VMT, energy consumption, and GHG emissions forecasting for passenger transportation

Transportation Research Part A 46 (2012) 487–500

Contents lists available at SciVerse ScienceDirect

Transportation Research Part A journal homepage: www.elsevier.com/locate/tra

VMT, energy consumption, and GHG emissions forecasting for passenger transportation Aikaterini Rentziou b, Konstantina Gkritza a,⇑, Reginald R. Souleyrette c a

Department of Civil, Construction & Environmental Engineering, 404 Town Engineering Building, Iowa State University, Ames, IA 50011-3232, United States School of Civil Engineering, National Technical University of Athens, Athens, Greece c Department of Civil Engineering, University of Kentucky, 259 Raymond Building, Lexington, KY 40506, United States b

a r t i c l e

i n f o

Article history: Received 11 April 2011 Received in revised form 27 November 2011 Accepted 29 November 2011

Keywords: VMT Energy consumption GHG Passenger transportation SURE Random parameters

a b s t r a c t Globalization, greenhouse gas emissions and energy concerns, emerging vehicle technologies, and improved statistical modeling capabilities make the present moment an opportune time to revisit aggregate vehicle miles traveled (VMT), energy consumption, and greenhouse gas (GHG) emissions forecasting for passenger transportation. Using panel data for the 48 continental states during the period 1998–2008, the authors develop simultaneous equation models for predicting VMT on different road functional classes and examine how different technological solutions and changes in fuel prices can affect passenger VMT. Moreover, a random coefficient panel data model is developed to estimate the influence of various factors (such as demographics, socioeconomic variables, fuel tax, and capacity) on the total amount of passenger VMT in the United States. To assess the influence of each significant factor on VMT, elasticities are estimated. Further, the authors investigate the effect of different policies governing fuel tax and population density on future energy consumption and GHG emissions. The presented methodology and estimation results can assist transportation planners and policy-makers in determining future energy and transportation infrastructure investment needs. Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction Passenger transportation accounts for almost 18% of the energy consumption and 22% of the total greenhouse gas (GHG) emissions in the United States (US); and approximately 63% of the transportation sector’s total energy consumption and 73% of the transportation sector’s total GHG emissions (US Department of Energy, 2008; US Environmental Protection Agency, 2008). Fuel use is predominantly related to vehicle miles traveled (VMT), the most commonly cited measure of passenger transportation in the US. As the availability of energy and funding resources for new infrastructure decreases, accurate forecasting of VMT is crucial for effective energy and transportation investment planning. Globalization, GHG emissions and energy concerns, emerging vehicle technologies, and improved statistical modeling capabilities make the present moment an opportune time to revisit aggregate VMT, energy consumption, and GHG emissions forecasting for passenger transportation. The influence of various factors on VMT has been investigated extensively in previous studies. Population is considered to be one of the most important factors influencing passenger transportation, and population growth has been the principal contributor to the growth of VMT (Fulton et al., 2000; Greene et al., 1995; Heanue, 1998; Maples, 2001; National Surface Transportation Policy, 2007a,b; Noland, 2001; Polzin et al., 2004; Souleyrette et al., 1995). Moreover, various other demographic and socioeconomic factors have been examined in previous studies and found to contribute to a change in VMT, such ⇑ Corresponding author. Tel.: +1 515 294 2343; fax: +1 515 294 7424. E-mail addresses: [email protected] (A. Rentziou), [email protected] (K. Gkritza), [email protected] (R.R. Souleyrette). 0965-8564/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.tra.2011.11.009

488

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500

as income (Brazil and Purvis, 2009; Brownstone and Golob, 2009; Burchell et al., 2002; Fulton et al., 2000; Greene et al., 1995; Heanue, 1998; Hu et al., 2000; Kweon and Kockelman, 2004; Liddle, 2009; Litman, 2005; Mabe, 2007; Maples, 2001; McGuckin and Liss, 2005; National Surface Transportation Policy, 2007a,b; Noland, 2001), age (Brownstone and Golob, 2009; Greene et al., 1995; Litman, 2005; Maples, 2001; National Surface Transportation Policy, 2007b), gender (Bagley and Mokhtarian, 2002; Burchell et al., 2002; Greene et al., 1995; Maples, 2001; National Surface Transportation Policy, 2007b), race (Brownstone and Golob, 2009; Contrino and McGuckin, 2009; Litman, 2005), level of education, and number of children (Brownstone and Golob, 2009). For example, VMT would increase with increasing income, level of education, higher number of children in a household, and be higher in predominantly White households. Population age structure’s influence on VMT has been found to be significant but variable across cohorts; for example, Liddle (2011) found a positive relationship between VMT and young adults (20–34 years old), and a negative one for the other cohorts. Note that age is highly correlated with number of workers and number of drivers per household. The increase in the number of workers and number of drivers per household would increase VMT as well (Barr, 2000; Brazil and Purvis, 2009; Brownstone and Golob, 2009; Heanue, 1998; Hu et al., 2000; Kweon and Kockelman, 2004). In addition to demographic and socioeconomic characteristics, fuel cost or trip cost also have been identified as factors affecting VMT (Congressional Budget Office, 2008a,b; Greene et al., 1995; Heanue, 1998; Hu et al., 2000; Liddle, 2009; Mabe, 2007; Maples, 2001; Noland, 2001; Southworth, 2001). However, it has been noted that increased fuel cost affects VMT only when that cost is high, for example, as high as the fuel prices experienced in 2008 (Congressional Budget Office, 2008a,b). Moreover, it has been shown that the elasticity of transport-related fuel demand in the US declines as incomes increase (Small and Van Dender, 2007), and overall travel demand (captured by VMT) with respect to fuel costs is highly inelastic (Congressional Budget Office, 2008a,b; Mabe, 2007; Noland, 2001). The historical (1966–2004) long-run elasticity of VMT with respect to fuel cost has been estimated at 0.210, while the corresponding moderately recent elasticity (2000– 2004) is much lower (0.057) (Small and Van Dender, 2007). In addition, vehicle availability and number of vehicles contribute to an increase in VMT (Bagley and Mokhtarian, 2002; Brazil and Purvis, 2009; Kweon and Kockelman, 2004; McGuckin and Liss, 2005; National Surface Transportation Policy, 2007a; Polzin, 2006; Souleyrette et al., 1995). Past work (Choo and Mokhtarian, 2007; Litman, 2005; Southworth, 2001) has also indicated the positive and negative impacts that new technologies that enabled telecommuting have had on VMT. Land use is considered to be one of the major factors influencing VMT, and various indicators, such as density or urban development, have been examined in past work (Barr, 2000; Brownstone and Golob, 2009; Chatman, 2008; Committee for the Study on the Relationships among Development Patterns, Vehicle Miles Traveled, and Energy Consumption, 2009; Fang, 2008; Holtzclaw, 1994; Kweon and Kockelman, 2004; National Surface Transportation Policy, 2007b; Newman and Kenworthy, 1999; Su, 2010). The influence of density varies from study to study, with elasticity values ranging from as low as 4–6% (Barr, 2000) up to 25–30% for a doubling of the density (Holtzclaw, 1994). Compact development and mixed land uses result in lower VMT (Cervero and Duncan, 2006; Ewing et al., 2007; Gómez-Ibánez and Humphrey, 2010), while suburbanization and sprawl development have the opposite effect on VMT (Burchell et al., 2002; National Surface Transportation Policy, 2007b; Southworth, 2001). VMT is lower in urbanized areas than in rural areas (Brazil and Purvis, 2009), and it is even lower closer to central business districts (Boarnet and Crane, 2001). The effect of transit availability and level of accessibility on VMT has also been documented in past studies (Cervero and Duncan, 2006; Holtzclaw, 1994; Krizek, 2003; Kweon and Kockelman, 2004; Litman, 2005). The length of the road network and existing capacity are two additional factors affecting VMT. The number of lane miles is typically used to estimate the effect of road network length on VMT (Fulton et al., 2000; Noland and Cowart, 2000; Noland, 2001), and the estimated elasticity ranges between 0.3 and 0.6 in the short-run and between 0.7 and 1.0 in the long-run (Noland, 2001). Increased lane mileage, road density, and capacity (Contrino and McGuckin, 2009; Strathman et al., 2000; Su, 2010) lead to an increase in VMT and can induce significant additional travel (Noland, 2001). Moreover, the decreased travel time due to increased capacity contributes to an increase in VMT (Barr, 2000; Transportation Research Board, 1995), while urban congestion has a negative and statistically significant effect on travel demand (Su, 2010). A positive effect on VMT has been observed with increases in trip length (McGuckin and Liss, 2005; Polzin et al., 2004). While previous studies have investigated the effect of demographic and socioeconomic characteristics, land use, road capacity, and fuel prices on VMT, the effect of technological solutions (such as telecommuting and alternative fuel vehicles) and fuel tax on VMT has not been fully examined. For example, the use of alternative fuel vehicles is expected to impact VMT, given the attributes of these vehicles (such as low fuel cost, or limited range of trip) as well as the specific characteristics and perceptions of the owners of these vehicles, such as increased environmental awareness or level of income (Turcksin et al., 2008), enhancing the need for the analysis of this technology’s impact on VMT. Using panel data for the 48 continental states during the period 1998–2008, the authors develop simultaneous equation models for predicting passenger VMT on different road functional classes and area classifications and examine how changes in various factors (such as demographics, socioeconomic variables, fuel cost, fuel tax, capacity and technological solutions) can affect passenger trips across the nation. To assess the influence of each significant factor on VMT, direct- and cross-elasticities are estimated. Further, the authors propose a methodology to investigate how different policies governing fuel tax and density might affect future energy consumption and GHG emissions in the US. First, a random coefficient panel data model is developed to estimate the total amount of passenger VMT in the US as a function of demographics, socioeconomic variables, fuel cost, and capacity. The development of a random parameter model for estimating total VMT allows the coefficients to vary so as to

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500

489

account for variation across population that cannot be captured by ordinary regression models. Then, the VISION software, developed by Argonne National Laboratory, is used for estimating the reduction in energy consumption and GHG emissions as a result of two hypothetical policies that can have a significant impact on VMT: increased state fuel tax and increased density. The estimated models of passenger VMT and results can assist transportation planners and policy-makers in predicting future energy and transportation infrastructure investment needs. 2. Data Various data sources were used in this study. VMT data for the 48 continental states in the US (excluding Alaska and Hawaii) from 1998–2008 were obtained from the Federal Highway Administration (FHWA) and were based on the Highway Performance Measurement System (HPMS). Note that this paper examines only VMT by passenger cars, motorcycles, light trucks and buses and, as such, truck VMT data are excluded. The VMT data were disaggregated by road type and urban and rural classifications.1 Information about demographic (population, age, race) and socioeconomic factors (income, percentage of people working at home, density) for each state was obtained from the US Census Bureau (1990, 2000, 2008). Data on fuel cost were based on the ‘‘Monthly Motor Fuel Reported by States’’ available from FHWA. Data on fuel taxes, highway lane miles, level of congestion (volume over capacity ratio), and vehicle registrations by state were obtained from FHWA Highway Statistics. Lastly, data for alternative fuel vehicles by state, which include compressed natural gas (CNG), electricity, ethanol, methanol, liquefied natural gas (LNG), and liquefied petroleum gas (LPG)/propane were provided by the US Department of Energy and the Energy Information Administration (2010a). Table 1 presents the descriptive statistics for select variables used in this study. The mean values represent the averages over the data sample used in this study (i.e., 48 states over 11 years, or a total of 528 observations). During the analysis period, total VMT in rural areas decreased, while urban VMT increased. Total population grew by 11%, while the percentage of urban population increased from 72% to 79%. The distribution of population by gender (49% male, 51% female) did not change, but the population of people 65 and older increased at a faster rate than the population of people younger than 18 years old. The Hispanic population experienced the greatest increase compared to other minority populations. The percentage of people working at home experienced a fivefold increase since 2000, mainly due to the increased use of the Internet, which has facilitated telecommuting. Fuel cost has increased since 1998, with the increase being higher during recent years and especially higher after 2004. Average density (population per square mile) also increased by 7.5%. Lastly, total rural lane mileage decreased slightly, while total urban lane mileage has increased since 2000. 3. Methodology VMT is a continuous variable that can take on several values. Because the dependent variable is thus continuous, a linear regression model was developed to determine the factors affecting the dependent variable. To determine elasticity for each variable, log-linear regression models were estimated. Logarithmic transformations of the independent variables were introduced in the models to minimize any heteroskedasticity in the cross-sectional data that can arise from combining data from states of different size and population (Noland, 2001). 3.1. Simultaneous equation models In many cases, transportation data are best modeled using a system of interrelated equations. Simultaneous equation models are used in cases where the dependent variable of one equation is the independent variable of the other equation or in cases where the dependent variables are correlated. If dependent variables are analyzed separately, correlation between regressors and disturbances will arise, violating key assumptions of best linear unbiased estimators (Washington et al., 2011). 3.1.1. Seemingly unrelated regression equations Seemingly unrelated regression equation (SURE) modeling is one category of equation model system that is useful in cases where the dependent variables are considered as a group but do not have a direct interaction (as would variables in simultaneous equations). This model is appropriate in cases where the factors that influence dependent variables are the same or where the dependent variables share common characteristics. In such cases, the equations are seemingly unrelated, but contemporaneous correlation of error terms will exist. If equations are estimated separately by ordinary least squares (OLS), coefficients are consistent but not efficient. Efficient parameters can be achieved only by considering the contemporaneous correlation among the disturbances (Washington et al., 2011). The general equation for seemingly unrelated models is (Zellner, 1962)

yi ¼ X i bi þ ei ;

1

i ¼ 1; . . . ; M

ð1Þ

The reader is referred to FHWA (1989) for additional information on the definition and characteristics of the functional systems for urban and rural areas.

490

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500

Table 1 Descriptive statistics for panel data (528 observations). Variables

Mean (or percentage)

Standard deviation

Vehicle miles traveled in rural areas (billions) Interstate Principal arterial Minor arterial Collector Total

4.07 4.26 3.08 3.79 15.2

2.87 3.31 2.47 3.10 11.05

Vehicle miles traveled in urban areas (billions) Interstate Freeways Other principal arterials Minor arterials Major collector Total

7.89 3.79 8.28 6.82 3.00 29.78

10.08 7.69 9.75 7.95 3.71 38.00

Population (millions) Percentage of urban population Percentage of White population Percentage of Black and African–American population Percentage of Hispanic or Latino population Percentage of Asian population Population under 18 years old (millions) Population 65 years old and over Percentage of male population Percentage of female population Income per capita (real dollars) Percentage of population working at home (telecommuting)a

5.99 67.03 79.09 10.11 8.47 2.33 1.51 746,516 49.18 50.82 31,021 17.25

6.41 15.51 10.19 9.53 9.41 2.04 1.66 768,776 0.66 0.66 5859 11.41

Fuel cost (cents/gallon) Fuel tax-state (cents/gallon) Total fuel tax (cents/gallon) Density (population per square mile) Vehicle registrations (millions) Vehicles per capita Percentage of alternative fuel vehicles

192.69 20.85 39.25 189.05 2.80 0.46 0.23

68.47 4.85 4.85 253.84 3.16 0.07 0.15

Lane miles-Rural Interstate Principal arterial Minor arterial Collector Total

642.29 2002.06 2824.17 8834.17 14,302.7

386.30 1228.40 1939.00 6801.41 9847.39

Lane miles-Urban Interstate Freeways Other principal arterial Minor arterial Major collector Total

302.53 206.35 1196.13 1988.52 2028.49 5722.02

258.85 291.63 1200.28 1933.05 2125.44 5710.2

V/C-rural interstateb 0.80–0.95 >0.95 Percentage of congested milesc

24.61 10.82 7.61

41.60 19.85 11.74

V/C-rural principal arterialb 0.80–0.95 >0.95 Percentage of congested miles

21.20 19.21 3.02

33.27 30.31 5.09

V/C-rural minor arterialb 0.80–0.95 >0.95 Percentage of congested miles

18.23 16.49 1.74

51.36 38.78 3.38

V/C-rural collectorb 0.80–0.95 >0.95 Percentage of congested miles Percentage of congested miles on rural roads

12.54 10.32 0.37 1.42

43.83 30.93 0.9 1.96

V/C-urban interstateb 0.80–0.95

52.59

63.30

491

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500 Table 1 (continued)

a b c d

Variables

Mean (or percentage)

Standard deviation

>0.95 Percentage of congested miles

60.90 30.28

84.94 18.72

V/C-urban freewaysb 0.80–0.95 >0.95 Percentage of congested milesd

26.63 29.43 20.19

54.32 67.22 15.54

V/C-urban other principal arterialb 0.80–0.95 >0.95 Percentage of congested miles

83.99 75.00 11.71

102.34 102.68 7.72

V/C-urban minor arterialb 0.80–0.95 >0.95 Percentage of congested miles

102.77 111.55 9.34

136.58 141.16 5.76

V/C-urban major collectorb 0.80–0.95 >0.95 Percentage of congested miles Percentage of congested miles on urban roads

61.97 78.18 5.86 10.05

92.92 111.56 4.01 4.99

Based on 432 observations. V/C: volume over capacity ratio. Based on 517 observations. Based on 477 observations.

where yi is the T  1 vector of observed values on the ith dependent variable; Xi the T  pi matrix with rank pi of observations on pi independent variables; bi the pi  1 vector of unknown regression coefficients; ei is the T  1 vector of error terms. It is assumed that e = (e1, e2, . . ., eM) has a multivariate normal density with mean E[e] = 0 and covariance E½ee0  ¼ R  IT ¼ V. Using generalized least squares (see Pindyck and Rubinfeld, 1998; Washington et al., 2011) a best linear unbiased estimator is obtained as:

^ ¼ ½X 0 ðR  IT Þ1  X1  X 0 ðR  IT Þ1  Y b

ð2Þ

For the present analysis, the SURE model was selected to estimate VMT on different road functional classes in rural and urban areas. This type of model was considered appropriate for this study because VMT values for different functional classes on either rural or urban roads are highly correlated (correlation coefficients range from 0.787 to 0.973). The authors estimated cross-elasticities to quantify and update these relationships among different functional classes. Moreover, the factors that affect VMT are common among the functional classes (another reason for selecting the SURE methodology). A dummy variable for each state was included in the models so that a fixed-effects,2 or dummy-variable, SURE model could be estimated and the unmeasured factors that affect VMT and are associated with each state could be accounted for. Note that, while this methodology was used in (Noland, 2001) for predicting total VMT (both passenger and freight) on three road types (Interstates, arterials, and collectors), the impact of one road type’s lane miles or capacity on the VMT of another road type was not examined. Lastly, dynamic adjustment of VMT over time is not considered in this study and as such, the estimated effects are intended to be interpreted as short-term effects. 3.2. Panel data regression models with random parameters In addition to estimating VMT by functional class and type of area, the authors estimated an aggregate model for total passenger VMT for forecasting future energy consumption and GHG emissions. Panel data models with random parameters were chosen for the analysis of total VMT because this methodology provides various benefits and overcomes some of the limitations of time-series and cross-section studies (such as heterogeneity and multicollinearity problems) (Baltagi, 2008). 3.2.1. Two-way error component regression model For panel data, the equation for a two-way error component model is written (Baltagi, 2008)

Y it ¼ a þ X 0it b þ li þ kt þ mit ;

i ¼ 1; . . . ; n;

t ¼ 1; . . . ; T

ð3Þ

2 The authors examined the applicability of first differencing (FD) in lieu of fixed effects (FE) estimation. Following Wooldridge (2002, p. 284), the FE estimator is more efficient under the assumption that the error terms are serially uncorrelated. However, the Durbin–Watson statistics for the SURE models (under the FE estimation) in Tables 2 and 3 were all around 1.8–1.9, which indicate very little presence of serial correlation (0.05–0.1).

492

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500

where a is the constant; Xit the set of explanatory variables, independent of mit for all i, t. li the unobserved cross-sectional specific effect; kt the unobserved time effects; mit the random disturbances; b is the coefficients of explanatory variables. The two-way error component regression model simultaneously accounts for the effects of different states and the effect of time (year) on the dependent variable. The li and kt variables are assumed to be fixed parameters to be estimated, and the mit variables are random disturbances, following the usual regression assumptions (Washington et al., 2011). The two-way error component regression model is appropriate to estimate the influence of any time-specific effect that is not included in the regression, such as the effect of hurricane Katrina in 2005 (Baltagi, 2008). 3.2.2. Random parameters Variables estimated with panel data and two-way error models have constant coefficients. However, in many cases the cross-sectional units examined possess different unobserved demographic and socioeconomic factors that result in response variables that vary over time and different cross-sectional units (Hsiao, 2003). In a random parameters and random coefficients approach, specific assumptions are made about the distributions that each coefficient follows, and the model is far more flexible and has more degrees of freedom than a fixed-effects slope coefficient approach (Biorn et al., 1998). If the model parameters are needed to account for individual cross-sectional unit heterogeneity and for specific time periods, the equation of the developed model should be written (Washington et al., 2011)

Y it ¼ Rðbkit X kit Þuit

ð4Þ

where,

uit ¼ li þ kt þ mit ;

i ¼ 1; . . . ; n;

t ¼ 1; . . . ; T

ð5Þ

bkit random coefficients such that

bkit ¼ bk þ aki þ kkt

ð6Þ

The aki and kkt variables are allowed to be random variables and introduce proper stochastic specifications. The random coefficients reduce the number of parameters to be estimated substantially, while still allowing the coefficients to differ from unit to unit and/or from time to time (Hsiao and Pesaran, 2004). 4. Estimation results of VMT forecasting models 4.1. Estimation of VMT on rural roads A SURE model was estimated to examine simultaneously the effect of different factors on VMT on four different functional classes (Interstate, principal arterial, minor arterial, collector) in rural areas. Table 2 shows the estimation results for VMT on different functional classes of rural roads. The dependent variable is the natural logarithm of VMT (log-VMT). The number of vehicle registrations and the amount of lane miles are likely to be endogenous in our estimation. To resolve this estimation problem, we estimated regression models to exogenously predict the number of vehicle registrations and the amount of lane miles as a function of income, race, gender, percentage of people telecommuting, density, fuel cost, fuel tax, and percentage of alternative fuel vehicles. The predicted values were then introduced in the SURE model.3 The variables presented in Table 2 are significant at the 95% confidence interval or higher. Fuel taxes, percentage of alternative fuel vehicles, as well as additional demographic variables were not found significant on any of the four equations of the SURE model. According to the estimation results shown in Table 2, race, gender, and income per capita are the socioeconomic factors affecting VMT on rural roads, a result consistent with previous studies (Bagley and Mokhtarian, 2002; Brazil and Purvis, 2009; Brownstone and Golob, 2009; Contrino and McGuckin, 2009; Fulton et al., 2000; Greene et al., 1995; Heanue, 1998; Liddle, 2009; Mabe, 2007; Maples, 2001; McGuckin and Liss, 2005; National Surface Transportation Policy, 2007a; Noland, 2001). A 1% increase in the income per capita would increase VMT on rural collector roads by 0.853%; the magnitude of this effect is similar to that noted in (Noland, 2001). Interestingly, the increase in male population has a positive effect on VMT on Interstates and a negative effect on VMT on collector roads. This may indicate that female drivers mainly undertake shorter trips to local destinations (traveled on rural collector roads), while male drivers dominate long-distance travel. Moreover, a 1% increase in fuel cost would decrease VMT on Interstates by 0.035%. This elasticity value is lower compared to other estimates reported in the literature. Also, new technological solutions that reduce the need for travel, expressed as the percentage of people telecommuting, result in a decrease in VMT on collector roads, as might be expected. Higher population density leads to a decrease in VMT on rural principal arterials. The effect has been indicated in past studies but mostly, in urban settings where non-motorized modes of transportation are prevalent. The amount of lane miles and length of the network were found to influence VMT on Interstate, minor arterial, and collector roads significantly. This result is consistent with the results of previous studies (Fulton et al., 2000; Noland and Cowart, 2000; Noland, 2001). The adopted model specification allowed the authors to investigate cross-elasticities. For 3 The test of the relevance of the instrumental variables (Stock and Watson, 2003, p. 350) suggested that our instruments were valid in this particular application.

493

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500 Table 2 SURE model estimation results for log-VMT on rural roads. Independent variables

Interstate

Principal arterial

Minor arterial

Collector

Constant Percentage of White population Percentage of African American or Black population Percentage of Hispanic or Latino population Percentage of male population Natural logarithm of income per capita Percentage of population working at home (telecommuting) Natural logarithm of fuel cost

2.137 0.024** 0.025** 0.003** 0.025*

3.863 0.013** 0.01**

1.481 0.024** 0.019**

13.049

Density Natural logarithm of Interstate lane miles Natural logarithm of minor arterial lane miles Natural logarithm of collector lane miles Natural logarithm of vehicle registrations Percentage of congested miles on Interstates Percentage of congested miles on minor arterials Alabama Arizona Arkansas California Colorado Connecticut Florida Georgia Illinois Indiana Iowa Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Montana Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina North Dakota Ohio Oregon Pennsylvania South Carolina South Dakota Tennessee Texas Utah Vermont Washington West Virginia Wisconsin Wyoming Year 2000 Year 2002 Goodness of fit measure (R2) F-statistic (p-value) Durbin–Watson (autocorrelation) Number of observations * ** a

Variables significant at the 95% confidence interval. Variables significant at the 99% confidence interval. Indicates cross-elasticity.

0.220** 0.853** 0.003**

0.035** 0.001** 0.244** 0.336** 0.479**

0.669** 0.0004**,a

0.432**,a 0.376**

0.648** 0.005**,a

0.169** 0.097** 0.205**

0.157** 0.189**

0.159** 0.104**

0.068** 0.190**

0.284** 0.140**

0.230** 0.085**

0.255** 0.427** 0.142**

0.109** 0.097** 0.076** 0.148**

0.107** 0.137**

0.095**

0.238** 0.043* 0.223** 0.126** 0.240** 0.134** 0.253** 0.145**

0.258** 0.053**

0.144** 0.123** 0.279** 0.467** 0.115**

0.643**

0.149**

0.153** 0.087*

0.104** 0.111** 0.180** 0.155** 0.200**

0.224** 0.096**

0.087** 0.125**

0.163** 0.286** 0.214** 0.092* 0.250**

0.232** 0.102** 0.435** 0.286** 0.122** 0.211** 0.214** 0.247**

0.010* 0.869 117.18 (0.0000) 1.804 (0.098) 419

0.130** 0.172** 0.363** 0.017** 0.011**

0.899 90.62 (0.0000) 1.887 (0.056) 419

0.879 125.47 (0.0000) 1.843 (0.078) 419

0.910 137.38 (0.0000) 1.757 (0.121) 419

494

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500

example, the amount of minor arterial lane miles affects VMT on collectors, and, as shown in Table 2, this effect (elasticity of 0.432) is higher than the effect of collector lane miles on VMT on collector roads (elasticity of 0.376). This result highlights the significance of examining the rural highway network as a system for investment planning purposes. Two other results that support this statement include the influence of Interstate congestion on VMT on principal arterials and the influence of minor arterial congestion on VMT on collector roads. The increased percentage of congested miles on Interstates increases the amount of VMT on principal arterials as travel demand shifts, and, likewise, the increased percentage of congested miles on minor arterials shifts demand to collector roads. Moreover, an increase in the number of vehicles results in an increase in VMT on Interstates, principal arterials, and minor arterials, with an elasticity that ranges from 0.48 to 0.67; this finding is well-documented in past work (Bagley and Mokhtarian, 2002; Brazil and Purvis, 2009; Kweon and Kockelman, 2004; McGuckin and Liss, 2005; National Surface Transportation Policy, 2007a; Polzin, 2006; Souleyrette et al., 1995). Lastly, VMT on rural roads varies by state and over time, as indicated by the different coefficients for different states and years. 4.2. Estimation of VMT on urban roads A SURE model was estimated to examine simultaneously the effect of different factors on VMT on the five different functional classes in urban areas. Table 3 shows the estimation results for VMT on different functional classes of urban roads. Note that the dependent variable is the natural logarithm of VMT (log-VMT). As was the case with the estimation of VMT on rural roads, predicted values for the number of vehicle registrations and amount of lane miles were introduced in the SURE model.4 The variables included in Table 3 are significant at the 95% confidence interval or higher. The percentage of population working at home and other demographic variables were not found significant when introduced in this model. Gender and race are significant socioeconomic factors affecting VMT on urban roads, a result also consistent with previous studies. It is interesting that in the case of urban roads, an increase in male population results in a decrease in VMT on Interstates and principal arterials, while in the case of rural roads, an increase in male population results in an increase in VMT on Interstates and a decrease in VMT on collectors. It is difficult to speculate what may be driving these differences in the coefficient values across road types. Disaggregate demographic information at the household-level (such as distribution of single adult households by area) could shed some light on these findings. Turning to the effect of fuel cost and fuel taxes on urban VMT, the analysis showed that increasing fuel cost would result in a decrease in VMT on Interstates, while increase of fuel tax would result in a decrease of VMT on freeways, principal arterials and minor arterials in urban areas. The effect across all four road classes is significant and highly inelastic. Population density is another important factor determining urban VMT, and an increase in density results in a decrease in urban VMT on minor arterials. A 1% increase in vehicle registrations leads to a 1.333% increase in VMT on freeways, while the effect of vehicle registrations on VMT on principal arterials and collectors is still positive but lower than that on freeways. Interestingly, an increase in the percentage of alternative fuel vehicles would result in a decrease in VMT on Interstates and an increase in VMT on freeways. This result could be attributed to either the characteristics of alternative fuel vehicles (being vehicles with different capabilities and limitations than other traffic) or to the attributes of alternative fuel vehicle owners, such as increased environmental awareness, that may impact the travel patterns. In addition, alternative fuel vehicles include plug-in electric vehicles that have a specific range of miles that can be driven during one trip, and, as such, that range would determine the length of trip and influence VMT on different functional classes. Regarding the length of the network, increasing the number of lane miles increases VMT on the corresponding functional classes. The elasticity ranges from 0.272 to 0.531, values much higher than the elasticity estimated for rural VMT with respect to network miles. This finding suggests that demand in urban areas is more sensitive to changes in supply, as might have been anticipated due to the higher levels of congestion as well as greater availability of other modes of transportation in urban areas than in rural areas. In addition, expanding the freeway network has a positive effect on VMT on Interstates, with corresponding cross-elasticity of 0.116. Lastly, VMT on urban roads varies by state, as indicated through the different coefficients for different states. 4.3. Estimation of total VMT To estimate future energy consumption and GHG emissions for passenger transportation, a random coefficient panel data model, which estimates the effect of various factors on total VMT, was developed. Table 4 shows the estimation results for total passenger VMT in the US. Note that the dependent variable is the natural logarithm of VMT (log-total VMT). As was the case with the SURE models, the number of vehicle registrations and the amount of lane miles were exogenously predicted.5 The variables included in Table 4 are significant at the 95% confidence interval or higher. The variables for urban population, fuel tax, vehicle registrations, percentage of alternative fuel vehicles, and length of network (rural and urban lane miles) have a constant parameter. According to Table 4, increases in fuel tax would decrease 4 The test of the relevance of the instrumental variables (Stock and Watson, 2003, p. 350) suggested that our instruments were valid in this particular application. 5 The test of the relevance of the instrumental variables (Stock and Watson, 2003, p. 350) suggested that our instruments were valid in this particular application. Furthermore, the Hausman specification test (Baltagi, 2008, p. 72) for exogeneity of the regressors rejected the null hypothesis (there are no endogenous variables or that endogeneity does not affect the estimation) and concluded that it is necessary to use an instrumental variables method.

495

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500 Table 3 SURE model estimation results for log-VMT on urban roads. Independent variables

Interstate

Freeway

Principal arterial

Minor arterial

Collector

Constant Percentage Percentage Percentage Percentage Percentage Percentage

11.640 0.007**

3.039 0.011** 0.041** 0.051** 0.067**

10.564 0.007**

8.644 0.009** 0.011**

6.071 0.009**

0.011** 0.061** 0.109**

0.009**

0.310*

0.205**

0.274** 0.016**

of of of of of of

urban population White population Black of African American population Hispanic or Latino population Asian population male population

Natural logarithm of fuel cost Natural logarithm of fuel tax-state Density Natural logarithm of Interstate lane miles Natural logarithm of freeway lane miles Natural logarithm of principal arterial lane miles Natural logarithm of minor arterial lane miles Natural logarithm of collector lane miles Natural logarithm of vehicle registrations Percentage of alternative fuel vehicles Percentage of congested miles on collector roads

* ** a

0.008** 0.060** 0.085** 0.088**

0.531** 0.116**,a

0.272** 0.369** 0.449**

0.072**

1.333** 0.391** **

0.387** 0.138**

0.449** 0.003**,a

Arizona Arkansas California Colorado Connecticut Delaware Florida Georgia Illinois Indiana Kansas Kentucky Louisiana Maine Maryland Massachusetts Michigan Minnesota Mississippi Missouri Nebraska Nevada New Hampshire New Jersey New Mexico New York North Carolina Ohio Oklahoma Oregon Pennsylvania Rhode Island South Carolina Texas Utah Vermont Virginia Washington West Virginia Wisconsin Wyoming

0.188

**

0.183

**

Goodness of fit measure (R2) F-statistic (p-value) Durbin–Watson (autocorrelation) Number of observations

0.958 304.24 (0.0000) 1.779 (0.110) 387

Variables significant at the 95% confidence interval. Variables significant at the 99% confidence interval. Indicates cross-elasticity.

0.014**

1.086 0.676** 0.153** 0.080** 0.398** 0.308**

0.301** 0.264**

0.409**

0.204** 0.631**

0.117** 0.224**

0.154**

0.622** 0.066** 0.091** 0.194** 0.186* 0.254** 0.213** 0.248** 0.066* 0.287** 0.292**

0.486** 0.183** 0.062* 0.104** 0.091** 0.450** 0.134** 0.180** 0.166**

0.320** 0.317** 0.365**

0.101** 0.285**

0.624**

0.185**

0.746**

0.223**

0.302** 0.935**

0.101**

1.170**

0.168**

0.303** 0.255** 0.057*

0.399**

0.506**

0.250** 0.148** 0.115** 0.181** 0.331**

0.346**

0.121** 0.269**

0.304**

0.182** 0.650**

0.077** 0.887 130.16 (0.0000) 1.854 (0.073) 387

0.936 111.03 (0.0000) 1.754 (0.123) 387

0.917 155.94 (0.0000) 1.875 (0.063) 387

0.875 154.52 (0.0000) 1.906 (0.047) 387

496

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500

Table 4 Random coefficients-panel data model estimation results for total log-VMT. Independent variables

Coefficient

Non-random parameters Percentage of urban population Natural logarithm of fuel tax-state Natural logarithm of vehicle registrations Percentage of alternative fuel vehicles Natural logarithm of total rural lane miles Natural logarithm of total urban lane miles

0.0003** 0.034** 0.016** 0.011* 0.083** 0.267**

Random parameters

* **

Mean

Standard error

Constant Natural logarithm of population Percentage of White population Percentage of Hispanic or Latino population Natural logarithm of income per capita Natural logarithm of density

4.560** 0.663** 0.0004** 0.006** 0.051** 0.003*

0.006 0.150 0.004 0.007 0.0215 0.006

Goodness of fit measure (R2) Chi-squared statistic (p-value) Number of observations

0.991 2202.59 (0.0000) 528

Variables significant at the 95% confidence interval. Variables significant at the 99% confidence interval.

VMT; however, our elasticity estimate is lower compared to other estimates reported in the literature. An increase in number of vehicle registrations and an increase in the percentage of alternative fuel vehicles would result in an increase in VMT. Also, an increase in urban population would increase total VMT. Lastly, the length of the network also affects travel demand, as expected and noted in previous studies (Fulton et al., 2000; Noland and Cowart, 2000; Noland, 2001). A 1% increase in rural lane miles results in a 0.083% increase in VMT, while the effect of increasing urban lane miles is much higher; a 1% increase in urban lane miles results in a 0.267% increase in VMT. This result is probably due to the larger number of people affected by road network supply in urban areas and the higher number of trips that can be induced in those areas, compared to rural areas. Turning to random parameters, demographic and socioeconomic factors have a normally distributed parameter. Population is the dominant factor affecting VMT. The parameter for population is normally distributed and positive for 99.9% of the population. Race is another demographic characteristic that affects travel demand, and its parameter is normally distributed. The effect of an increase in the White population on total VMT is not uniform; the coefficient for the White population is less than zero for 54% of cases and greater than zero for 46% of cases. Likewise, the coefficient for the Hispanic or Latino population is less than zero for 80% of cases and greater than zero for 20% of cases. These findings show that there is significant heterogeneity in the travel patterns of different population groups across states and over time, which could be picking up cultural differences that might not be captured by other variables in the model (such as income or area of residence). Income per capita is another socioeconomic factor with a normally distributed parameter. The estimated parameter indicates that an increase in income per capita would mostly result in an increase in travel demand (99% of the population). The difference in the magnitude of the influence of income per capita on VMT across different areas can probably be attributed to the differing cost of living across areas, as well as other factors, such as lifestyle, availability of transit, and development of the area. Similar factors could also affect the coefficient estimated for density. The parameter for density is normally distributed, with a mean of 0.003 and a standard deviation of 0.006. Given these estimates, the effect of higher density on VMT remains negative for 69% of cases and positive for 31% of cases.

5. Estimation of future energy consumption and GHG emissions As noted in the introduction, energy consumption and GHG emissions from passenger transportation in the US is substantial. Because climate change and diminishing energy resources have become major global challenges, predicting future energy needs and reducing energy consumption and GHG emissions are crucial. The objective of this analysis is to demonstrate a methodology for estimating the reduction in energy consumption and GHG emissions as a result of two hypothetical policies: an increased state fuel tax and increased density. Both factors were found to have a significant impact on VMT, and their growth can be influenced through policies. The Argonne National Laboratory has developed for the US Department of Energy the VISION model, which provides estimates of the potential energy use, oil use, and carbon emission impacts of advanced light- and heavy-duty vehicle technologies and alternative fuels through the year 2050 (Singh et al., 2003; Ward, 2008). The model consists of two Excel workbooks: a Base Case of US highway fuel use and carbon emissions to 2050 and a copy (of the Base Case) that can be modified to reflect

497

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500 Table 5 The potential impact of increasing fuel tax and density on energy consumption and GHG emissions. Policy 1: Increased state fuel tax Estimated elasticity of demand (VMT) Estimated annual VMT growth factor Estimated energy consumption (% reduction) Estimated GHG emissions (% reduction) Policy 2: Increased density Estimated elasticity of demand (VMT) Estimated annual VMT growth factor Estimated energy consumption (% reduction) Estimated GHG emissions (% reduction)

0.034 0.999 Light duty vehiclesa

Passenger cars

38.9% 39.4%

8.4% 8.4%

0.003 1.0067 Light duty vehiclesa

Passenger cars

30.9% 31.5%

3.7% 3.7%

a The US fleet of light duty vehicles consists of cars and light trucks, including minivans, sport utility vehicles (SUVs) and trucks with gross vehicle weight less than 8500 lb (Energy Information Administration, 2005).

alternative assumptions about advanced vehicle and alternative fuel market penetration. VISION estimates the energy consumption in the horizon year based on the amount of VMT in the base year, predictions regarding the amount of VMT for the years between the base and the horizon years, and assumptions about advanced vehicle and alternative fuel market penetration (from the Energy Information Administration, 2010b). The horizon year selected for this analysis is 2040. To predict VMT in the horizon year, a growth factor is applied to the VMT in the base year (2008). The growth factor between 2008 and 2040 is estimated as the average of the historical growth rates during the following periods: 1998–2003 (1.0153% or 1.53% increase per year) and 2003–2008 (1.004% or 0.4% increase per year). The estimated annual VMT growth factor for the period 2008–2040 would be 1.0097 (or 0.97% increase per year). The total VMT in 2003 and 2008 and the estimated total VMT for 2040 are 2.175, 2.227, and 2.919 trillions, respectively. 5.1. Policy 1: Increased state fuel tax The authors studied the effect that indexing state fuel tax to inflation would have on energy consumption and GHG emissions. The average state fuel tax across the 48 states and over our analysis period was equal to 20.85 cents per gallon. In 2010 dollars, an equivalent average tax would be 27.42 cents per gallon, which corresponds to a 31.5% increase. As shown in Table 4, the elasticity of VMT with respect to fuel tax is equal to 0.034. As such, a 31.5% increase in fuel tax would be expected to result in a 1.1% decrease in VMT in the near term. The modified growth VMT factor would in that case be (1– 0.011)  1.0097 = 0.999. This modified growth factor was used as an input in VISION for the year 2015. It was assumed that the effect of this policy on VMT would show gradually from 2011 to 2015. As such, this growth factor was applied gradually from 2011 to 2015. The estimated annual VMT growth factor for the period up to 2040 (1.0097) was applied for the years 2008–2010, and gradually from 2016 to 2040. Table 5 shows the reduction in energy consumption and GHG emissions as a result of implementing this policy. Increasing the fuel tax by 31.5% would decrease energy consumption and GHG emissions from light-duty vehicles by 38.9% and 39.4%, respectively. The decrease in energy consumption and GHG emissions from passenger cars would be lower and equal to 8.4%. While the feasibility of this policy in the near term is questionable, an increased fuel tax has the potential of achieving significant reductions in energy consumption and GHG emissions in the long term. 5.2. Policy 2: Increased density Change in land use is recognized to have a significant impact on travel demand and intensity. The effect that a policy encouraging the development of denser areas in the US would have on travel demand, energy consumption and GHG emissions is examined under this scenario. Table 4 shows that a 1% increase in density would result in a 0.003% decrease in VMT. A 100% increase in density would be expected to result in a 0.3% decrease in VMT. The modified growth VMT factor would in that case be (1–0.003)  1.0097 = 1.0067. This modified growth factor was used as an input in VISION for the year 2015. It was assumed that the effect of this policy on VMT would show gradually from 2011 to 2015. As such, this growth factor was applied gradually from 2011 to 2015. The estimated annual VMT growth factor for the period up to 2040 (1.0097) was applied for the years 2008–2010, and gradually from 2016 to 2040. Table 5 shows the reduction of energy consumption and GHG emissions as a result of implementing this policy. The reduction of energy consumption and GHG emissions from lightduty vehicles would be 30.9% and 31.5%, respectively. The decrease in energy consumption and GHG emissions from passenger cars would be 3.7%. In view of the recent land use trends (average density grew by 7.5% from 1998 to 2008), this scenario might seem very aggressive and uncertain in the near term. However, for the purpose of illustrating the methodology, it was of interest to choose an aggressive scenario that could show significant potential for reductions in future energy consumption and GHG emissions.

498

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500

6. Conclusions This paper estimated simultaneous equation models for predicting passenger VMT on different road functional classes in urban and rural areas. In addition to demographic and socioeconomic factors, fuel cost, density, and network length, all of which have been examined in previous research, this paper investigated the influence of technological solutions, such as telecommuting and alternative fuel vehicles, on travel demand. This paper also explored how hypothetical changes in fuel tax and density would affect passenger VMT and the corresponding impacts on future energy consumption and GHG emissions. The estimation results confirmed the effect of well-established contributing factors on VMT, such as population, race, gender, urban population, income per capita, number of vehicle registrations, and density. The authors found that the magnitude of the effect of these factors on total passenger VMT is not uniform, and for certain groups (White and Hispanic or Latino populations) the effect can be either positive or negative. The influence of lane miles on VMT has also been indicated in previous work (Fulton et al., 2000; Noland, 2001), but the estimated elasticity values in previous research were higher in most cases than the elasticity estimates in the present study. Note that our elasticity estimates are comparable to short-term elasticities reported in the literature. Moreover, the authors found that an increase in congestion on one functional road class would shift demand to another functional class (that may serve as substitute), and that VMT on most classes of urban roads have higher elasticity values with respect to lane miles than do VMT on rural roads. The results also showed that the effects of fuel cost and fuel tax are not uniform across the various functional road classes and confirmed that VMT values are highly inelastic to fuel cost changes. The effect of telecommuting and alternative fuel vehicles on VMT was also found to be significant. Increasing the percentage of people telecommuting would decrease short-distance trips traveled on collector roads in rural areas. The number of alternative fuel vehicles has a non-uniform effect on VMT on urban Interstates and freeways, with a decrease in VMT on Interstates and an increase on freeways. The aggregate effect of alternative fuel vehicles on total VMT was found to be positive. These findings could be attributed to either the characteristics of alternative fuel vehicles (a vehicle type with unique performance characteristics, range, and fuel costs) or to the specific attributes of alternative fuel vehicle owners. Additional analysis that would separate battery electric vehicles from other alternative fuel vehicles might help explain these differences in travel patterns. This paper also presented a methodology for estimating future energy consumption and GHG emissions from passenger transportation and presented the results associated with two transportation policies: increasing the state fuel tax by 31.5% and doubling the current levels of population density. While the feasibility of these policies in the near term is uncertain, both have the potential of achieving significant reductions in energy consumption and GHG emissions in the long term. However, some caution should be exercised when viewing the results associated with these two policies because they are subject to the limitations of the data, the assumptions the authors adopted about the estimated growth factors for VMT, and the inherent limitations of the VISION model that was used. More specifically, future energy consumption and GHG emission reductions were estimated based on VMT in the base year (provided in VISION), changes in the annual VMT growth factor (user-defined), and assumptions about advanced vehicle and alternative fuel market penetration (from Energy Information Administration, 2010b). These estimates can be updated and the assumptions can be relaxed as more information becomes available in the future. Acknowledgements This material is based upon work supported by the National Science Foundation under Grant No. 0835989. The authors would like to thank Anant Vyas of Argonne National Laboratory for his assistance with the VISION software, and Robert Rozycki of Federal Highway Administration for providing the HPMS data for this study. References Bagley, M.N., Mokhtarian, P.L., 2002. The impact of residential neighborhood type on travel behavior: a structural equation modeling approach. The Annals of Regional Science 36 (2), 279–297. Baltagi, B., 2008. Econometric Analysis of Panel Data, fourth ed. John Wiley & Sons Ltd, Chichester, West Sussex, UK. Barr, L., 2000. Testing for the significance of induced highway travel demand in metropolitan areas. Transportation Research Record 1706, 4–8. Biorn, E., Lindquist, K., Skjerpen, T., 1998. Random Coefficients and Unbalanced Panels: An Application on Data from Norwegian Chemical Plants. Discussion Paper No. 235, Statistics Norway, Research Department. (accessed May 2010). Boarnet, M., Crane, R., 2001. The influence of land use on travel behavior: specification and estimation strategies. Transportation Research Part A (35), 823– 845. Brazil, H., Purvis, C., 2009. BASSTEGG – Sketch Planning Charrette/GIS Models for Predicting Household Vehicle Miles of Travel (VMT) and Greenhouse Gas (CO2) Emissions. In: 2009 Transportation Planning, Land Use, and Air Quality Conference, July 28–29, Denver, Colorado. Brownstone, D., Golob, T., 2009. The impact of residential density on vehicle usage and energy consumption. Journal of Urban Economics 65, 91–98. Burchell, R., Lowenstein, G., Dolphin, W.R., Galley, C.C., Downs, A., Seskin, S., Still, K.G., Moore, T., 2002. Cost of Sprawl—2000. Transit Cooperative Research Program Report 74. Transportation Research Board, National Research Council, Washington, DC. Cervero, R., Duncan, M., 2006. Which reduce vehicle travel more: jobs-housing balance or retail-housing mixing? Journal of the American Planning Association 72 (4), 475–490. Chatman, D., 2008. Deconstructing development density: quality, quantity and price effects on household non-work travel. Transportation Research Part A 42, 1008–1030.

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500

499

Choo, S., Mokhtarian, P.L., 2007. Telecommunications and travel demand supply: aggregate structural equation models for the US. Transportation Research Part A 41, 4–18. Committee for the Study on the Relationships among Development Patterns, Vehicle Miles Traveled, and Energy Consumption Driving and the Built Environment, 2009. The effects of Compact Development on Motorized Travel, Energy Use and CO2 Emissions. Transportation Research Board, Board on Energy and Environmental Systems, Washington DC. Congressional Budget Office, 2008a. Effects of Gasoline Prices on Driving Behavior and Vehicle Markets. Congress of the United States, Washington, DC. Congressional Budget Office, 2008b. Climate-Change Policy and CO2 Emissions from Passenger Vehicles. Economic and Budget Issue Brief. Congress of the United States, Washington, DC. Contrino, H., McGuckin, N., 2009. Travel demand in the context of growing diversity. TR News 264, 4–9. Energy Information Administration, 2005. Fuel economy of the light duty vehicle fleet. . Energy Information Administration, 2010a. Historical Data: Alternative Transportation Fuels (ATF) and Alternative Fueled Vehicles (AFV). Table V-3. US Department of Energy. . Energy Information Administration, 2010b. Annual Energy Outlook 2010. . Ewing, R., Bartholomew, K., Winkelman, S., Walters, J., Chen, D., 2007. Growing Cooler: The Evidence on Urban Development and Climate Change. Urban Land Institute, Washington, DC. Fang, H., 2008. A discrete–continuous model of households’ vehicle choice and usage, with an application to the effects of residential density. Transportation Research Part B 42, 736–758. Federal Highway Administration, 1989. Functional Classification Guidelines, Section II: Concepts, Definitions, and System Characteristics. Federal Highway Administration (FHWA). . Federal Highway Administration. Monthly Motor Fuel Reported by States. Federal Highway Administration (FHWA). . Federal Highway Administration. Highway Statistics. Tables HM-20, VM-2, MV-1, MF-121T, MF-205, MF-5, HM-42. Federal Highway Administration (FHWA). . Fulton, L., Noland, B., Meszler, D., Thomas, J., 2000. A statistical analysis of induced travel effects in the US Mid-Atlantic region. Journal of Transportation and Statistics 3 (1), 1–14. Gómez-Ibánez, J.A., Humphrey, N., 2010. Driving and the built environment: the effects of compact development on motorized travel, energy use and CO2 emissions. TRB special report. TR News 268, 24–28. Greene, D., Chin, S., Gibson, R., 1995. Aggregate Vehicle Travel Forecasting Model. Oak Ridge National Laboratory, Oak Ridge, Tennessee. Heanue, K., 1998. Highway Capacity and Induced Travel: Evidence and Implications. Transportation Research Circular 481. Transportation Research Board, National Research Council, Washington DC, pp. 33–45. Holtzclaw, J., 1994. Using Residential Patterns and Transit to Decrease Auto Dependence and Costs. National Resources Defense Council, Sierra Club, San Francisco. Federal Highway Administration. Highway Performance Monitoring System (HPMS). Federal Highway Administration (FHWA). . Hsiao, C., 2003. Analysis of Panel Data, second ed. Cambridge University Press, Cambridge, UK. Hsiao, C., Pesaran, M.H., 2004. Random Coefficient Panel Data Models. CESifo Working Paper No. 1233. (accessed May 2010). Hu, P., Jones, D., Reuscher, T., Schmoyer, R., Truett, L., 2000. Projecting Fatalities in Crashes Involving Older Drivers, 2000–2025. Oak Ridge National Laboratory, Oak Ridge, Tennessee. Krizek, K.J., 2003. Residential relocation and changes in urban travel: does neighborhood-scale urban form matter? Journal of the American Planning Association 69 (3), 265–279. Kweon, Y., Kockelman, K., 2004. Nonparametric Regression Estimation of Household VMT. Presented at the 2004 Annual Meeting of Transportation Research Board. Liddle, B., 2009. Long-run relationship among transport demand, income, and gasoline price for the US. Transportation Research Part D 14, 73–82. Liddle, B., 2011. Consumption-driven environmental impact and age structure change in OECD countries: a cointegration-STIRPAT analysis. Demographic Research 24 (30), 00. Litman, T., 2005. The Future isn’t What it Used to Be: Changing Trends and Their Implications for Transport Planning. Victoria Transport Policy Institute. Mabe, K., 2007. Impact of Gas Prices on Vehicle Miles Traveled. A Preliminary Economic Inquiry, Presentation in Farmers Insurance Group, September 2007. Maples, J., 2001. The Transportation Sector Model of the National Energy Modeling System. Office of Integrated and Forecasting, Energy Information Administration. US Department of Energy, Washington DC. McGuckin, N., Liss, S., 2005. Aging cars, aging drivers: important findings from the national household travel survey. ITE Journal 75 (9), 30–37. National Surface Transportation Policy and Revenue Study Commission, 2007a. Implications of Rising Household Income on Passenger Travel Demand. Commission Briefing Paper 4A-06. Prepared by: Section 1909 Commission Staff. National Surface Transportation and Revenue Study Commission, 2007b. Transportation for Tomorrow. Report of the National Surface Transportation and Revenue Study Commission. Newman, P., Kenworthy, J., 1999. Costs of automobile dependence: global survey of cities. Transportation Research Record 1670, 17–26. Noland, R., 2001. Relationships between highway capacity and induced vehicle travel. Transportation Research Part A 35, 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. Pindyck, R.S., Rubinfeld, D.L., 1998. Econometric Models and Economic Forecasts, fourth ed. Irwin/McGraw-Hill, New Jersey. Polzin, S., 2006.The Case for Moderate Growth in Vehicle Miles of Travel: A Critical Juncture in US Travel Behavior Trends. Prepared for US Department of Transportation. Polzin, S., Chu, X., Toole-Holt, L., 2004. Forecasts of future vehicle miles of travel in the United States. Transportation Research Record 1895, 147–155. Singh, M., Vyas, A., Steiner, E., 2003. VISION Model: Description of the Model Used to Estimate the Impact of Highway Vehicle Technologies and Fuels on Energy Use and Carbon Emission to 2050. Center for Transportation Research, Energy Systems Division. Argonne National Laboratory, Argonne, Illinois. Small, K., Van Dender, K., 2007. Long-Run Trends in Transport Demand, Fuel Price Elasticities, and Implications of the Oil Outlook for Transport Policy, Discussion Paper No. 2007-16. Souleyrette, R., Hans, Z., Garrison, W., Wazny, L., 1995. Analysis of trends underlying urban/regional impacts of traffic growth. Journal of Urban Planning and Development 158, 171. Southworth, F., 2001. On the potential impacts of land use change policies on automobile vehicle miles of travel. Energy Policy 29, 1271–1283. Stock, J.H., Watson, M.W., 2003. Introduction to Econometrics. Pearson Education, Inc., Boston, MA. Strathman, J., Dueker, J., Sanchez, T., Zhang, J., Riis, A., 2000. Analysis of induced travel in the 1995 NPTS. US Environmental Protection Agency, Office of Transportation and Air Quality, Washington, DC. Su, Q., 2010. Travel demand in the US urban areas: a system dynamic panel data approach. Transportation Research Part A 44, 110–117. Transportation Research Board, 1995. Expanding metropolitan highways: implications for air quality and energy use. Special Report 245. National Research Council. National Academy Press, Washington, DC. Turcksin, L.S., VanMoll, C., Macharis, N., Sergeant, VanMierlo, J., 2008. How green is the car purchase decision? A review. In: Proc 10th Int Conf Application Advances Technology Transportation, Athens, Greece, May 27–31. US Census Bureau Census, 1990. Summary Tape File 1 (STF 1). . US Census Bureau Census, 2000. Summary File 1 (SF 1) and Summary File 3 (SF 3) .

500

A. Rentziou et al. / Transportation Research Part A 46 (2012) 487–500

US Census Bureau. American Community Survey, 2008. . US Department of Energy, Transportation Energy Data Book. (2008). Edition 28. Chapter 2: Energy. . US Environmental Protection Agency, 2008. US Inventory of Greenhouse Gas Emissions and Sinks: 1990 to 2006. . Ward, J., 2008. VISION 2008 User’s Guide. Vehicle Technologies Program. US Department of Energy, Energy Efficiency and Renewable Energy. Washington, S.P., Karlaftis, M.G., Mannering, F.L., 2011. Statistical and Econometric Methods for Transportation Data Analysis, second ed. Chapman and Hall/CRC, Boca Raton, FL. Wooldridge, J., 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press, Cambridge, MA. Zellner, A., 1962. An efficient method of estimating seemingly unrelated regressions and tests for aggregation bias. Journal of the American Statistical Association 57, 348–368.