- Email: [email protected]
Effects of additional capacity on vehicle kilometers of travel in the U.S.: Evidence from National Household Travel Surveys
Journal of Transport Geography 66 (2018) 1–9
Contents lists available at ScienceDirect
Journal of Transport Geography journal homepage: www.elsevier.com/locate/jtrangeo
Effects of additional capacity on vehicle kilometers of travel in the U.S.: Evidence from National Household Travel Surveys
MARK
Edmund J. Zolnik Schar School of Policy and Government, George Mason University, 3351 Fairfax Drive, MS 3B1, Arlington, VA, 22201, USA
A R T I C L E I N F O
A B S T R A C T
Keywords: Induced demand Vehicle kilometers of travel Capacity Multilevel models
Adding capacity is one policy mechanism to alleviate congestion. However, the empirical evidence strongly suggests that additional capacity only makes congestion worse. This study analyzes the differential effects of additional freeway capacity versus additional arterial capacity on vehicle kilometers of travel (VKT) in metropolitan areas across the U.S. The analysis uses vehicle data and household data from the 2001 and the 2009 National Household Travel Surveys (NHTS) and includes stock and flow measures of road capacity, road congestion, commuter demand, and economic growth for metropolitan areas. Taking into account differences between metropolitan areas on each measure, the study adopts a novel multilevel model approach to estimate how additional capacity affects VKT. Results indicate that adding more arterial capacity slightly decreases VKT over a lag period from six years (1995 to 2001) to eight years (2001 to 2009), probably because adding arterials shortens routes between origins and destinations more so than adding freeways. Consistent with expectations, VKT is lower in more congested metropolitan areas, and in metropolitan areas that got more congested. Results also indicate that rebound effects (higher fuel-economy vehicles are driven much more than lower fuel-economy vehicles) will at least partially offset the demand management benefits of (gasoline) price sensitivity (higher gasoline prices decrease VKT).
1. Introduction Congestion is a major transportation problem in metropolitan areas across the U.S. There are many different strategies to mitigate congestion, including adding more capacity (freeways and arterials). Whichever strategy is adopted, it is important to account for all of the potential impacts. For example, in the case of adding more capacity, it is important to account for indirect and long-term impacts of induced demand for private-vehicle travel. In order to account fully for such impacts, this study explores the differential effects of additional freeway capacity versus additional arterial capacity on private-vehicle travel demand in metropolitan areas across the U.S. in 2001 and in 2009. Freeways and arterials are complementary functional types of capacity in the road network, but the differential effects of additional capacity are important to explore for many reasons. First, the economic returns on different functional types of capacity are not the same. Returns for interstate highways are greater than returns for non-interstate major roads (arterials and collectors). Likewise, returns for the latter are greater than returns for local roads (Jiwattanakulpaisarn et al., 2012). Second, the effects of different functional types of capacity on vehicle miles of travel (VMT), the most common measure of privatevehicle travel demand in the U.S. (Rentziou et al., 2012), are not the
same. For example, the travel-time benefits of new collectors are greater than the travel-time benefits of new interstates or new arterials (Noland, 2001). Third, the effect of additional capacity is not the same in different metropolitan areas. That is, the effect tends to be greater in metropolitan areas where the percentage increase in capacity (freeway and arterial) is larger (Noland and Cowart, 2000). The organization of the paper is as follows. The next section briefly reviews the induced demand literature. The data section lists the data sources for the vehicle level, the household level, and the metropolitan area level, respectively. The methodology section introduces the multilevel model in the study and also hypothesizes the effects of the vehicle-level, the household-level, and the metropolitan area-level independent variables. The results section presents the outcomes of the multilevel model. The discussion section focuses on the policy implications of the results. Finally, the conclusions section highlights the contributions of the results to the induced demand literature, as well as the most fruitful direction for future research. 2. Background Economic theory on induced demand—additional capacity attracts more traffic (AASHO, 1957)—suggests that additional capacity
E-mail address: [email protected]. https://doi.org/10.1016/j.jtrangeo.2017.10.020 Received 15 January 2015; Received in revised form 6 October 2017; Accepted 26 October 2017 0966-6923/ © 2017 Elsevier Ltd. All rights reserved.
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
Table 1 Key literature on induced demand. Author(s), year
Area(s)
Downs, 1962 Hills, 1996 Goodwin, 1996 Noland and Lem, 2002 Hansen and Huang, 1997 Noland and Cowart, 2000 Noland, 2001
US UK UK UK/US
Methodology
Results Law of peak-hour traffic congestion—traffic and congestion are in equilibrium. Additional capacity induces demand by changing network accessibility. Demand induced by 10% in the short term and by 20% in the long term. Lane-mile elasticities range from +0.3 to +0.6 regardless of the data or the methodology.
FE
Marginal effect of lane-miles of state highway on VMT is greatest in largest CMSAs/MSAs.
US
C/CEA/(C) MSA MSA
1938–1994 1994–1998 1993–2001 1973–1990
Theoretical Theoretical Review Review
1982–1996
2SLS
US
State
1984–1996
FE/SURE
CA
Unit(s)
Year(s)
Cervero and Hansen, 2002 Hymel et al., 2010 Duranton and Turner, 2011 Su, 2011
CA
C/CEA
1976–1997
SE
Lane-mile additions account for 15% of annual VMT growth with great variation between MSAs. Induced demand effects of lower functional types of new capacity (collectors) is greater than the induced demand effects of higher functional types of new capacity (interstates and arterials). Induced demand effect of +0.6 and induced investment effect of +0.3.
US US
State MSA
1966–2004 1983/1993/2003
3SLS 3SLS
Total road length induces demand by 3.7% in the short run and by 18.6% in the long run. VKT increases proportional to additional interstate capacity.
US
State
2001–2008
DPD
Rentziou et al., 2012
US
State
1998–2008
SURE/RE
Short-run and long-run rebound effects of 3% and 11%, respectively. Short-run and long-run road capacity per capita effects of 7% and 26%, respectively. VMT elasticity with respect to lane-miles higher on urban versus rural roads.
Area abbreviations: CA = California; UK = United Kingdom; and US = United States. Unit abbreviations are: C/CEA = County/County Equivalent Area; (C)MSA = (Consolidated) Metropolitan Statistical Area. Methodology abbreviations are: 2SLS = Two-Stage Least Squares; 3SLS = Three-Stage Least Squares; DPD = Dynamic Panel Data; FE = Fixed Effects; RE = Random Effects; SE = Simultaneous Equations; and SURE = Seemingly Unrelated Regression Equations.
capacity. For example, increases in traffic on the new routes are not offset by decreases in traffic on the old routes that provide alternatives either on a short-term or a long-term basis. The latter empirical result contradicts Downs' (1962) theory that older commuter routes will experience less congestion. A review of the empirical literature on induced demand from the U.K. and from the U.S. by Noland and Lem (2002) showed that lane-mile elasticities range from +0.3 to +0.6, regardless of the data or the methodology. The most methodologically sophisticated citations in the induced demand literature also account for the rebound effect (Jevons, 1865; Khazzoom, 1980; Alcott, 2005; Sorrell, 2007). The rebound effect is “the interaction of energy use with the efficiency of energy use: lower the energy required to do something, and you will do a bit more of that thing” (Schipper, 2000, 351). The most conservative estimates of the short-term (one-year) rebound effect for private vehicles from the literature are about 10% (Greening et al., 2000; Sorrell et al., 2009; Chakravarty et al., 2013). However, the least conservative estimates of the long-term rebound effect for private vehicles are about 30%. Such a range of estimates on the magnitude of the rebound effect is a source of controversy in academic and in policy circles (Tierney, 2011; Turner, 2013). Some argue that the rebound effect is in decline (Small and Van Dender, 2007) or that it is just a distraction to divert attention from efforts to enact stricter energy-efficiency regulations (Gillingham et al., 2013). Others argue that energy efficiency standards like Corporate Average Fuel Economy (CAFE) standards in the U.S. are not cost effective because of the rebound effect (Frondel and Vance, 2013) and, perhaps, greater energy efficiency leads to backfire—more usage offsets any energy savings due to efficiency improvements (Jenkins et al., 2011; Tierney, 2011). Regardless of the academic and the policy debates on the magnitude of the rebound effect, two examples from the induced demand literature (Hymel et al., 2010; Su, 2011) confirm that the rebound effect affects private-vehicle travel demand. In order to heed Hills' (1996) call to specify a realistically complex mathematical model of induced demand, the study pools disaggregate cross-sectional data from the National Household Travel Survey (NHTS) for two time points (2001 and 2009) to account for short-term and for long-term travel behavior changes in response to additional capacity. Barr (2000) also uses disaggregate cross-sectional data from an older version of the NHTS, known as the Nationwide Personal Transportation Survey, to model induced demand, but only for one time point (1995).
increases traffic because travel times decrease. The effect on travel times is due to the following (short-term and/or long-term) behavioral changes in response to less congestion:
• taking different routes for the same trip; • making the same trips at different times of the day; • using different modes for the same trip; • substituting destinations for the same (shopping or recreational) trip; and • making more trips (TRB, 1995). Table 1 summarizes the key literature on induced demand, and the citations here are by no means exhaustive. Rather, the citations highlight the trajectory of the literature from the theoretical to the empirical, with the trend toward methodological sophistication in the latter. The law of peak-hour traffic congestion (Downs, 1962, 393) states that “congestion rises to meet maximum capacity.” The law assumes the following with regard to commuter decision making:
• commuters seek to minimize travel times cognizant of income, monetary cost, place of residence, and personal comfort constraints; • the law of inertia rules—unless events in the environment compel change mode choices and route choices are habitual; events in the environment which compel mode choice and route • choice changes are those that decrease travel times; and • commuters are of two types—those who are passive and those who are active in seeking out different routes to save themselves time. Hills (1996) spelled out exactly what is meant by induced demand in order to distinguish generated traffic from induced traffic and to relate the latter phenomenon to the range of travel behavior responses to additional capacity. The cumulative route choices of all private-vehicle drivers generate traffic on a network whereas the addition of capacity to a network, by design, changes accessibility and induces traffic.
A review of the extant empirical literature on induced demand by Goodwin (1996) showed that additional capacity induces demand by 10% in the short term and by 20% in the long term. However, induced demand is dependent on the context, size, and location of the new
2
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
Table 2 Descriptive statistics for vehicle-, household-, and (C)MSA-levels of analysis. Level
n
Vehicle
130,489
Variable
Category
Dependent VKT Independent Age (years) Emissions (kilograms) Gasoline price ($ per liter) Mileage (kilometers per liter) Type
Automobile Van Sports Utility Vehicle Pickup Truck 2001 2009
Year Household
Mean
SD
18,392.51
15,959.37
9.20 18,497.21 0.78 8.82 55.93% 8.51% 20.15% 15.42% 16.23% 83.77%
6.75 15,706.25 0.10 2.60
66,445 Independent Income Less than $25,000 $25,000 to $49,999 $50,000 to $74,999 $75,000 to $99,999 Greater than or Equal to $100,000
10.81% 22.04% 18.59% 17.10% 31.46% 2.55
1.22
Arterial Arterial Change Freeway Freeway Change
6,988 682 2,380 179
6,466 648 2,073 177
Travel Time Index Travel Time Index Change
1.22 0.04
0.07 0.02
Commuters (1,000) Commuters Change (1,000)
1,159 197
1,074 169
Gross Metropolitan Product Gross Metropolitan Product Change
110.5 25.1
94.7 21.7
Yes No
35.98% 64.02%
Northeast Midwest South West
14.50% 8.40% 45.03% 32.07%
Vehicle count (C)MSA
44 Independent Capacity (lane-kilometers)
Congestion
Demand
Growth ($1,000,000,000)
Rail
Region
VKT is vehicle kilometers of travel.
empirical literature suggests that the induced demand effect is not the same at the metropolitan area level across the U.S. (Noland and Cowart, 2000). The induced demand effect is not homogeneous because travel behavior responses are not directly attributable to the additional capacity; rather, such responses are directly attributable to the resultant congestion reduction from the additional capacity (Litman, 2001). Besides congestion, Handy and Boarnet (2014) suggest that the induced demand effect may vary nationally by the size of the metropolitan area and by gasoline price. A random effects model (Duncan and Jones, 2000) of induced demand allows for variation in travel behavior responses to additional capacity at the metropolitan area level across the U.S.—the intercepts for each metropolitan area vary randomly. Besides the advantage of specifying a more realistically complex model of induced demand that nests travel behavior within metropolitan areas, another advantage is that a random effects specification accounts for the independence of observations. If observations within metropolitan areas are not independent but are correlated, then the assumption that the observations are independent and identically distributed is problematic. A multilevel model adds a second variance term to account for the nesting of observations and so extends the random part of the
The use of disaggregate data is important because most mathematical models of induced demand effects in the economics literature use aggregate cross-sectional data or aggregate cross-sectional/time-series data. The overreliance on aggregate data draws criticism from transportation planners who argue that the focus ought to be on how drivers respond to additional capacity so as to capture behavioral changes best and, ultimately, understand induced demand effects (Noland and Lem, 2002). To that end, the use of disaggregate data in this study contributes to a better balance in the empirical literature on induced demand. The study also contributes to the trend toward more methodological sophistication in the empirical literature on induced demand via the adoption of a novel multilevel approach (Noland, 2001). Such an approach is novel because mathematical models in the empirical literature on induced demand tend to assume that the effect is fixed at the metropolitan area level—the intercepts for each metropolitan area are independently fixed while the slopes are the same across metropolitan areas. In other words, travel behavior responses to additional capacity are the same at the metropolitan area level across the U.S. (Noland, 2001). However, a fixed effect specification is not realistic because the 3
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
Fig. 1. Historical vehicle kilometers of travel. USDOT, 2017.
gasoline prices in dollars per liter are real 2009 gasoline prices in dollars per liter. Mileage is fuel economy in kilometers per liter. Type is vehicle type: automobile (automobile, car, or station wagon; van (cargo, mini, or passenger); SUV; or pickup truck. Year is zero if the vehicle is from the 2009 NHTS subsample, the larger of the two subsamples also known as the referent category, one otherwise. Data at the household level (n = 66,445) include information on the characteristics of each household (Table 2). Total income is the total household income in dollars. Vehicle count is the number of household vehicles. Data at the (C)MSA level (n = 44) include stock and flow measures of road capacity, road congestion, commuter demand, and economic growth as well as commuter rail and census region (Table 2). The stock versions are for 2001 and 2009, respectively. The flow versions are differences from 1995 to 2001 and from 2001 to 2009, respectively, reflective of the time gap between successive NHTSs. The measures of road capacity are the lane-kilometers of freeways and the lane-kilometers of arterials from the Highway Performance Monitoring System (HPMS) (TTI, 2012). The measure of road congestion is the Travel Time Index (TTI) (TTI, 2014). The TTI is the ratio of travel times under peak travel conditions to travel times under nonpeak travel conditions (Schrank et al., 2012). The numerator and the denominator of the TTI are in units of time, so the ratio is unitless. This allows for the comparison of trips of different lengths to estimate how peak conditions affect travel times. Commuter demand is the number of private-vehicle commuters from the NHTS (TTI, 2014). Economic growth is the gross metropolitan product (GMP) in billions of dollars (USCM, 2000, 2003, 2010). GMP is the market value of all final goods and services produced within a (C)MSA in a given year. The flow version of GMP for 2001 is the difference from 1997 to 2001, not from 1995 to 2001, due to data availability limitations. Commuter rail is one if the (C)MSA has commuter rail, zero otherwise. The census regions are the Northeast, Midwest, South, and West regions of the U.S. The next section introduces the multilevel model in the study.
model (Bullen et al., 1997). The following section lists the data sources for the vehicle level, the household level, and the metropolitan area level, respectively. 3. Data Data are from the 2001 and the 2009 NHTS, respectively (USDOT, 2001, 2009). The 2001 and 2009 NHTS are cross-sectional surveys of travel behavior in the U.S. The sample sizes for the respective vehicle and household files in the 2001 NHTS are 139,382 and 69,817 respectively, and the sample sizes for the respective vehicle and household files in the 2009 NHTS are 309,163 and 150,147. In order to nest vehicles accurately within households in Consolidated Metropolitan Statistical Areas (CMSA) or Metropolitan Statistical Areas (MSA), hereafter known as a (C)MSA, those with missing data at the vehicle level, the household level, or the (C)MSA level are excluded from the respective subsamples. Application of the above selection criterion to the 2001 NHTS left a subsample of 21,179 vehicles nested within 11,032 households nested within 44 (C)MSAs. Application of the above selection criterion to the 2009 NHTS left a subsample of 109,310 vehicles nested within 55,413 households nested within 44 (C)MSAs. Pooling the 2001 subsample with the 2009 subsample nests 130,489 vehicles within 66,445 households within 44 (C)MSAs. To be clear, the 44 (C)MSAs in the 2001 subsample and the 44 (C)MSAs in the 2009 subsample are the same. Data at the vehicle level (n = 130,489) include information on each household's vehicles (Table 2). Vehicle kilometers of travel (VKT) is the best estimate of the kilometers each vehicle was driven over the past twelve months. Age is vehicle age in years. Emissions are annual carbon dioxide (CO2) emissions in kilograms (USEPA, 2014). Gasoline price is an estimate of the price of gasoline in dollars per liter for each vehicle (car, van, sports utility vehicle (SUV), or pickup truck) from the Energy Information Administration (USEIA, 2014). In order to harmonize gasoline prices for 2001 and for 2009 (Stern, 1984), nominal 2001 4
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
(Raudenbush and Bryk, 2002). Within each household, vehicle-level VKT is a function of vehicle-level independent variables plus an individual-level error term:
Table 3 Regression coefficient estimates for vehicle-, household-, and (C)MSA-levels of analysis from full, random-intercepts models with STOCK and FLOW versions of capacity, congestion, demand, and growth independent variables. Variable
Category
Vehicle Age (years) Emissions (kilograms) Gasoline Price ($ per liter) Mileage (kilometers per liter) Type Automobile Van Sports utility vehicle Pickup truck
Yvhm = β0 hm + β1 hm W1 vhm + …+βAhm WAvhm + rvhm
FLOW Coefficient (SE)
−11.98 (1.73)a +0.96 (0.001)a −2,277.43 (480.37)a +1,429.62 (4.85)a
−11.95 (3.31)a +0.96 (0.01)a −2,899.31 (1,493.15)c +1,429.41 (31.79)a
Referent −673.83 (39.84)a −1,807.82 (30.94)a −2,008.66 (33.94)a
Referent −676.21 (61.94)a −1,807.74 (59.77)a
where Yvhm is the VKT for vehicle v in household h in (C)MSA m; β0hm is the y-intercept term for household h in (C)MSA m; βAhm are a = 1, …, A vehicle-level coefficients; WAvhm is the independent variable a for vehicle v in household h and (C)MSA m; and rvhm is the vehicle-level random effect term. At level 2, the variation between households is a function of household-level independent variables plus a household-level error term:
−2,006.88 (68.01)a
β0 hm = π00 m + π01m X1 hm + …+π0 Cm XChm + e0 hm
−672.28 (120.61)a Referent
−824.79 (356.32)b Referent
Household Income Less than $25,000 $25,000 to $49,999 $50,000 to $74,999 $75,000 to $99,999 Greater than or equal to $100,000 Vehicle Count (C)MSA Intercept Capacity (lane-kilometers) Arterial Arterial change Freeway Freeway change Congestion Travel time index
+280.69 +251.35 +166.79 +152.36 Referent
(39.55)a (30.68)a (31.29)a (31.68)a
+282.46 +253.41 +169.17 +153.97 Referent
(35.74)a (37.58)a (39.10)a (28.69)a
+29.56 (9.22)a
+29.67 (9.69)a
+18,961.38 (28.86)a
+19,035.00 (49.04)a
At level 3, the variation between (C)MSAs is a function of (C)MSAlevel independent variables plus a (C)MSA-level error term:
π00m = γ000 + γ001 Z1m + …+γ00B ZBm + u 00m
(3)
where γ000 is the y-intercept term for (C)MSA m; γ00B are b = 1, …, B (C)MSA-level coefficients; ZBm are (C)MSA-level independent variables; and u00m is the (C)MSA-level random effect term.
+0.02 (0.01) −0.06 (0.03)c +0.003 (0.01)
In the multilevel models the regression coefficients at level 1 and at level 2 are fixed. These three-level models are known as random-intercepts models. The random-intercepts models estimate how household characteristics such as total income and metropolitan area characteristics such as road capacity affect vehicle-level VKT. The following subsection hypothesizes on the effects of the vehiclelevel, the household-level, and the (C)MSA-level independent variables.
−0.23 (0.17) −1,714.99 (291.76)a −2,494.51 (813.32)a
Travel time index change Demand Commuters (1,000) Commuters change (1,000) Growth ($1,000,000,000) Gross metropolitan product Gross metropolitan product change Rail Yes No Region Northeast Midwest South West
(2)
where π00m is the y-intercept term for (C)MSA m; π0Cm are c = 1, …, C household-level coefficients; XChm is the independent variable c for household h in (C)MSA m; and e0hm is the household-level random effect term.
Year 2001 2009
(1)
STOCK Coefficient (SE)
−0.17 (0.06)a −0.20 (0.12)
+0.81 (0.35)
4.1. Hypothesized effects of vehicle-, household-, and (C)MSA-level independent variables
b
At the vehicle level, less fuel-efficient, older vehicles will be driven less than more fuel-efficient, newer vehicles. Consistent with the trend toward greater price sensitivity (Litman, 2013), the effect of gasoline prices will be negative. Consistent with the rebound effect where use increases with greater energy efficiency (Schipper, 2000) the effect of mileage will be positive. Higher fuel-economy automobiles will be driven more than lower fuel-economy vans, SUVs, or pickup trucks. VKT will be higher in 2009 than in 2001 even though aggregate demand in 2009 is below the historical trend (Fig. 1). At the household level, the empirical evidence strongly suggests that VKT increases with income (Liddle, 2009). The relationship between VKT and vehicle count is expected to be positive because the number of vehicles per household grew slightly from the 2001 NHTS subsample (2.52) to the 2009 NHTS subsample (2.56). At the (C)MSA level, VKT is expected to be higher in (C)MSAs that have more freeway lane-kilometers and more arterial lane-kilometers and in (C)MSAs that have added more freeway lane-kilometers and added more arterial lane-kilometers (Noland and Cowart, 2000). VKT is expected to be lower in more congested (C)MSAs and in (C)MSAs that got more congested. VKT is also expected to be lower in (C)MSAs that have more commuters and in (C)MSAs that added more commuters due to congestion effects (Handy and Boarnet, 2014). VKT is expected to be
+2.54 (1.18)b
+259.36 (44.45)a Referent
+58.69 (48.91) Referent
+24.04 (46.56) −5.99 (42.93) Referent +207.81 (48.75)a
+110.69 (74.76) −39.21 (58.01) Referent +145.22 (103.60)
Dependent variable is VKT. Referent category represents the highest-frequency category. Standard errors appear in parentheses. STOCK versions of capacity, congestion, demand, and growth independent variables at (C)MSA level are for 2001 and for 2009, respectively. FLOW versions of capacity, congestion, demand, and growth independent variables at (C)MSA level are differences from 2001 to 2009, respectively. a Indicates significance at 99% confidence level. b Indicates significance at 95% confidence level. c Indicates significance at 90% confidence level.
4. Methodology The VKT multilevel models in the study are three-level models of vehicles (v) nested within households (h) nested within (C)MSAs (m) 5
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
Fig. 2. Map of residuals from the full, random-intercepts model with STOCK versions of capacity, congestion, demand, and growth independent variables at (C)MSA level.
year regression coefficient is negative; VKT is at least −670 km lower in 2001 than in 2009. The results at the household level are generally inconsistent with the hypothesized effects. Inconsistent with empirical evidence, VKT decreases with income; it is highest in the lowest-income category (less than $25,000). The effect of vehicle count on VKT is positive, as expected; a one standard deviation increase in vehicle count (1.22) increases VKT by about +40 km. The results at the (C)MSA level are somewhat consistent with the hypothesized effects. VKT is lower, not higher, in (C)MSAs that have added more arterial lane-kilometers; a one standard deviation increase in the change in arterial lane-kilometers (648 lane-kilometers) decreases VKT by about −40 km. VKT is indeed lower in more congested (C)MSAs and in (C)MSAs that got more congested; a one standard deviation increase in the TTI (0.07) decreases VKT by about −120 km and a one standard deviation increase in the change in TTI (0.02) decreases VKT by about −50 km. Consistent with congestion effects, the relationship between VKT and commuter demand is negative; a one standard deviation increase in commuters (1,074) decreases VKT by about −180 km. VKT is indeed higher in high-growth (C)MSAs and in rapidly-growing (C)MSAs; a one standard deviation increase in GMP (94.7) increases VKT by about +80 km and a one standard deviation increase in the change in GMP (21.7) increases VKT by about +60 km. Maps of the residuals from the full, random-intercepts models with STOCK (Fig. 2) and with FLOW (Fig. 3) versions of capacity, congestion,
higher in high-growth and in rapidly-growing (C)MSAs because growth indirectly increases demand for private-vehicle usage. VKT is expected to be lower in (C)MSAs with commuter rail because of the effect of mode choice on VMT (Holtzclaw, 1991, 1994). Census region controls for regional differences in freeway lane-kilometers and in arterial lanekilometers at the (C)MSA level across the U.S.
5. Results Random-Intercepts Models Random-intercepts models help to understand how household-level independent variables like total income and (C)MSA-level independent variables like road capacity affect vehicle-level VKT (Table 3). The results at the vehicle level are generally consistent with the hypothesized effects. Older vehicles are indeed driven less than newer vehicles; a one standard deviation increase in age (6.75 years) decreases VKT by about −80 km. The effect of gasoline price is negative, as expected; a one standard deviation increase in gasoline prices (0.10 dollars per liter) decreases VKT by at least −230 km. Consistent with the rebound effect, the effect of mileage on VKT is positive; a one standard deviation increase in mileage (2.60 km per liter) increases VKT by about +3,700 km. Consistent with expectations, vans (about −670 km), SUVs (about −1,800 km), and pickup trucks (about −2,000 km) are driven less than automobiles. Finally, as expected, the 6
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
Fig. 3. Map of residuals from the full, random-intercepts model with FLOW versions of capacity, congestion, demand, and growth independent variables at (C)MSA level.
capacity from 1995 to 2001 (+13.09%), but ranks only 22nd in percentage increase in arterial capacity from 2001 to 2009 (+12.79%). Such results suggest that the demand-inducing effects of freeway capacity and of arterial capacity are sensitive to the scale of the capacity in the respective (C)MSAs (Handy and Boarnet, 2014). Indeed, the differences in the scale of the capacity for freeways (about five times) and for arterials (about ten times) between Washington, D.C. and Greensboro are substantial. Amongst the lower-demand-than-expected (C)MSAs, Boston, Philadelphia, and San Antonio rank relatively high on percentage change in congestion in 2001 and in 2009. For example, Boston ranks 6th in percentage increase in congestion from 1995 to 2001 (+4.69%), Philadelphia ranks 10th in percentage increase in congestion from 2001 to 2009 (+1.61%), and San Antonio ranks 17th in percentage increase in congestion from 1995 to 2001 (+3.25%).
demand, and growth independent variables at the (C)MSA level help to illustrate the national differences in private-vehicle travel demand. First, as a check on the specification of the respective models, no clustering of high-residual and/or low-residual (C)MSAs is evident so both full, random-intercepts models account for national variation in private-vehicle travel demand. Second, on the one hand, the (C)MSAs where the observed demand is highest relative to the expected demand (based on the full, random-intercepts model specifications) are Washington, D.C. and Greensboro. On the other hand, the (C)MSAs where the observed demand is lowest relative to the expected demand (based on the full, random-intercepts model specifications) are Boston, Philadelphia, and San Antonio. Such results are not entirely consistent with the assertion that induced demand effects are greatest where the percentage increase in capacity is larger (Noland and Cowart, 2000). Amongst the higher-demand-than-expected (C)MSAs, on the one hand, Washington, D.C. ranks only 38th in percentage increase in freeway capacity from 1995 to 2001 (+7.96%), and only 35th in percentage increase in freeway capacity from 1995 to 2001 (+2.52%). Further, Washington, D.C. ranks only 28th in percentage increase in arterial capacity from 1995 to 2001 (+7.16%), and ranks only 19th in percentage increase in arterial capacity from 2001 to 2009 (+12.79%). On the other hand, Greensboro ranks only 18th in percentage increase in freeway capacity from 1995 to 2001 (+7.55%), but ranks 7th in percentage increase in freeway capacity from 2001 to 2009 (+22.53). Further, Greensboro ranks 13th in percentage increase in arterial
6. Discussion The absence of a demand-inducing effect from freeway capacity or from arterial capacity—in fact, the opposite is true for additional arterial capacity—is inconsistent with the empirical literature on induced demand. However, the context for the study is important to acknowledge. The first decade of the 21st century coincided with two noteworthy shocks: Hurricane Katrina and the Great Recession. The former greatly affected road capacity in one of the metropolitan areas (New Orleans), and the latter greatly affected aggregate demand in the U.S. 7
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
year lag periods in the study may not be long enough to capture the demand-inducing effects of additional road capacity.
from 2001 to 2009. In the former case, the percentage change in freeway capacity and in arterial capacity in New Orleans from 2001 to 2009 is negative: −10.61% and −3.37%, respectively. In the latter case, the percentage change in GMP for four of the forty-four (C)MSAs from 2001 to 2009 is negative: Rochester (−0.22%); Buffalo (−10.23%); Grand Rapids (−35.49%); and Greensboro (−53.54%). Second, the proportional reduction in error (PRE) from a null model (without independent variables at the vehicle level, the household level, or the metropolitan area level) to a full model (with independent variables at the vehicle level, the household level, and the metropolitan area level) shows that the vehicle level accounts for the majority of the variation in VKT (by about −94.39%). Such a result suggests that most of the variation in VKT is attributable to vehicle characteristics such as gasoline price, not to household characteristics such as total income or to metropolitan area characteristics such as road capacity. Notwithstanding the context of the study and the relative utility of the respective vehicle, household, and metropolitan area levels of independent variables to account for variation in private-vehicle travel demand, the policy implications of first the vehicle-level results and second the metropolitan area-level results are as follows. At the vehicle level, the implication of the positive result on the effect of gasoline price on VKT is that a popular policy tool to mitigate private-vehicle travel demand, that is, higher gasoline prices due to higher (federal and/or state) gasoline taxes, should be very efficacious. However, the rebound effect should, at a minimum, offset the demand management benefits of (gasoline) price sensitivity since higher fuel-economy vehicles are driven much more than lower fuel-economy vehicles. In the extreme, the strong rebound effect may equal or may surpass such benefits. At the metropolitan area level, controlling for worsening congestion, a lag period of from six to eight years is sufficient to capture the slight demand-reducing effects of adding more arterial capacity; that is, additional arterial capacity allows drivers to ultimately discover new routes to shorten the same trips.
References AASHO (American Association of State Highway Officials), 1957. A Policy on Arterial Highways in Urban Areas. AASHO, Washington, DC. Alcott, B., 2005. Jevon's paradox. Ecol. Econ. 54 (1), 9–21. Barr, L., 2000. Testing for the significance of induced highway travel demand in metropolitan areas. Transp. Res. Rec. 1706, 1–8. Bullen, N., Jones, K., Duncan, C., 1997. Modelling complexity: analysing between-individual and between-place variation—a multilevel tutorial. Environ. Plan. A 29 (4), 585–609. Cervero, R., Hansen, M., 2002. Induced travel demand and induced road investment: a simultaneous equation analysis. J. Transp. Econ. Pol. 36 (3), 469–490. Chakravarty, D., Dasgupta, S., Roy, J., 2013. Rebound effect: how much to worry? Curr. Opin. Environ. Sustain. 5 (2), 216–228. Downs, A., 1962. The law of peak-hour expressway congestion. Traffic Q. 16 (3), 393–409. Duncan, C., Jones, K., 2000. Using multilevel models to model heterogeneity: potential and pitfalls. Geogr. Anal. 32 (4), 279–305. Duranton, G., Turner, M., 2011. The fundamental law of road congestion: evidence from US cities. Am. Econ. Rev. 101 (6), 2616–2652. Frondel, M., Vance, C., 2013. Don't belittle the rebound effect. Nature 494 (7438), 430. Gillingham, K., Kotchen, M., Rapson, D., Wagner, G., 2013. The rebound effect is overplayed. Nature 493 (7433), 475–476. Goodwin, P., 1996. Empirical evidence on induced traffic: a review and synthesis. Transportation 23 (1), 35–54. Greening, L., Greene, D., Difiglio, C., 2000. Energy efficiency and consumption—the rebound effect—a survey. Energy Pol. 28 (6–7), 389–401. Handy, S., Boarnet, M., 2014. Impact of Highway Capacity and Induced Travel on Passenger Vehicle Use and Greenhouse Gas Emissions. Air Resources Board, Sacramento, CA. Hansen, M., Huang, Y., 1997. Road supply and traffic in California urban areas. Transp. Res. A 31 (3), 205–218. Hills, P., 1996. What is induced traffic? Transportation 23 (1), 5–16. Holtzclaw, J., 1991. Explaining Urban Density and Transit Impacts on Auto Use. National Resources Defense Council, San Francisco, CA. Holtzclaw, J., 1994. Using Residential Patterns and Transit to Decrease Auto Dependence and Costs. National Resources Defense Council, San Francisco, CA. Hymel, K., Small, K., Van Dender, K., 2010. Induced demand and rebound effects in road transport. Transp. Res. B 44 (10), 1220–1241. Jenkins, J., Nordhaus, T., Shellenberger, M., 2011. Energy Emergence: Rebound & Backfire as Emergent Phenomena. The Breakthrough Institute, Oakland, CA. Jevons, W., 1865. The Coal Question: An Enquiry Concerning the Progress of the Nation, and the Probable Exhaustion of Our Coal-Mines. Macmillan, London, UK. Jiwattanakulpaisarn, P., Noland, R., Graham, D., 2012. Marginal productivity of expanding highway capacity. J. Transp. Econ. Pol. 46 (3), 333–347. Khazzoom, J., 1980. Economic implications of mandated efficiency in standards for household appliances. Energy J. 1 (4), 21–40. Liddle, B., 2009. Long-run relationship among transport demand, income, and gasoline price for the US. Transp. Res. D 14 (2), 73–82. Litman, T., 2001. Generated traffic: implications for transportation planning. ITE J. 71 (4), 38–47. Litman, T., 2013. Changing North American vehicle-travel price sensitivities: implications for transport and energy policy. Transp. Policy 28, 2–10. Noland, R., 2001. Relationships between highway capacity and induced vehicle travel. Transp. Res. A 35 (1), 47–72. Noland, R., Cowart, W., 2000. Analysis of metropolitan highway capacity and the growth in vehicle miles of travel. Transportation 27 (4), 363–390. Noland, R., Lem, L., 2002. A review of the evidence for induced travel and changes in transportation and environmental policy in the US and the UK. Transp. Res. Part D: Transp. Environ. 7 (1), 1–26. Raudenbush, S., Bryk, A., 2002. Hierarchical Linear Models: Applications and Data Analysis Methods. Sage, Thousand Oaks, CA. Rentziou, A., Gkritza, K., Souleyrette, R., 2012. VMT, energy consumption, and GHG emissions forecasting for passenger transportation. Transp. Res. A: Policy Pract. 46 (3), 487–500. Schipper, L., 2000. On the rebound: the interaction of energy efficiency, energy use and economic activity. An introduction. Energy Pol. 28 (6–7), 351–353. Schrank, D., Eisele, W., Lomax, T., 2012. Urban Mobility Report. Texas Transportation Institute, College Station, TX. Small, K., Van Dender, K., 2007. Fuel efficiency and motor vehicle travel: the declining rebound effect. Energy J. 28 (1), 25–51. Sorrell, S., 2007. The Rebound Effect: An Assessment of the Evidence for Economy-Wide Energy Savings from Improved Energy Efficiency. UK Energy Research Centre, London, UK. Sorrell, S., Dimitropoulos, J., Sommerville, M., 2009. Empirical estimates of the direct rebound effect: a review. Energy Pol. 37 (4), 1356–1371. Stern, P., 1984. Improving Energy Demand Analysis. National Academy Press, Washington, DC. Su, Q., 2011. Induced motor vehicle travel from improved fuel efficiency and road expansion. Energy Pol. 39 (11), 7257–7264. Tierney, J., 2011. When Energy Efficiency Sullies the Environment. The New York Times
7. Conclusions The contributions of the results to the empirical literature on induced demand are attributable to the data and to the design of the study. The use of disaggregate data on travel behavior unmasks how drivers, rather than groups of drivers, respond to changes in road capacity. The decision to disaggregate freeway lane-kilometers from arterial lane-kilometers rather than to aggregate both to proxy for stock and for flow measures of road capacity (Noland and Cowart, 2000) highlights the differential effects of freeways and of arterials on VKT. To that end, the most important contribution of the study to the induced travel demand literature is that adding more arterial capacity actually decreases VKT. However, the demand-reducing effects of additional arterial capacity are not as consequential to private-vehicle travel behavior as are vehicle characteristics such as gasoline price. The most fruitful direction for future research is to explore the suggestion that the demand-inducing effect of adding road capacity is dependent on the scale of the preexisting road capacity. To return to the example in the discussion section, the same percentage increase in capacity may induce more demand in one metropolitan area versus another metropolitan area where the lane-kilometers of preexisting capacity is greater. One approach with the potential to capture differences in the scale of the capacity between metropolitan areas is to use a geographic information system to map where additional road capacity is added and how that new capacity affects the balance of the road network across urban, suburban, and exurban locations. The expectation is that adding road capacity in more peripheral locations will engender more private-vehicle travel. Another approach is to pool disaggregate data from the next NHTS with the disaggregate data from the 2001 NHTS and from the 2009 NHTS to extend the lag period for the respective flow measures of road capacity, road congestion, commuter demand, and economic growth. To that end, the six year and the eight 8
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
USCM (U.S. Conference of Mayors), 2010. U.S. Metro Economies: Pace of Economic Recovery: GMP and Jobs. IHS Global Insight, Lexington, MA. USDOT (U.S. Department of Transportation), 2017. Historical VMT from 1970. Washington, DC Online at: https://www.fhwa.dot.gov/policyinformation/travel_ monitoring/tvt.cfm (accessed 09.28.17). USDOT (U.S. Department of Transportation), 2001. 2001 National Household Travel Survey. USDOT, Washington, DC Online at: http://nhts.ornl.gov/download.shtml# 2001 (accessed 11.11.14). USDOT (U.S. Department of Transportation), 2009. 2009 National Household Travel Survey. USDOT, Washington, DC Online at: http://nhts.ornl.gov/download.shtml# 2009 (accessed 11.11.14). USEIA (U.S. Energy Information Administration), 2014. Energy Efficiency Information. USEIA, Washington, DC Online at: https://www.eia.gov/ (accessed 11.11.14). USEPA (U.S. Environmental Protection Agency), 2014. Greenhouse Gas Emissions from a Typical Passenger Vehicle. USEPA, Washington, DC.
March 7. Online at: http://www.nytimes.com/2011/03/08/science/08tier.html. TRB (Transportation Research Board), 1995. Expanding Metropolitan Highways: Implications for Air Quality and Energy Use. National Academy Press, Washington, DC. TTI (Texas Transportation Institute), 2012. Congestion Data for Your City. TTI, College Station, TX Online at: http://mobility.tamu.edu/ums/congestion-data/ (accessed 08.01.15). TTI (Texas Transportation Institute), 2014. Congestion Data for Your City. TTI, College Station, TX Online at: http://mobility.tamu.edu/ums/congestion-data/ (accessed 04.24.16). Turner, K., 2013. “Rebound” effects from increased energy efficiency: a time to pause and reflect. Energy J. 34 (4), 25–42. USCM (U.S. Conference of Mayors), 2000. U.S. Metro Economies: The Engines of America's Growth. Standard & Poor's, Lexington, MA. USCM (U.S. Conference of Mayors), 2003. The Role of Metro Areas in the US Economy. Global Insight, Lexington, MA.
9
Contents lists available at ScienceDirect
Journal of Transport Geography journal homepage: www.elsevier.com/locate/jtrangeo
Effects of additional capacity on vehicle kilometers of travel in the U.S.: Evidence from National Household Travel Surveys
MARK
Edmund J. Zolnik Schar School of Policy and Government, George Mason University, 3351 Fairfax Drive, MS 3B1, Arlington, VA, 22201, USA
A R T I C L E I N F O
A B S T R A C T
Keywords: Induced demand Vehicle kilometers of travel Capacity Multilevel models
Adding capacity is one policy mechanism to alleviate congestion. However, the empirical evidence strongly suggests that additional capacity only makes congestion worse. This study analyzes the differential effects of additional freeway capacity versus additional arterial capacity on vehicle kilometers of travel (VKT) in metropolitan areas across the U.S. The analysis uses vehicle data and household data from the 2001 and the 2009 National Household Travel Surveys (NHTS) and includes stock and flow measures of road capacity, road congestion, commuter demand, and economic growth for metropolitan areas. Taking into account differences between metropolitan areas on each measure, the study adopts a novel multilevel model approach to estimate how additional capacity affects VKT. Results indicate that adding more arterial capacity slightly decreases VKT over a lag period from six years (1995 to 2001) to eight years (2001 to 2009), probably because adding arterials shortens routes between origins and destinations more so than adding freeways. Consistent with expectations, VKT is lower in more congested metropolitan areas, and in metropolitan areas that got more congested. Results also indicate that rebound effects (higher fuel-economy vehicles are driven much more than lower fuel-economy vehicles) will at least partially offset the demand management benefits of (gasoline) price sensitivity (higher gasoline prices decrease VKT).
1. Introduction Congestion is a major transportation problem in metropolitan areas across the U.S. There are many different strategies to mitigate congestion, including adding more capacity (freeways and arterials). Whichever strategy is adopted, it is important to account for all of the potential impacts. For example, in the case of adding more capacity, it is important to account for indirect and long-term impacts of induced demand for private-vehicle travel. In order to account fully for such impacts, this study explores the differential effects of additional freeway capacity versus additional arterial capacity on private-vehicle travel demand in metropolitan areas across the U.S. in 2001 and in 2009. Freeways and arterials are complementary functional types of capacity in the road network, but the differential effects of additional capacity are important to explore for many reasons. First, the economic returns on different functional types of capacity are not the same. Returns for interstate highways are greater than returns for non-interstate major roads (arterials and collectors). Likewise, returns for the latter are greater than returns for local roads (Jiwattanakulpaisarn et al., 2012). Second, the effects of different functional types of capacity on vehicle miles of travel (VMT), the most common measure of privatevehicle travel demand in the U.S. (Rentziou et al., 2012), are not the
same. For example, the travel-time benefits of new collectors are greater than the travel-time benefits of new interstates or new arterials (Noland, 2001). Third, the effect of additional capacity is not the same in different metropolitan areas. That is, the effect tends to be greater in metropolitan areas where the percentage increase in capacity (freeway and arterial) is larger (Noland and Cowart, 2000). The organization of the paper is as follows. The next section briefly reviews the induced demand literature. The data section lists the data sources for the vehicle level, the household level, and the metropolitan area level, respectively. The methodology section introduces the multilevel model in the study and also hypothesizes the effects of the vehicle-level, the household-level, and the metropolitan area-level independent variables. The results section presents the outcomes of the multilevel model. The discussion section focuses on the policy implications of the results. Finally, the conclusions section highlights the contributions of the results to the induced demand literature, as well as the most fruitful direction for future research. 2. Background Economic theory on induced demand—additional capacity attracts more traffic (AASHO, 1957)—suggests that additional capacity
E-mail address: [email protected]. https://doi.org/10.1016/j.jtrangeo.2017.10.020 Received 15 January 2015; Received in revised form 6 October 2017; Accepted 26 October 2017 0966-6923/ © 2017 Elsevier Ltd. All rights reserved.
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
Table 1 Key literature on induced demand. Author(s), year
Area(s)
Downs, 1962 Hills, 1996 Goodwin, 1996 Noland and Lem, 2002 Hansen and Huang, 1997 Noland and Cowart, 2000 Noland, 2001
US UK UK UK/US
Methodology
Results Law of peak-hour traffic congestion—traffic and congestion are in equilibrium. Additional capacity induces demand by changing network accessibility. Demand induced by 10% in the short term and by 20% in the long term. Lane-mile elasticities range from +0.3 to +0.6 regardless of the data or the methodology.
FE
Marginal effect of lane-miles of state highway on VMT is greatest in largest CMSAs/MSAs.
US
C/CEA/(C) MSA MSA
1938–1994 1994–1998 1993–2001 1973–1990
Theoretical Theoretical Review Review
1982–1996
2SLS
US
State
1984–1996
FE/SURE
CA
Unit(s)
Year(s)
Cervero and Hansen, 2002 Hymel et al., 2010 Duranton and Turner, 2011 Su, 2011
CA
C/CEA
1976–1997
SE
Lane-mile additions account for 15% of annual VMT growth with great variation between MSAs. Induced demand effects of lower functional types of new capacity (collectors) is greater than the induced demand effects of higher functional types of new capacity (interstates and arterials). Induced demand effect of +0.6 and induced investment effect of +0.3.
US US
State MSA
1966–2004 1983/1993/2003
3SLS 3SLS
Total road length induces demand by 3.7% in the short run and by 18.6% in the long run. VKT increases proportional to additional interstate capacity.
US
State
2001–2008
DPD
Rentziou et al., 2012
US
State
1998–2008
SURE/RE
Short-run and long-run rebound effects of 3% and 11%, respectively. Short-run and long-run road capacity per capita effects of 7% and 26%, respectively. VMT elasticity with respect to lane-miles higher on urban versus rural roads.
Area abbreviations: CA = California; UK = United Kingdom; and US = United States. Unit abbreviations are: C/CEA = County/County Equivalent Area; (C)MSA = (Consolidated) Metropolitan Statistical Area. Methodology abbreviations are: 2SLS = Two-Stage Least Squares; 3SLS = Three-Stage Least Squares; DPD = Dynamic Panel Data; FE = Fixed Effects; RE = Random Effects; SE = Simultaneous Equations; and SURE = Seemingly Unrelated Regression Equations.
capacity. For example, increases in traffic on the new routes are not offset by decreases in traffic on the old routes that provide alternatives either on a short-term or a long-term basis. The latter empirical result contradicts Downs' (1962) theory that older commuter routes will experience less congestion. A review of the empirical literature on induced demand from the U.K. and from the U.S. by Noland and Lem (2002) showed that lane-mile elasticities range from +0.3 to +0.6, regardless of the data or the methodology. The most methodologically sophisticated citations in the induced demand literature also account for the rebound effect (Jevons, 1865; Khazzoom, 1980; Alcott, 2005; Sorrell, 2007). The rebound effect is “the interaction of energy use with the efficiency of energy use: lower the energy required to do something, and you will do a bit more of that thing” (Schipper, 2000, 351). The most conservative estimates of the short-term (one-year) rebound effect for private vehicles from the literature are about 10% (Greening et al., 2000; Sorrell et al., 2009; Chakravarty et al., 2013). However, the least conservative estimates of the long-term rebound effect for private vehicles are about 30%. Such a range of estimates on the magnitude of the rebound effect is a source of controversy in academic and in policy circles (Tierney, 2011; Turner, 2013). Some argue that the rebound effect is in decline (Small and Van Dender, 2007) or that it is just a distraction to divert attention from efforts to enact stricter energy-efficiency regulations (Gillingham et al., 2013). Others argue that energy efficiency standards like Corporate Average Fuel Economy (CAFE) standards in the U.S. are not cost effective because of the rebound effect (Frondel and Vance, 2013) and, perhaps, greater energy efficiency leads to backfire—more usage offsets any energy savings due to efficiency improvements (Jenkins et al., 2011; Tierney, 2011). Regardless of the academic and the policy debates on the magnitude of the rebound effect, two examples from the induced demand literature (Hymel et al., 2010; Su, 2011) confirm that the rebound effect affects private-vehicle travel demand. In order to heed Hills' (1996) call to specify a realistically complex mathematical model of induced demand, the study pools disaggregate cross-sectional data from the National Household Travel Survey (NHTS) for two time points (2001 and 2009) to account for short-term and for long-term travel behavior changes in response to additional capacity. Barr (2000) also uses disaggregate cross-sectional data from an older version of the NHTS, known as the Nationwide Personal Transportation Survey, to model induced demand, but only for one time point (1995).
increases traffic because travel times decrease. The effect on travel times is due to the following (short-term and/or long-term) behavioral changes in response to less congestion:
• taking different routes for the same trip; • making the same trips at different times of the day; • using different modes for the same trip; • substituting destinations for the same (shopping or recreational) trip; and • making more trips (TRB, 1995). Table 1 summarizes the key literature on induced demand, and the citations here are by no means exhaustive. Rather, the citations highlight the trajectory of the literature from the theoretical to the empirical, with the trend toward methodological sophistication in the latter. The law of peak-hour traffic congestion (Downs, 1962, 393) states that “congestion rises to meet maximum capacity.” The law assumes the following with regard to commuter decision making:
• commuters seek to minimize travel times cognizant of income, monetary cost, place of residence, and personal comfort constraints; • the law of inertia rules—unless events in the environment compel change mode choices and route choices are habitual; events in the environment which compel mode choice and route • choice changes are those that decrease travel times; and • commuters are of two types—those who are passive and those who are active in seeking out different routes to save themselves time. Hills (1996) spelled out exactly what is meant by induced demand in order to distinguish generated traffic from induced traffic and to relate the latter phenomenon to the range of travel behavior responses to additional capacity. The cumulative route choices of all private-vehicle drivers generate traffic on a network whereas the addition of capacity to a network, by design, changes accessibility and induces traffic.
A review of the extant empirical literature on induced demand by Goodwin (1996) showed that additional capacity induces demand by 10% in the short term and by 20% in the long term. However, induced demand is dependent on the context, size, and location of the new
2
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
Table 2 Descriptive statistics for vehicle-, household-, and (C)MSA-levels of analysis. Level
n
Vehicle
130,489
Variable
Category
Dependent VKT Independent Age (years) Emissions (kilograms) Gasoline price ($ per liter) Mileage (kilometers per liter) Type
Automobile Van Sports Utility Vehicle Pickup Truck 2001 2009
Year Household
Mean
SD
18,392.51
15,959.37
9.20 18,497.21 0.78 8.82 55.93% 8.51% 20.15% 15.42% 16.23% 83.77%
6.75 15,706.25 0.10 2.60
66,445 Independent Income Less than $25,000 $25,000 to $49,999 $50,000 to $74,999 $75,000 to $99,999 Greater than or Equal to $100,000
10.81% 22.04% 18.59% 17.10% 31.46% 2.55
1.22
Arterial Arterial Change Freeway Freeway Change
6,988 682 2,380 179
6,466 648 2,073 177
Travel Time Index Travel Time Index Change
1.22 0.04
0.07 0.02
Commuters (1,000) Commuters Change (1,000)
1,159 197
1,074 169
Gross Metropolitan Product Gross Metropolitan Product Change
110.5 25.1
94.7 21.7
Yes No
35.98% 64.02%
Northeast Midwest South West
14.50% 8.40% 45.03% 32.07%
Vehicle count (C)MSA
44 Independent Capacity (lane-kilometers)
Congestion
Demand
Growth ($1,000,000,000)
Rail
Region
VKT is vehicle kilometers of travel.
empirical literature suggests that the induced demand effect is not the same at the metropolitan area level across the U.S. (Noland and Cowart, 2000). The induced demand effect is not homogeneous because travel behavior responses are not directly attributable to the additional capacity; rather, such responses are directly attributable to the resultant congestion reduction from the additional capacity (Litman, 2001). Besides congestion, Handy and Boarnet (2014) suggest that the induced demand effect may vary nationally by the size of the metropolitan area and by gasoline price. A random effects model (Duncan and Jones, 2000) of induced demand allows for variation in travel behavior responses to additional capacity at the metropolitan area level across the U.S.—the intercepts for each metropolitan area vary randomly. Besides the advantage of specifying a more realistically complex model of induced demand that nests travel behavior within metropolitan areas, another advantage is that a random effects specification accounts for the independence of observations. If observations within metropolitan areas are not independent but are correlated, then the assumption that the observations are independent and identically distributed is problematic. A multilevel model adds a second variance term to account for the nesting of observations and so extends the random part of the
The use of disaggregate data is important because most mathematical models of induced demand effects in the economics literature use aggregate cross-sectional data or aggregate cross-sectional/time-series data. The overreliance on aggregate data draws criticism from transportation planners who argue that the focus ought to be on how drivers respond to additional capacity so as to capture behavioral changes best and, ultimately, understand induced demand effects (Noland and Lem, 2002). To that end, the use of disaggregate data in this study contributes to a better balance in the empirical literature on induced demand. The study also contributes to the trend toward more methodological sophistication in the empirical literature on induced demand via the adoption of a novel multilevel approach (Noland, 2001). Such an approach is novel because mathematical models in the empirical literature on induced demand tend to assume that the effect is fixed at the metropolitan area level—the intercepts for each metropolitan area are independently fixed while the slopes are the same across metropolitan areas. In other words, travel behavior responses to additional capacity are the same at the metropolitan area level across the U.S. (Noland, 2001). However, a fixed effect specification is not realistic because the 3
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
Fig. 1. Historical vehicle kilometers of travel. USDOT, 2017.
gasoline prices in dollars per liter are real 2009 gasoline prices in dollars per liter. Mileage is fuel economy in kilometers per liter. Type is vehicle type: automobile (automobile, car, or station wagon; van (cargo, mini, or passenger); SUV; or pickup truck. Year is zero if the vehicle is from the 2009 NHTS subsample, the larger of the two subsamples also known as the referent category, one otherwise. Data at the household level (n = 66,445) include information on the characteristics of each household (Table 2). Total income is the total household income in dollars. Vehicle count is the number of household vehicles. Data at the (C)MSA level (n = 44) include stock and flow measures of road capacity, road congestion, commuter demand, and economic growth as well as commuter rail and census region (Table 2). The stock versions are for 2001 and 2009, respectively. The flow versions are differences from 1995 to 2001 and from 2001 to 2009, respectively, reflective of the time gap between successive NHTSs. The measures of road capacity are the lane-kilometers of freeways and the lane-kilometers of arterials from the Highway Performance Monitoring System (HPMS) (TTI, 2012). The measure of road congestion is the Travel Time Index (TTI) (TTI, 2014). The TTI is the ratio of travel times under peak travel conditions to travel times under nonpeak travel conditions (Schrank et al., 2012). The numerator and the denominator of the TTI are in units of time, so the ratio is unitless. This allows for the comparison of trips of different lengths to estimate how peak conditions affect travel times. Commuter demand is the number of private-vehicle commuters from the NHTS (TTI, 2014). Economic growth is the gross metropolitan product (GMP) in billions of dollars (USCM, 2000, 2003, 2010). GMP is the market value of all final goods and services produced within a (C)MSA in a given year. The flow version of GMP for 2001 is the difference from 1997 to 2001, not from 1995 to 2001, due to data availability limitations. Commuter rail is one if the (C)MSA has commuter rail, zero otherwise. The census regions are the Northeast, Midwest, South, and West regions of the U.S. The next section introduces the multilevel model in the study.
model (Bullen et al., 1997). The following section lists the data sources for the vehicle level, the household level, and the metropolitan area level, respectively. 3. Data Data are from the 2001 and the 2009 NHTS, respectively (USDOT, 2001, 2009). The 2001 and 2009 NHTS are cross-sectional surveys of travel behavior in the U.S. The sample sizes for the respective vehicle and household files in the 2001 NHTS are 139,382 and 69,817 respectively, and the sample sizes for the respective vehicle and household files in the 2009 NHTS are 309,163 and 150,147. In order to nest vehicles accurately within households in Consolidated Metropolitan Statistical Areas (CMSA) or Metropolitan Statistical Areas (MSA), hereafter known as a (C)MSA, those with missing data at the vehicle level, the household level, or the (C)MSA level are excluded from the respective subsamples. Application of the above selection criterion to the 2001 NHTS left a subsample of 21,179 vehicles nested within 11,032 households nested within 44 (C)MSAs. Application of the above selection criterion to the 2009 NHTS left a subsample of 109,310 vehicles nested within 55,413 households nested within 44 (C)MSAs. Pooling the 2001 subsample with the 2009 subsample nests 130,489 vehicles within 66,445 households within 44 (C)MSAs. To be clear, the 44 (C)MSAs in the 2001 subsample and the 44 (C)MSAs in the 2009 subsample are the same. Data at the vehicle level (n = 130,489) include information on each household's vehicles (Table 2). Vehicle kilometers of travel (VKT) is the best estimate of the kilometers each vehicle was driven over the past twelve months. Age is vehicle age in years. Emissions are annual carbon dioxide (CO2) emissions in kilograms (USEPA, 2014). Gasoline price is an estimate of the price of gasoline in dollars per liter for each vehicle (car, van, sports utility vehicle (SUV), or pickup truck) from the Energy Information Administration (USEIA, 2014). In order to harmonize gasoline prices for 2001 and for 2009 (Stern, 1984), nominal 2001 4
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
(Raudenbush and Bryk, 2002). Within each household, vehicle-level VKT is a function of vehicle-level independent variables plus an individual-level error term:
Table 3 Regression coefficient estimates for vehicle-, household-, and (C)MSA-levels of analysis from full, random-intercepts models with STOCK and FLOW versions of capacity, congestion, demand, and growth independent variables. Variable
Category
Vehicle Age (years) Emissions (kilograms) Gasoline Price ($ per liter) Mileage (kilometers per liter) Type Automobile Van Sports utility vehicle Pickup truck
Yvhm = β0 hm + β1 hm W1 vhm + …+βAhm WAvhm + rvhm
FLOW Coefficient (SE)
−11.98 (1.73)a +0.96 (0.001)a −2,277.43 (480.37)a +1,429.62 (4.85)a
−11.95 (3.31)a +0.96 (0.01)a −2,899.31 (1,493.15)c +1,429.41 (31.79)a
Referent −673.83 (39.84)a −1,807.82 (30.94)a −2,008.66 (33.94)a
Referent −676.21 (61.94)a −1,807.74 (59.77)a
where Yvhm is the VKT for vehicle v in household h in (C)MSA m; β0hm is the y-intercept term for household h in (C)MSA m; βAhm are a = 1, …, A vehicle-level coefficients; WAvhm is the independent variable a for vehicle v in household h and (C)MSA m; and rvhm is the vehicle-level random effect term. At level 2, the variation between households is a function of household-level independent variables plus a household-level error term:
−2,006.88 (68.01)a
β0 hm = π00 m + π01m X1 hm + …+π0 Cm XChm + e0 hm
−672.28 (120.61)a Referent
−824.79 (356.32)b Referent
Household Income Less than $25,000 $25,000 to $49,999 $50,000 to $74,999 $75,000 to $99,999 Greater than or equal to $100,000 Vehicle Count (C)MSA Intercept Capacity (lane-kilometers) Arterial Arterial change Freeway Freeway change Congestion Travel time index
+280.69 +251.35 +166.79 +152.36 Referent
(39.55)a (30.68)a (31.29)a (31.68)a
+282.46 +253.41 +169.17 +153.97 Referent
(35.74)a (37.58)a (39.10)a (28.69)a
+29.56 (9.22)a
+29.67 (9.69)a
+18,961.38 (28.86)a
+19,035.00 (49.04)a
At level 3, the variation between (C)MSAs is a function of (C)MSAlevel independent variables plus a (C)MSA-level error term:
π00m = γ000 + γ001 Z1m + …+γ00B ZBm + u 00m
(3)
where γ000 is the y-intercept term for (C)MSA m; γ00B are b = 1, …, B (C)MSA-level coefficients; ZBm are (C)MSA-level independent variables; and u00m is the (C)MSA-level random effect term.
+0.02 (0.01) −0.06 (0.03)c +0.003 (0.01)
In the multilevel models the regression coefficients at level 1 and at level 2 are fixed. These three-level models are known as random-intercepts models. The random-intercepts models estimate how household characteristics such as total income and metropolitan area characteristics such as road capacity affect vehicle-level VKT. The following subsection hypothesizes on the effects of the vehiclelevel, the household-level, and the (C)MSA-level independent variables.
−0.23 (0.17) −1,714.99 (291.76)a −2,494.51 (813.32)a
Travel time index change Demand Commuters (1,000) Commuters change (1,000) Growth ($1,000,000,000) Gross metropolitan product Gross metropolitan product change Rail Yes No Region Northeast Midwest South West
(2)
where π00m is the y-intercept term for (C)MSA m; π0Cm are c = 1, …, C household-level coefficients; XChm is the independent variable c for household h in (C)MSA m; and e0hm is the household-level random effect term.
Year 2001 2009
(1)
STOCK Coefficient (SE)
−0.17 (0.06)a −0.20 (0.12)
+0.81 (0.35)
4.1. Hypothesized effects of vehicle-, household-, and (C)MSA-level independent variables
b
At the vehicle level, less fuel-efficient, older vehicles will be driven less than more fuel-efficient, newer vehicles. Consistent with the trend toward greater price sensitivity (Litman, 2013), the effect of gasoline prices will be negative. Consistent with the rebound effect where use increases with greater energy efficiency (Schipper, 2000) the effect of mileage will be positive. Higher fuel-economy automobiles will be driven more than lower fuel-economy vans, SUVs, or pickup trucks. VKT will be higher in 2009 than in 2001 even though aggregate demand in 2009 is below the historical trend (Fig. 1). At the household level, the empirical evidence strongly suggests that VKT increases with income (Liddle, 2009). The relationship between VKT and vehicle count is expected to be positive because the number of vehicles per household grew slightly from the 2001 NHTS subsample (2.52) to the 2009 NHTS subsample (2.56). At the (C)MSA level, VKT is expected to be higher in (C)MSAs that have more freeway lane-kilometers and more arterial lane-kilometers and in (C)MSAs that have added more freeway lane-kilometers and added more arterial lane-kilometers (Noland and Cowart, 2000). VKT is expected to be lower in more congested (C)MSAs and in (C)MSAs that got more congested. VKT is also expected to be lower in (C)MSAs that have more commuters and in (C)MSAs that added more commuters due to congestion effects (Handy and Boarnet, 2014). VKT is expected to be
+2.54 (1.18)b
+259.36 (44.45)a Referent
+58.69 (48.91) Referent
+24.04 (46.56) −5.99 (42.93) Referent +207.81 (48.75)a
+110.69 (74.76) −39.21 (58.01) Referent +145.22 (103.60)
Dependent variable is VKT. Referent category represents the highest-frequency category. Standard errors appear in parentheses. STOCK versions of capacity, congestion, demand, and growth independent variables at (C)MSA level are for 2001 and for 2009, respectively. FLOW versions of capacity, congestion, demand, and growth independent variables at (C)MSA level are differences from 2001 to 2009, respectively. a Indicates significance at 99% confidence level. b Indicates significance at 95% confidence level. c Indicates significance at 90% confidence level.
4. Methodology The VKT multilevel models in the study are three-level models of vehicles (v) nested within households (h) nested within (C)MSAs (m) 5
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
Fig. 2. Map of residuals from the full, random-intercepts model with STOCK versions of capacity, congestion, demand, and growth independent variables at (C)MSA level.
year regression coefficient is negative; VKT is at least −670 km lower in 2001 than in 2009. The results at the household level are generally inconsistent with the hypothesized effects. Inconsistent with empirical evidence, VKT decreases with income; it is highest in the lowest-income category (less than $25,000). The effect of vehicle count on VKT is positive, as expected; a one standard deviation increase in vehicle count (1.22) increases VKT by about +40 km. The results at the (C)MSA level are somewhat consistent with the hypothesized effects. VKT is lower, not higher, in (C)MSAs that have added more arterial lane-kilometers; a one standard deviation increase in the change in arterial lane-kilometers (648 lane-kilometers) decreases VKT by about −40 km. VKT is indeed lower in more congested (C)MSAs and in (C)MSAs that got more congested; a one standard deviation increase in the TTI (0.07) decreases VKT by about −120 km and a one standard deviation increase in the change in TTI (0.02) decreases VKT by about −50 km. Consistent with congestion effects, the relationship between VKT and commuter demand is negative; a one standard deviation increase in commuters (1,074) decreases VKT by about −180 km. VKT is indeed higher in high-growth (C)MSAs and in rapidly-growing (C)MSAs; a one standard deviation increase in GMP (94.7) increases VKT by about +80 km and a one standard deviation increase in the change in GMP (21.7) increases VKT by about +60 km. Maps of the residuals from the full, random-intercepts models with STOCK (Fig. 2) and with FLOW (Fig. 3) versions of capacity, congestion,
higher in high-growth and in rapidly-growing (C)MSAs because growth indirectly increases demand for private-vehicle usage. VKT is expected to be lower in (C)MSAs with commuter rail because of the effect of mode choice on VMT (Holtzclaw, 1991, 1994). Census region controls for regional differences in freeway lane-kilometers and in arterial lanekilometers at the (C)MSA level across the U.S.
5. Results Random-Intercepts Models Random-intercepts models help to understand how household-level independent variables like total income and (C)MSA-level independent variables like road capacity affect vehicle-level VKT (Table 3). The results at the vehicle level are generally consistent with the hypothesized effects. Older vehicles are indeed driven less than newer vehicles; a one standard deviation increase in age (6.75 years) decreases VKT by about −80 km. The effect of gasoline price is negative, as expected; a one standard deviation increase in gasoline prices (0.10 dollars per liter) decreases VKT by at least −230 km. Consistent with the rebound effect, the effect of mileage on VKT is positive; a one standard deviation increase in mileage (2.60 km per liter) increases VKT by about +3,700 km. Consistent with expectations, vans (about −670 km), SUVs (about −1,800 km), and pickup trucks (about −2,000 km) are driven less than automobiles. Finally, as expected, the 6
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
Fig. 3. Map of residuals from the full, random-intercepts model with FLOW versions of capacity, congestion, demand, and growth independent variables at (C)MSA level.
capacity from 1995 to 2001 (+13.09%), but ranks only 22nd in percentage increase in arterial capacity from 2001 to 2009 (+12.79%). Such results suggest that the demand-inducing effects of freeway capacity and of arterial capacity are sensitive to the scale of the capacity in the respective (C)MSAs (Handy and Boarnet, 2014). Indeed, the differences in the scale of the capacity for freeways (about five times) and for arterials (about ten times) between Washington, D.C. and Greensboro are substantial. Amongst the lower-demand-than-expected (C)MSAs, Boston, Philadelphia, and San Antonio rank relatively high on percentage change in congestion in 2001 and in 2009. For example, Boston ranks 6th in percentage increase in congestion from 1995 to 2001 (+4.69%), Philadelphia ranks 10th in percentage increase in congestion from 2001 to 2009 (+1.61%), and San Antonio ranks 17th in percentage increase in congestion from 1995 to 2001 (+3.25%).
demand, and growth independent variables at the (C)MSA level help to illustrate the national differences in private-vehicle travel demand. First, as a check on the specification of the respective models, no clustering of high-residual and/or low-residual (C)MSAs is evident so both full, random-intercepts models account for national variation in private-vehicle travel demand. Second, on the one hand, the (C)MSAs where the observed demand is highest relative to the expected demand (based on the full, random-intercepts model specifications) are Washington, D.C. and Greensboro. On the other hand, the (C)MSAs where the observed demand is lowest relative to the expected demand (based on the full, random-intercepts model specifications) are Boston, Philadelphia, and San Antonio. Such results are not entirely consistent with the assertion that induced demand effects are greatest where the percentage increase in capacity is larger (Noland and Cowart, 2000). Amongst the higher-demand-than-expected (C)MSAs, on the one hand, Washington, D.C. ranks only 38th in percentage increase in freeway capacity from 1995 to 2001 (+7.96%), and only 35th in percentage increase in freeway capacity from 1995 to 2001 (+2.52%). Further, Washington, D.C. ranks only 28th in percentage increase in arterial capacity from 1995 to 2001 (+7.16%), and ranks only 19th in percentage increase in arterial capacity from 2001 to 2009 (+12.79%). On the other hand, Greensboro ranks only 18th in percentage increase in freeway capacity from 1995 to 2001 (+7.55%), but ranks 7th in percentage increase in freeway capacity from 2001 to 2009 (+22.53). Further, Greensboro ranks 13th in percentage increase in arterial
6. Discussion The absence of a demand-inducing effect from freeway capacity or from arterial capacity—in fact, the opposite is true for additional arterial capacity—is inconsistent with the empirical literature on induced demand. However, the context for the study is important to acknowledge. The first decade of the 21st century coincided with two noteworthy shocks: Hurricane Katrina and the Great Recession. The former greatly affected road capacity in one of the metropolitan areas (New Orleans), and the latter greatly affected aggregate demand in the U.S. 7
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
year lag periods in the study may not be long enough to capture the demand-inducing effects of additional road capacity.
from 2001 to 2009. In the former case, the percentage change in freeway capacity and in arterial capacity in New Orleans from 2001 to 2009 is negative: −10.61% and −3.37%, respectively. In the latter case, the percentage change in GMP for four of the forty-four (C)MSAs from 2001 to 2009 is negative: Rochester (−0.22%); Buffalo (−10.23%); Grand Rapids (−35.49%); and Greensboro (−53.54%). Second, the proportional reduction in error (PRE) from a null model (without independent variables at the vehicle level, the household level, or the metropolitan area level) to a full model (with independent variables at the vehicle level, the household level, and the metropolitan area level) shows that the vehicle level accounts for the majority of the variation in VKT (by about −94.39%). Such a result suggests that most of the variation in VKT is attributable to vehicle characteristics such as gasoline price, not to household characteristics such as total income or to metropolitan area characteristics such as road capacity. Notwithstanding the context of the study and the relative utility of the respective vehicle, household, and metropolitan area levels of independent variables to account for variation in private-vehicle travel demand, the policy implications of first the vehicle-level results and second the metropolitan area-level results are as follows. At the vehicle level, the implication of the positive result on the effect of gasoline price on VKT is that a popular policy tool to mitigate private-vehicle travel demand, that is, higher gasoline prices due to higher (federal and/or state) gasoline taxes, should be very efficacious. However, the rebound effect should, at a minimum, offset the demand management benefits of (gasoline) price sensitivity since higher fuel-economy vehicles are driven much more than lower fuel-economy vehicles. In the extreme, the strong rebound effect may equal or may surpass such benefits. At the metropolitan area level, controlling for worsening congestion, a lag period of from six to eight years is sufficient to capture the slight demand-reducing effects of adding more arterial capacity; that is, additional arterial capacity allows drivers to ultimately discover new routes to shorten the same trips.
References AASHO (American Association of State Highway Officials), 1957. A Policy on Arterial Highways in Urban Areas. AASHO, Washington, DC. Alcott, B., 2005. Jevon's paradox. Ecol. Econ. 54 (1), 9–21. Barr, L., 2000. Testing for the significance of induced highway travel demand in metropolitan areas. Transp. Res. Rec. 1706, 1–8. Bullen, N., Jones, K., Duncan, C., 1997. Modelling complexity: analysing between-individual and between-place variation—a multilevel tutorial. Environ. Plan. A 29 (4), 585–609. Cervero, R., Hansen, M., 2002. Induced travel demand and induced road investment: a simultaneous equation analysis. J. Transp. Econ. Pol. 36 (3), 469–490. Chakravarty, D., Dasgupta, S., Roy, J., 2013. Rebound effect: how much to worry? Curr. Opin. Environ. Sustain. 5 (2), 216–228. Downs, A., 1962. The law of peak-hour expressway congestion. Traffic Q. 16 (3), 393–409. Duncan, C., Jones, K., 2000. Using multilevel models to model heterogeneity: potential and pitfalls. Geogr. Anal. 32 (4), 279–305. Duranton, G., Turner, M., 2011. The fundamental law of road congestion: evidence from US cities. Am. Econ. Rev. 101 (6), 2616–2652. Frondel, M., Vance, C., 2013. Don't belittle the rebound effect. Nature 494 (7438), 430. Gillingham, K., Kotchen, M., Rapson, D., Wagner, G., 2013. The rebound effect is overplayed. Nature 493 (7433), 475–476. Goodwin, P., 1996. Empirical evidence on induced traffic: a review and synthesis. Transportation 23 (1), 35–54. Greening, L., Greene, D., Difiglio, C., 2000. Energy efficiency and consumption—the rebound effect—a survey. Energy Pol. 28 (6–7), 389–401. Handy, S., Boarnet, M., 2014. Impact of Highway Capacity and Induced Travel on Passenger Vehicle Use and Greenhouse Gas Emissions. Air Resources Board, Sacramento, CA. Hansen, M., Huang, Y., 1997. Road supply and traffic in California urban areas. Transp. Res. A 31 (3), 205–218. Hills, P., 1996. What is induced traffic? Transportation 23 (1), 5–16. Holtzclaw, J., 1991. Explaining Urban Density and Transit Impacts on Auto Use. National Resources Defense Council, San Francisco, CA. Holtzclaw, J., 1994. Using Residential Patterns and Transit to Decrease Auto Dependence and Costs. National Resources Defense Council, San Francisco, CA. Hymel, K., Small, K., Van Dender, K., 2010. Induced demand and rebound effects in road transport. Transp. Res. B 44 (10), 1220–1241. Jenkins, J., Nordhaus, T., Shellenberger, M., 2011. Energy Emergence: Rebound & Backfire as Emergent Phenomena. The Breakthrough Institute, Oakland, CA. Jevons, W., 1865. The Coal Question: An Enquiry Concerning the Progress of the Nation, and the Probable Exhaustion of Our Coal-Mines. Macmillan, London, UK. Jiwattanakulpaisarn, P., Noland, R., Graham, D., 2012. Marginal productivity of expanding highway capacity. J. Transp. Econ. Pol. 46 (3), 333–347. Khazzoom, J., 1980. Economic implications of mandated efficiency in standards for household appliances. Energy J. 1 (4), 21–40. Liddle, B., 2009. Long-run relationship among transport demand, income, and gasoline price for the US. Transp. Res. D 14 (2), 73–82. Litman, T., 2001. Generated traffic: implications for transportation planning. ITE J. 71 (4), 38–47. Litman, T., 2013. Changing North American vehicle-travel price sensitivities: implications for transport and energy policy. Transp. Policy 28, 2–10. Noland, R., 2001. Relationships between highway capacity and induced vehicle travel. Transp. Res. A 35 (1), 47–72. Noland, R., Cowart, W., 2000. Analysis of metropolitan highway capacity and the growth in vehicle miles of travel. Transportation 27 (4), 363–390. Noland, R., Lem, L., 2002. A review of the evidence for induced travel and changes in transportation and environmental policy in the US and the UK. Transp. Res. Part D: Transp. Environ. 7 (1), 1–26. Raudenbush, S., Bryk, A., 2002. Hierarchical Linear Models: Applications and Data Analysis Methods. Sage, Thousand Oaks, CA. Rentziou, A., Gkritza, K., Souleyrette, R., 2012. VMT, energy consumption, and GHG emissions forecasting for passenger transportation. Transp. Res. A: Policy Pract. 46 (3), 487–500. Schipper, L., 2000. On the rebound: the interaction of energy efficiency, energy use and economic activity. An introduction. Energy Pol. 28 (6–7), 351–353. Schrank, D., Eisele, W., Lomax, T., 2012. Urban Mobility Report. Texas Transportation Institute, College Station, TX. Small, K., Van Dender, K., 2007. Fuel efficiency and motor vehicle travel: the declining rebound effect. Energy J. 28 (1), 25–51. Sorrell, S., 2007. The Rebound Effect: An Assessment of the Evidence for Economy-Wide Energy Savings from Improved Energy Efficiency. UK Energy Research Centre, London, UK. Sorrell, S., Dimitropoulos, J., Sommerville, M., 2009. Empirical estimates of the direct rebound effect: a review. Energy Pol. 37 (4), 1356–1371. Stern, P., 1984. Improving Energy Demand Analysis. National Academy Press, Washington, DC. Su, Q., 2011. Induced motor vehicle travel from improved fuel efficiency and road expansion. Energy Pol. 39 (11), 7257–7264. Tierney, J., 2011. When Energy Efficiency Sullies the Environment. The New York Times
7. Conclusions The contributions of the results to the empirical literature on induced demand are attributable to the data and to the design of the study. The use of disaggregate data on travel behavior unmasks how drivers, rather than groups of drivers, respond to changes in road capacity. The decision to disaggregate freeway lane-kilometers from arterial lane-kilometers rather than to aggregate both to proxy for stock and for flow measures of road capacity (Noland and Cowart, 2000) highlights the differential effects of freeways and of arterials on VKT. To that end, the most important contribution of the study to the induced travel demand literature is that adding more arterial capacity actually decreases VKT. However, the demand-reducing effects of additional arterial capacity are not as consequential to private-vehicle travel behavior as are vehicle characteristics such as gasoline price. The most fruitful direction for future research is to explore the suggestion that the demand-inducing effect of adding road capacity is dependent on the scale of the preexisting road capacity. To return to the example in the discussion section, the same percentage increase in capacity may induce more demand in one metropolitan area versus another metropolitan area where the lane-kilometers of preexisting capacity is greater. One approach with the potential to capture differences in the scale of the capacity between metropolitan areas is to use a geographic information system to map where additional road capacity is added and how that new capacity affects the balance of the road network across urban, suburban, and exurban locations. The expectation is that adding road capacity in more peripheral locations will engender more private-vehicle travel. Another approach is to pool disaggregate data from the next NHTS with the disaggregate data from the 2001 NHTS and from the 2009 NHTS to extend the lag period for the respective flow measures of road capacity, road congestion, commuter demand, and economic growth. To that end, the six year and the eight 8
Journal of Transport Geography 66 (2018) 1–9
E.J. Zolnik
USCM (U.S. Conference of Mayors), 2010. U.S. Metro Economies: Pace of Economic Recovery: GMP and Jobs. IHS Global Insight, Lexington, MA. USDOT (U.S. Department of Transportation), 2017. Historical VMT from 1970. Washington, DC Online at: https://www.fhwa.dot.gov/policyinformation/travel_ monitoring/tvt.cfm (accessed 09.28.17). USDOT (U.S. Department of Transportation), 2001. 2001 National Household Travel Survey. USDOT, Washington, DC Online at: http://nhts.ornl.gov/download.shtml# 2001 (accessed 11.11.14). USDOT (U.S. Department of Transportation), 2009. 2009 National Household Travel Survey. USDOT, Washington, DC Online at: http://nhts.ornl.gov/download.shtml# 2009 (accessed 11.11.14). USEIA (U.S. Energy Information Administration), 2014. Energy Efficiency Information. USEIA, Washington, DC Online at: https://www.eia.gov/ (accessed 11.11.14). USEPA (U.S. Environmental Protection Agency), 2014. Greenhouse Gas Emissions from a Typical Passenger Vehicle. USEPA, Washington, DC.
March 7. Online at: http://www.nytimes.com/2011/03/08/science/08tier.html. TRB (Transportation Research Board), 1995. Expanding Metropolitan Highways: Implications for Air Quality and Energy Use. National Academy Press, Washington, DC. TTI (Texas Transportation Institute), 2012. Congestion Data for Your City. TTI, College Station, TX Online at: http://mobility.tamu.edu/ums/congestion-data/ (accessed 08.01.15). TTI (Texas Transportation Institute), 2014. Congestion Data for Your City. TTI, College Station, TX Online at: http://mobility.tamu.edu/ums/congestion-data/ (accessed 04.24.16). Turner, K., 2013. “Rebound” effects from increased energy efficiency: a time to pause and reflect. Energy J. 34 (4), 25–42. USCM (U.S. Conference of Mayors), 2000. U.S. Metro Economies: The Engines of America's Growth. Standard & Poor's, Lexington, MA. USCM (U.S. Conference of Mayors), 2003. The Role of Metro Areas in the US Economy. Global Insight, Lexington, MA.
9