 2008, the 2005 NETS data was a logical choice given our focus on the relationship between freight activity and employment subcenters. Nonetheless, at the risk that the 2008 recession could significantly affect the implications of this research, Table 1 presents a brief comparison of the employment subcenters in the area of study for 2005 and 2009. During this period, the total number of employment subcenters was reduced from 53 to 46, while the two largest subcenters underwent a substantial increase in surface area, especially the second largest subcenter, although this was not the case for the remaining subcenters. In terms of employment, however, the change is much less pronounced for all subcenters, reflecting some degree of stability in terms of economic activity in the region. 3.2. Freight transportation data The freight data were obtained from the Southern California Association of Governments (SCAG). SCAG developed the heavy duty truck (HDT) model primarily using heavy duty truck trip data collected by Cambridge Systematics. The freight data consist of link-based truck flow data (a total of 68,324 valid links) for the Los Angeles region. We

Fig. 1. Study area showing the hexagons with employment data. (Source: Authors' Analysis of 2005 National Establishment Time Series).

270

M.G. Boarnet et al. / Journal of Transport Geography 60 (2017) 267–276

Fig. 2. Employment subcenters in the Los Angeles region. (Source: Authors' Analysis of 2005 National Establishment Time Series).

calculated freight travel distance as the main outcome variable by multiplying the link-based truck flow by the length of each link. Aggregate freight vehicle kilometers travelled (VKT) was then calculated by summing up all freight travel distance for all links within each employment hexagon. The aggregate value represents the total freight distance travelled within each employment hexagon. The calculation of the freight VKT variable is shown in the following equation. n

Freight VKT by truck type ¼ ∑ V i;k  li i

Where, Vi,k is per link daily truck volume for k type of truck, li refers to length of each link i in km, k denotes types of freight truck: 1 = all types; 2 = light duty, 3 = medium duty, 4 = heavy duty, i refers to each freight flow link, and n is total number of links within one employment hexagon. The freight data obtained from the SCAG are the best data on truck activities in the Los Angeles region. The freight data are an estimation of truck activities from a freight model based on multiple data sources,

including proprietary truck surveys, truck GPS, commodity flow surveys, and port terminal truck surveys. Because it is modelled data, we validated the SCAG data by comparing it against the annual average daily truck traffic (AADT) from a truck survey conducted by the California Department of Transportation (Caltrans). The truck AADT data were obtained from a manual truck count, which includes a partial day, 24-h, 7-day and continuous vehicle classification counts conducted annually on all highways in the State of California. The partial day and 24-h counts are conducted on high volume, urban highways, and the 7-day counts are on low volume, rural highways. The resulting counts were adjusted to an estimate of annual average daily truck traffic by accounting for seasonal and weekly variation. Because the Caltrans data are limited to trucks on highways, we computed an average truck AADT for each employment hexagon using the Caltrans data and compared the resulting values against the average SCAG freight flow and truck VKT on highways only (excluding arterial freight traffic) for each matching hexagon. The SCAG data and the Caltrans data are positively correlated (0.65) at the 5% significance level. We also performed a log- and square root-transformation of

Table 1 Subcenter summary statistics. Total employment Subcenter

2005

2009

1 2 Other subcentersa

1,091,789 563,287 38,008

1,107,139 605,284 36,791

a

Values in row correspond to the average in category.

Share of total employment Percent change 1.41% 7.46% −3.20%

2005

2009

30.38% 15.68% 1.06%

33.24% 18.17% 1.10%

Surface area in square miles

Percentage point change

2005

2009

Percent change

2.86% 2.49% 0.04%

62 39 3.63

69 49 3.82

11.29% 25.64% 5.23%

M.G. Boarnet et al. / Journal of Transport Geography 60 (2017) 267–276

both SCAG data and Caltrans measures to satisfy the normality assumption of Pearson's correlation. The correlation analysis using the transformed data indicate that the Caltrans truck AADT data are highly correlated with the SCAG truck VKT data, with a correlation coefficient of 0.76 at the 5% significance level. This suggests that the SCAG freight data are a reasonable data source for our study, providing good estimates of both intra- and inter-regional freight activities in the Los Angeles region. Another reason to favor the SCAG freight data over the Caltrans truck AADT data is that, even though the Caltrans data reflect freight activity after the 2008 recession, it does so partially. Due to how sparsely the data are collected at highways only, using the Caltrans data would reduce the number of observations from 5054 to 369 employment hexagons, a total reduction of 93% of the hexagons were we to use the Caltrans AADT data. Table 2 is a descriptive summary of daily truck flow and VKT data. The truck flows and VKT were grouped into three truck types: lightheavy (8500–14,000 lbs. gross vehicle weight, GVW); medium-heavy (14,001–33,000 lbs. GVW); and heavy-heavy (N33,000 lbs. GVW). Heavy-duty truck type has the highest mean daily truck flow per link (mean truck flow = 621), indicating that heavy-duty trucks make up the majority of the SCAG freight travel data. The truck flow data are highly skewed to the right, with a disproportionately large volume of data clustered at zero. This suggests that many links have little or no truck flow present on a typical day but some links contain very high truck activities, with daily mean values reaching up to 22,776 trucks per day. The mean daily truck VKT per hexagon is about 12,700 km. Again, the heavy-duty trucks account for most of the daily truck VKT. Similar to the flow data, the mean VKT values have a skewed distribution, with most values clustered at zero. Given that heavy-duty trucks make up the majority of the freight travel data, we limit our analysis to VKT from trucks of this size and exclude the observations for light and medium-duty trucks. 3.3. Intermodal freight facility data Intermodal freight transport utilizes two or more modes to form an integrated freight movement chain (Lowe, 2005). Intermodal freight facilities serve as important transfer stations between truck trailers or cargo containers and rail lines. These facilities typically consist of a rail yard, a container yard depot, a trucking terminal, and a warehousing facility. Large cargo containers transported by rail or truck are temporarily stored until they get shipped to other locations. Import/export goods are usually transported by truck, and interstate goods are transported to other cities by rail or truck. Because intermodal facilities act as a natural hub for heavy duty trucks, they are important factors for freight movement at the metropolitan level. We obtained geocoded intermodal facility data from the 2011 National Transportation Atlas Database (NTAD) maintained by U.S. Bureau of Transportation Statistics, a nationwide geographic database of transportation facilities, transportation networks, and associated infrastructure. This dataset includes spatial information for transportation modal networks and intermodal terminals, as well as the related

271

attribute information for these features. According to the NTAD data, there are about 93 intermodal freight terminals of various sizes and functions in the Los Angeles region. Of these terminals, we selected the seven largest intermodal terminals that operate rail-to-truck and truck-to-rail transloading facilities. Four of them are operated by Union Pacific Railroad, and three are operated by the competing Burlington Northern/Sante Fe Railroad. Although the NTAD data are from 2011, we confirmed that the seven major terminals included in our analysis existed before 2005 based on the California Air Resources Board's enforcement document on major rail yards in California (California Air Resources Board, 2005). 3.4. Regression model development In order to assess the effect that employment subcenters have on freight activity, we developed an ordinary least squares (OLS) model with heteroskedasticity robust standard errors. This entails discerning between the effect from employment subcenters and total employment in a hexagon and adjacent hexagons, while controlling for other land use variables such as the presence of freeways. We use hexagons as the unit of analysis. By limiting the analysis to those hexagons that registered employment, we used the 6491 mile2 hexagons as the unit of analysis. This also allows us to normalize effects by land area. The model was set up as follows: m

p

j¼1

k¼1

Y i ¼ β0 þ ∑ β j X ij þ ∑ βk Z ik þ ε i Where Yi refers to total VKT from heavy-duty trucks in each hexagon i, β0 is the intercept, Xij refers to the set of variables j that pertain to employment subcenters and total employment for each hexagon i, Zik refers to the set of other control variables related to land use and urban form characteristics in each hexagon i, and ε refers to the error term. We estimated 4 regressions that reflect different levels of disaggregation of the variables in the model above. The first regression is the base model that would estimate the simplest relation between truck VKT and the explanatory variables. Independent variables include the level of employment and its square value in each hexagon, since employment and freight activity have been found to have a non-linear relationship (Sánchez-Díaz et al., 2014). The regression also includes binary variables to identify the presence of highways and whether a hexagon is in an employment subcenter. In addition, to see how different industry sectors affect freight activity, we included the share of total employment in each hexagon that is in different industry sectors using the NAICS 2-digit codes. We focused particularly on employment in agriculture, construction, manufacturing, mining, professional services, retail, transportation, utilities and wholesale as independent variables for two reasons: (1) to reduce the possibility of collinearity in the model and (2) because this set of industries provides an opportunity to distinguish the potential effect that labor intensive and capital intensive industries have on freight activity. Finally, an additional control variable was added to measure the

Table 2 Summary of daily truck flow and VKT. Measures

Na

Mean

SD

Median

Daily flow (# of trucks) Light-duty flow Medium-duty flow Heavy-duty flow Daily truck VKT (distance in kilometers) Light-duty VKT Medium-duty VKT Heavy-duty VKT

68,968 68,968 68,968 68,968 5,609 5,609 5,609 5,609

972.95 196.86 154.75 621.35 12,793.91 2,299.82 1,743.93 8,750.16

2,446.20 402.02 321.15 1,760.23 26,710.76 4,061.74 3,177.89 20,002.08

189 62 44 77 1,910.26 554.72 378.43 903.32

Min 0 0 0 0 0 0 0 0

Max 22,776 3,512 2,729 17,397 559,342 50,969 31,410 476,962

a The unit of analysis for the truck flow is link; whereas for the truck VKT it is hexagons that have employment and freight links (Source: Authors' analysis of 2008 SCAG baseline freight data). The difference in the number of hexagons between this table and Table 3 is due to the fact that the number of hexagons in Table 3 excludes those with 0 VKT.

272

M.G. Boarnet et al. / Journal of Transport Geography 60 (2017) 267–276

effect from the distance to intermodal facilities from the centroid of a hexagon. Intermodal facilities are areas dedicated to the transshipment of freight cargo and are expected to be related to freight activity on nearby links. The effect from these facilities was represented with a continuous variable that measures the linear distance between each hexagon centroid and the nearest intermodal facility (from among the seven largest such facilities in the region). In the second regression, we substituted the subcenter binary variable with binary variables that specify if each hexagon is located within subcenters 1 or 2. As previously mentioned, the two largest subcenters are much larger than the others, both in land area and employment. Therefore, we divided the subcenter variable into three groups: 1) a binary variable indicating whether a hexagon is in the largest subcenter, located in the Downtown Los Angeles-Wilshire Avenue-Santa Monica corridor; 2) a binary variable indicating whether a hexagon is in the second largest subcenter, located in central Orange County; and 3) a binary variable indicating whether a hexagon is in any of the other subcenters excluding the largest two subcenters. The freeway binary variable was

also replaced with binary variables that specify whether there are 1 or 2 freeways present in each hexagon; no hexagon had 3 freeways or more within it. Regressions 3 and 4 add variables that measure the effect of employment in adjacent hexagons and thresholds in distance to intermodal facilities on total daily freight VKT in each hexagon. Total employment in adjacent hexagons and its squared value were also included in the model to consider employment clustering effects. Regression 4 also includes binary variables that specify whether each hexagon is located within distance bands of 1 mile to 10 miles from one of the seven largest intermodal facilities. Table 3 shows a complete description of the variables included in the regressions, and summary statistics for these variables are provided in Table 4. We excluded hexagons that have zero VKT from the analysis because these are far from sub-centers and, as a result, their relation to sub-centers should be different. Therefore, we focus on the effect of sub-centers conditional on having any VKT, to avoid issues of how zero VKT hexagons are likely quite different from positive VKT hexagons. Also, due to the overrepresentation of heavy trucks in the SCAG

Table 3 Regression variable descriptions. Variable name

Description

Variable type

Source

VKT Subcenter Subcenter_one Subcenter_two Subcenter_other

Daily vehicle kilometers travelled for heavy-duty trucks Indicates if hexagon is located in an employment subcenter Indicates if hexagon is located in employment subcenter 1 Indicates if hexagon is located in employment subcenter 2 Indicates if hexagon is located in an employment subcenter other than subcenter 1 or 2 Total employment in each hexagon in thousands of jobs Square of total employment per hexagon in thousands of jobs Total employment in adjacent hexagons in thousands of jobs Square of total employment in adjacent hexagons in thousands of jobs Share of total hexagon employment classified by NAICS (code = 11) as agriculture, forestry, fishing, and hunting Share of total hexagon employment classified by NAICS (code = 23) as construction Share of total hexagon employment classified by NAICS (code = 31–33) as manufacturing Share of total hexagon employment classified by NAICS (code = 21) as mining activities Share of total hexagon employment classified by NAICS (code = 54) as professional, scientific, and technical services Share of total hexagon employment classified by NAICS (code = 44–45) as retail trade Share of total hexagon employment classified by NAICS (code = 48–49) as transportation and warehousing Share of total hexagon employment classified by NAICS (code = 22) as utilities Share of total hexagon employment classified by NAICS (code = 42) as wholesale trade Presence of freeway in each hexagon Presence of only one freeway in each hexagon Presence of only two freeways in each hexagon Linear distance between hexagon centroid and intermodal facility (miles) Indicates whether distance between hexagon and intermodal facility is less than or equal to 1 mile Indicates whether distance between hexagon and intermodal facility is N1 mile and less than or equal to 2 miles Indicates whether distance between hexagon and intermodal facility is N2 miles and less than or equal to 3 miles Indicates whether distance between hexagon and intermodal facility is N3 miles and less than or equal to 4 miles Indicates whether distance between hexagon and intermodal facility is N4 miles and less than or equal to 5 miles Indicates whether distance between hexagon and intermodal facility is N5 miles and less than or equal to 6 miles Indicates whether distance between hexagon and intermodal facility is N6 miles or less than or equal to 7 miles Indicates whether distance between hexagon and intermodal facility is N7 miles or less than or equal to 8 miles Indicates whether distance between hexagon and intermodal facility is N8 miles or less than or equal to 9 miles Indicates whether distance between hexagon and intermodal facility is N9 miles or less than or equal to 10 miles

Continuous Binary; (1) if hexagon is located in subcenter and (0) if it is not Binary; (1) if hexagon is located in subcenter 1 and (0) if it is not Binary; (1) if hexagon is located in subcenter 2 and (0) if it is not Binary; (1) if hexagon is located in subcenter other than subcenters 1 or 2 and (0) if it is not Continuous Continuous Continuous Continuous Continuous

SCAG 2008 NETS 2005 NETS 2005 NETS 2005 NETS 2005 NETS 2005 NETS 2005 NETS 2005 NETS 2005 NETS 2005

Continuous

NETS 2005

Continuous

NETS 2005

Continuous

NETS 2005

Continuous

NETS 2005

Continuous

NETS 2005

Continuous

NETS 2005

Continuous

NETS 2005

Continuous

NETS 2005

Binary; (1) for presence of freeway and (0) for no freeway Binary; (1) for presence of one freeway and (0) otherwise Binary; (1) for presence of two freeways and (0) otherwise Continuous Binary; (1) if distance is less than or equal to 1 mile and (0) if it is not

SCAG 2008 SCAG 2008 SCAG 2008 NTAD 2011 NTAD 2011

Binary; (1) if distance is N1 mile and less than or equal to 2 miles and (0) if it is not Binary; (1) if distance is N2 miles and less than or equal to 3 miles and (0) if it is not Binary; (1) if distance is N3 miles and less than or equal to 4 miles and (0) if it is not Binary; (1) if distance is N4 miles and less than or equal to 5 miles and (0) if it is not Binary; (1) if distance is N5 miles and less than or equal to 6 miles and (0) if it is not Binary; (1) if distance is N6 miles and less than or equal to 7 miles and (0) if it is not Binary; (1) if distance is N7 miles and less than or equal to 8 miles and (0) if it is not Binary; (1) if distance is N8 miles and less than or equal to 9 miles and (0) if it is not Binary; (1) if distance is N9 miles and less than or equal to 10 miles and (0) if it is not

NTAD 2011

emp emp_sq tot_adj_emp tot_adj_emp_sq perc_Agri perc_Cons perc_Manu perc_Mini perc_Prof perc_Reta perc_Tran perc_Util perc_Whol FWY one_FWY two_FWY Dist_int_fac_mi dist_one dist_two dist_three dist_four dist_five dist_six dist_seven dist_eight dist_nine dist_ten

NTAD 2011 NTAD 2011 NTAD 2011 NTAD 2011 NTAD 2011 NTAD 2011 NTAD 2011 NTAD 2011

M.G. Boarnet et al. / Journal of Transport Geography 60 (2017) 267–276 Table 4 Regression variable summary statistics, for N = 5054 hexagons with VKT N 0 for heavy trucks, freight link, and employment. Variable name

Mean

Median

VKT Subcenter Subcenter_one Subcenter_two Subcenter_other emp emp_sq tot_adj_emp tot_adj_emp_sq perc_Agri perc_Cons perc_Manu perc_Mini perc_Prof perc_Reta perc_Tran perc_Util perc_Whol FWY one_FWY two_FWY Dist_int_fac_mi dist_one dist_two dist_three dist_four dist_five dist_six dist_seven dist_eight dist_nine dist_ten

9,711.048 0.125 0.019 0.014 0.092 1.710 17.659 9.925 361.648 0.039 0.101 0.076 0.003 0.073 0.105 0.035 0.006 0.053 0.265 0.138 0.126 26.185 0.018 0.026 0.028 0.030 0.032 0.033 0.033 0.035 0.034 0.033

1,148.028 0.000 0.000 0.000 0.000 0.349 0.121 3.235 10.462 0.000 0.050 0.024 0.000 0.048 0.070 0.007 0.000 0.024 0.000 0.000 0.000 20.780 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Minimum 0.892 0.000 0.000 0.000 0.000 0.001 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.096 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

Maximuma

Std. Dev.

476,962.460 1.000 1.000 1.000 1.000 122.073 14,901.817 244.303 59,683.956 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 168.914 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000

20,849.310 0.331 0.137 0.117 0.290 3.839 229.303 16.223 1,869.504 0.149 0.159 0.138 0.042 0.100 0.138 0.097 0.048 0.105 0.441 0.345 0.332 23.182 0.134 0.160 0.164 0.172 0.175 0.178 0.179 0.183 0.181 0.179

a A value of 1 (100%) for share of employment of each NAICS industry corresponds a low number of firms (never greater than 4) identified in hexagons where this value was registered.

freight data, we limited the analysis to the VKT generated by heavy trucks. While the results for all freight are qualitatively similar to those reported here for heavy trucks (available upon request from authors), we note that SCAG's freight model was designed for heavy duty trucks, and hence a heavy-duty truck analysis uses what is likely the most reliable part of the SCAG data. 3.5. Visualization of freight activities To visualize the relationship between freight travel and employment subcenters, we used linear interpolation to create a statistical surface of freight flow and overlaid employment subcenters on top of the interpolated surface. Previous studies have used similar interpolation techniques to estimate traffic intensity based on known traffic count data (Selby and Kockelman, 2013; Wang and Kockelman, 2009). To perform interpolation, we first converted the link-based freight flow data into point data using the centroid of the link as the geometric location of the point. These converted points were linearly interpolated using the inverse distance weighting (IDW) method, a deterministic approach to interpolate unknown points by assigning higher values for points close to known points and lower values for points far from the known points, hence the name inverse distance weighting. The resulting surface from this interpolation is presented in the Results and discussion section.

273

subcenter have, on average, approximately 1800 more daily heavyduty truck VKT, controlling for other factors, although this is significant at the 90% level. Also, a 1000 job increase in total hexagon employment is associated with a 264 daily VKT increase. In terms of the effect of employment share among industry sectors, larger shares in manufacture, retail, transportation, and wholesale have positive and statistically significant effects on hexagon daily VKT (wholesale being significant at the 90% level), while the employment shares in agriculture and professional services have negative and statistically significant effects at the 90% and 95% level, respectively. This suggests that freight activity is positively related to more capital intensive industries (e.g. manufacturing and retail) while land intensive industries (e.g. agriculture) or knowledge-based industries (e.g. professional services) are negatively associated with freight activity. The model suggests that the presence of a freeway in a given hexagon would increase daily freight VKT in the hexagon by 21,000 km. Results also show that linear distance to the nearest intermodal facility does not have a statistically significant effect on daily freight VKT. Compared to the first model, the second model has an improved model fit, with the R2 changing from 0.23 to 0.39. The model shows a negative association between subcenter 1 and freight VKT, with subcenter 1 being significant at the 90% level, while subcenter 2 is not statistically significant. Being located in any other subcenter, however, is associated with a significant increase in daily freight VKT of 2800 km. This confirms that subcenters 1 and 2 are different from the other subcenters in their relationship to heavy duty freight VKT. In addition, the presence of only one freeway in a hexagon is associated with an average increase in daily heavy duty freight VKT of approximately 36,700 km, while the presence of two freeways is associated with a lower average increase of approximately 4500 km. The effect of hexagon employment levels (Emp) is not statistically significant in Regression 2. The effect of employment share observed in the previous regression are similar here, with the exception that employment shares in construction now has a negative statistically significant effect at the 90% confidence level. The employment shares in agriculture, professional services, and wholesale are no longer significant. The association between freight VKT and linear distance with the nearest intermodal facility is also not statistically significant in this regression. In regressions 3 and 4, variables measuring total employment and the square of employment in adjacent hexagons are added to the model, but neither have a statistically significant association with the heavy duty freight VKT dependent variable. The effect on daily VKT from a hexagon being located in subcenter 1 is now negative and statistically significant at the 95% level or better in Regressions 3 and 4, suggesting that the larger share of employment in professional services in this subcenters or other characteristics of that centers (the downtown orientation) is associated with a reduction in freight activity when compared to the other subcenters. The other subcenters (subcenters 3 through 53) have a consistently positive relationship with freight VKT (Table 5). In terms of the effect of intermodal facilities, Regression 4 shows that hexagons that are within one mile from these facilities are associated with an increase in daily freight VKT of approximately 6000 km. The effects from employment share and presence of freeways remained similar in Regressions 3 and 4. The results of these regressions give no evidence for the effect of employment beyond the one-mile hexagon on heavy duty freight VKT, and the effect of intermodal facilities on freight VKT is confined to one mile as indicated by the results in Regression 4. 4.2. Visualization of freight activities

4. Results and discussion 4.1. Regression results Table 5 presents the results of the linear regression model. The first regression model shows that hexagons located within an employment

Fig. 3 shows an interpolation of the daily freight flow for all truck types, using the inverse distance weighting method. Areas not included in the map were left out of the image due to space limitation, but were not excluded from the analysis. Freight activities are generally high on freeway networks and concentrated in the Long Beach and Los Angeles

274

M.G. Boarnet et al. / Journal of Transport Geography 60 (2017) 267–276

Table 5 OLS regression coefficients of the freight VKT, heavy duty trucks. (1) Const Subcenter

(2)

1,958.16⁎⁎ (936.07) 1,832.83⁎ (945.78)

Subcenter_one Subcenter_two Subcenter_other Emp emp_sq

(3) 2,057.76⁎⁎⁎ (889.28)

2,556.98⁎⁎⁎ (626.52)

−3,709.18⁎ (1,911.17) −1,415.75 (1,396.53) 2,815.58⁎⁎⁎

−5,670.28⁎⁎⁎ (2,080.80) −2,598.49 (1,610.88) 2,499.13⁎⁎

−5,453.46⁎⁎ (2,072.94) −1,986.61 (1,595.97) 2,593.26⁎⁎

(984.45) 103.39 (109.85) 0.51 (0.96)

(1,076.2) 35.99 (123.72) 0.51 (1.02) 0.30 (40.27) 0.33 (0.26) −982.95 (1,156.3) −2,019.04⁎

−2,552.54⁎ (1,400.16) −2,344.83 (1,518.49) 6,363.01⁎⁎⁎

−1,103.15 (1,120.36) −2,079.98⁎

(1,076.13) 14.97 (123.82) 0.69 (1.01) 24.69 (39.81) 0.22 (0.25) −1,034.83 (1,133.15) −1,983.27⁎

(1,177.92) 3,843.81⁎⁎

(1,179.15) 3,946.01⁎⁎

(1,177.46) 3,109.79⁎

(1,802.42) 11,278.20 (8,416.10) −4,214.90⁎⁎ (2,186.11) 10,224.00⁎⁎

(1,659.40) 8,986.79 (6,010.78) −1,513.10 (1,842.38) 7,499.53⁎

(1,656.71) 8,984.09 (6,018.78) −1,520.13 (1,843.46) 7,520.08⁎

(1,665.42) 8,925.51 (5,975.20) −1,431.23 (1,840.87) 7,557.48⁎

(4,722.23) 11,951.80⁎⁎⁎ (4,350.51) −3,225.39 (5,104.80) 3,702.25⁎

(4,372.84) 10,411.60⁎⁎⁎ (3,737.09) −97.29 (4,053.26) 2,558.43 (1,958.98)

(4,382.73) 10,351.9⁎⁎⁎ (3,736.64) 3.69 (4,055.26) 2,438.11 (1,958.87)

(4,455.08) 9,046.04⁎⁎

0.19 (16.88)

36,794.80⁎⁎⁎ (1,254.64) 4,566.22⁎⁎⁎ (755.57) 8.17 (15.17)

36,779.4⁎⁎⁎ (1,273.27) 4,573.04⁎⁎⁎ (756.49) 10.81 (15.60)

5,054 0.2327 0.2306 71.42

5,054 0.3896 0.3875 82.77

5,054 0.3902 0.3879 77.94

263.98⁎⁎ (112.79) −2.05 (1.51)

tot_adj_emp tot_adj_emp_sq perc_Agri perc_Cons perc_Manu perc_Mini perc_Prof perc_Reta perc_Tran perc_Util perc_Whol FWY

(2,200.75) 21,053.40⁎⁎⁎ (868.05)

one_FWY two_FWY Dist_int_fac_mi

(4)

2,234.23⁎⁎⁎ (850.26)

dist_one dist_two dist_three dist_four dist_five dist_six dist_seven dist_eight dist_nine dist_ten Observations R-squared Adj. R-squared F statistic Standard errors in parenthesis. ⁎ p ≤ 0.1. ⁎⁎ p ≤ 0.05. ⁎⁎⁎ p ≤ 0.01.

(3,893.16) 33.53 (4,068.35) 2,002.45 (1,941.47)

36,789.30⁎⁎⁎ (1,288.31) 4,622.01⁎⁎⁎ (754.57)

5,978.89⁎⁎ (2,490.18) 185.40 (1,675.61) 936.84 (1,422.48) −1,895.57 (1,301.56) −1,281.34 (1,290.28) 1,061.34 (1,414.28) −525.38 (1,060.91) −662.73 (1,119.25) 600.34 (1,110.28) −224.77 (1,102.03) 5,054 0.3921 0.3887 55.11

M.G. Boarnet et al. / Journal of Transport Geography 60 (2017) 267–276

Downtown areas. The combination of the Port of Los Angeles and the Port of Long Beach is one of the top ten container port complexes in the world by volume. The nearby Long Beach Freeway is well integrated with the port facilities, providing direct connections among the ports, distribution centers, and the nearby intermodal facilities. The port and the Downtown areas, being part of the regional distribution centers, typically generate a substantial volume of freight traffic. Also these locations are close to intermodal terminal facilities which serve as the major loading and distribution center for both domestic and international containers. Other locations with high freight activities include the City of Commerce and the City of Industry. These locations are regional centers of employment, characterized by a mix of warehousing and manufacturing industries. The heat map shows a less apparent relationship between employment subcenters and freight activities. Some employment subcenters are located in areas with low freight activities. For example, the subcenter that stretches from the downtown Los Angeles to Santa Monica contains low freight activities. Likewise, subcenters located in Santa Ana and central Orange County have low to moderate freight activities. It is likely that some of these large subcenters are better suited for service industries which may not generate significant freight demand. Subcenters with a greater proportion of manufacturing industries may generate more freight demand and freight traffic. 5. Conclusion The analysis presented in this paper adds to the findings from previous research conducted at the firm level. Like in previous research, we found employment to be associated with freight activity. However, rather than focusing on employment at the level of the firm, we examined the spatial distribution of employment, other geographic characteristics, and their relationship to freight activity.

275

Freight VKT is larger when a hexagon is within a mile of an intermodal facility, but the effect does not persist over longer distances. This suggests that freight activity is dispersed over the freeway and road network, a finding reinforced by the heat map shown in Fig. 3. Note that the presence of a freeway in a hexagon is approximately six times as large as the coefficient on being within a mile of an intermodal freight facility. From a policy perspective, this implies that any negative externalities associated with freight travel are more likely associated with freeways than with the intermodal facilities that are transshipment hubs. We are not suggesting to ignore the impacts of intermodal facilities, but the regressions in Table 4 suggest that highway access can be a larger determinant of freight VKT. Looking more broadly at the economic geography of the region, employment shares in a hexagon are associated with freight VKT in the same hexagon in ways that are generally expected, with industries that are associated with production or sale of goods being associated with higher freight flows. Employment subcenters, the primary motivation for our study, are independently associated with increased freight VKT, while the two largest subcenters in the region have the opposite effect or no association with heavy duty freight VKT. This suggests that there is differentiation in the economic function and hence in the goods movement characteristics of different employment subcenters. This latter point is a topic that we suggest is ripe for further research. Our findings suggest directions for both future research and policy analysis. In terms of research, we note that the economic geography of a region clearly influences freight flows, and analyzing aggregate flows is an important topic for freight research. While that may seem obvious, for decades the literature on passenger transportation has moved from aggregate to disaggregate analyses, to illuminate behavioral elements that are obscured by aggregate, zone level, studies. Both aggregate and disaggregate analyses will be important, but we note that

Fig. 3. Heat map of freight flow, all truck types, using inverse distance weighting method. *Intermodal facility in San Bernardino is not shown in this map due to space limitation. (Source: 2008 SCAG baseline freight data; 2005 National Establishment Time-Series data. For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article)

276

M.G. Boarnet et al. / Journal of Transport Geography 60 (2017) 267–276

understanding how freight travel relates to economic clustering and urban form can be improved by aggregate (zone-level) analyses, because policy questions about freight are often explicitly geographic in nature. Given the intricate connection between urban geography and freight activity, planners and policy-makers need to assess how urban freight travel will be linked to changes in land use or infrastructure at what is usually the level of a small geographic zone and larger metropolitan geographies. The spatial pattern of employment concentrations and the highway network clearly matter for freight travel, providing some intuitive first steps to predicting zone-level freight VKT based on the geography of a region. While that is a start, future research should also consider factors which were outside the scope of this study, such as how zoning decisions and historic development patterns are associated with freight VKT. Freight VKT is tied to the economic geography of a metropolitan area. The results of this research indicate that the two largest employment subcenters are associated with lower freight VKT or are not associated with freight VKT levels, while other subcenters are associated with higher freight VKT. The magnitudes of the coefficients in Table 5 suggest that the effect of employment concentration in a tertiary sub-center (centers 3 through 53) could be approximately half the size of the effect of being within a mile of an intermodal facility. One implication is that policy-makers should be more alert to the ways that the spatial distribution of employment shapes freight travel patterns. The policy focus should broaden beyond the more traditionally obvious intermodal facilities and highways to consider how development, including the spatial pattern of employment, is associated with freight travel patterns. The results in this paper are a start, and policy-makers would benefit from additional model building that examines the association between freight VKT and the geography of employment. One final point is that our analysis is pre-recessionary because of data constraints and, even though the contrast of available data pre and post-recession suggest that the findings are relevant for post-recession dynamics, it is possible that other regions in the US could have had larger changes in their pattern of employment subcenters as a result of the 2008–2010 economic recession. We suggest future research in other metropolitan areas and other time periods, to examine the generalizability of the relationship that we found in Los Angeles. Acknowledgements We are grateful for funding support from the METRANS Transportation Center at USC (53-5701-6001), through the University Transportation Centers (UTC) program. Funding for this research, through the UTC program, was provided by the California Department of Transportation (65A0533 TO 001). The National Employment Time Series data used in this research were used under license from Walls and Associates. The opinions, findings, and conclusions in this paper are the authors alone and do not necessarily reflect positions of the funding entities. References Alonso, W., 1964. Location and Land Use. Harvard University Press, Cambridge, MA. Anas, A., Arnott, R., Small, K.A., 1998. Urban Spatial Structure. J. Econ. Lit. 36, 1426–1464. Badoe, D., Miller, E., 2000. Transportation–land-use interaction: empirical findings in North America, and their implications for modeling. Transp. Res. Part D: Transp. Environ. 5, 235–263. Bassok, A., Johnson, C., Kitchen, M., Maskin, R., Overby, K., Carlson, D., Goodchild, A., McCormack, E., Wygonik, E., 2013. NCFRP Report 24 - Smart Growth and Urban Goods Movement. Transportation Research Board, Washington, D.C. Bento, A.M., Cropper, M.L., Mobarak, A.M., Vinha, K., 2005. The effects of urban spatial structure on travel demand in the United States. Rev. Econ. Stat. 87, 466–478.

Boarnet, M.G., Crane, R., 2001. The influence of land use on travel behavior. Transp. Res. A Policy Pract. 35. Boarnet, M.G., Sarmiento, S., 1998. Can land-use policy really affect travel behaviour? A Study of the Link between Non-work Travel and Land-use Characteristics. Urban Stud. 35:1155–1169. http://dx.doi.org/10.1080/0042098984538. California Air Resources Board, 2005. ARB/Railroad Statewide Agreement: Particulate Emissions Reduction Program at Califronia Rail Yards. Cidell, J., 2010. Concentration and decentralization: the new geography of freight distribution in US metropolitan areas. J. Transp. Geogr. 18:363–371. http://dx.doi.org/10. 1016/j.jtrangeo.2009.06.017. Craig, S.G., Ng, P.T., 2001. Using quantile smoothing splines to identify employment subcenters in a multicentric urban area. J. Urban Econ. 49:100–120. http://dx.doi.org/10. 1006/juec.2000.2186. Dablanc, L., 2014. Logistics sprawl and urban freight planning issues in a major gateway city: the case of Los Angeles. In: Gonzalez-Feliu, J., Semet, F., Routhier, J.-L. (Eds.), Sustainable Urban Logistics: Concepts, Methods and Information Systems, EcoProduction. Springer, Berlin Heidelberg, Berlin, Heidelberg:pp. 49–69 http://dx. doi.org/10.1007/978-3-642-31788-0. Giuliano, G., Small, K.A., 1999. The determinants of growth of employment subcenters. J. Transp. Geogr. 7, 189–201. Giuliano, G., Redfearn, C., Agarwal, A., Li, C., Zhuang, D., 2007. Employment concentrations in Los Angeles, 1980–2000. Environ. Plan. A 39:2935–2957. http://dx.doi.org/10. 1068/a393. Giuliano, G., Redfearn, C., Agarwal, A., He, S., 2012. Network accessibility and employment centres. Urban Stud. 49:77–95. http://dx.doi.org/10.1177/0042098011411948. Glaeser, E.L., Kahn, M.E., 2001. Decentralized employment and the transformation of the American City (no. working paper 8117). NBER Working Paper No. 8117. National Bureau of Economic Research, Cambridge, MA. http://dx.doi.org/10.1353/urb.2001. 0008. Hesse, M., Rodrigue, J.P., 2004. The transport geography of logistics and freight distribution. J. Transp. Geogr. 12:171–184. http://dx.doi.org/10.1016/j.jtrangeo. 2003.12.004. Holguín-Veras, J., Jaller, M., Destro, L., Ban, X. (Jeff), Lawson, C., Levinson, H.S., 2011. Freight generation, freight trip generation, and perils of using constant trip rates. Transp. Res. Rec. 2224:68–81. http://dx.doi.org/10.3141/2224-09. Holguín-Veras, J., Jaller, M., Sánchez-Díaz, I., Campbell, S., Lawson, C.T., 2014. Freight generation and freight trip generation models. In: Tavasszy, L., Jong, G. de (Eds.), Modelling Freight Transport. Elsevier, Waltham, MA:pp. 43–63 http://dx.doi.org/10.1016/ B978-0-12-410400-6.00003-3. Iding, M.H., Meester, W.J., Tavasszy, L., 2002. Freight trip generation by firms. Pap. 42nd Eur. Congr. Reg. Sci. Assoc. Kane, J., Tomer, A., 2015. FAST act ushers new era for U.S. freight policy [WWW document]. Brookings Inst. URL. http://www.brookings.edu/blogs/the-avenue/posts/ 2015/12/28-fast-act-freight-policy-kane-tomer (accessed 1.1.15). Lowe, D., 2005. Intermodal Freight Transport. Elsevier Butterworth-Heinemann, Oxford, UK. McDonald, J.F., 1987. The identification of urban employment subcenters. J. Urban Econ. 21, 242–258. McDonald, J.F., McMillen, D.P., 1990. Employment subcenters and land values in a polycentric urban area: the case of Chicago. Environ. Plan. A 22:1561–1574. http://dx. doi.org/10.1068/a221561. McMillen, D.P., 2001. Nonparametric employment subcenter identification. J. Urban Econ. 50:448–473. http://dx.doi.org/10.1006/juec.2001.2228. Mills, E.S., 1967. An aggregative model of resource allocation in a metropolitan area. Am. Econ. Rev. 57, 197–210. Muth, R., 1969. Cities and Housing. University of Chicago Press, Chicago, IL. Naess, P., 2003. Urban structures and travel behaviour. Experiences from empirical research in Norway and Denmark. Eur. J. Transp. Infrastruct. Res. 3, 155–178. Puga, D., 2010. The magnitude and causes of agglomeration economies. J. Reg. Sci. 50, 203–219. Redfearn, C.L., 2007. The topography of metropolitan employment: identifying centers of employment in a polycentric urban area. J. Urban Econ. 61, 519–541. Rodrigue, J.-P., 2006a. Transport geography should follow the freight. J. Transp. Geogr. 14: 386–388. http://dx.doi.org/10.1016/j.jtrangeo.2006.06.003. Rodrigue, J.-P., 2006b. Challenging the derived transport-demand thesis: geographical issues in freight distribution. Environ. Plan. A 38:1449–1462. http://dx.doi.org/10. 1068/a38117. Sánchez-Díaz, I., Holguín-Veras, J., Wang, X., 2014. An exploratory analysis of spatial effects on freight trip attraction. Transportation (Amst) http://dx.doi.org/10.1007/ s11116-014-9570-1. Selby, B., Kockelman, K.M., 2013. Spatial prediction of traffic levels in unmeasured locations: applications of universal kriging and geographically weighted regression. J. Transp. Geogr. 29:24–32. http://dx.doi.org/10.1016/j.jtrangeo.2012.12.009. Wang, X., Kockelman, K.M., 2009. Forecasting network data spatial interpolation of traffic counts using texas data. Transp. Res. Rec. 100–108. http://dx.doi.org/10. 3141/2105-13. Woudsma, C., 2001. Understanding the movement of goods, not people: issues, evidence and potential. Urban Stud. 38:2439–2455. http://dx.doi.org/10.1080/00420980120094605.