WHO RIDE THE HIGH SPEED RAIL IN THE UNITED STATES THE ACELA EXPRESS CASE STUDY

Proceedings of he 2010 Join Rail Conference JRC2010 April 27-29, 2010, Urbana, Illinois, USA JRC2010-36236 WHO RIDE THE HIGH SPEED RAIL IN THE UNITED STATES THE ACELA EXPRESS CASE STUDY Zhenhua Chen The School of Public Policy George Mason Universiy Fairfax, VA U.S.A. ABSTRACT In his sudy, we focus on he Acela Express, and ry o find ou how seleced inernal and exernal facors affec he Acela Express s ridership. A wo-sage leas square regression model is inroduced in order o eliminae he endogeneiy problem caused by price and ridership. Also he Cochrane-Orcu Procedure is adoped o solve auocorrelaion. The resul shows ha icke price and rain on-ime performances, which are used o being hough as imporan facors affec ridership become insignifican, while oher facors like employmen of business and professional in he Norheas Corridor areas have higher influence on high speed rain ridership. The broader objecive of his research is o provide policy suggesions for building of an efficien high-speed rail nework ha can boh be profiable and solve pracical problems ha he conemporary ransporaion sysem faces. INTRODUCTION When Presiden Barack Obama was sworn ino Whie House, he se a prioriy on developing a high-speed rail plan for he counry, which is hough o be one of he soluions for addressing he increasing raffic congesion and improving he environmen [1,2]. There are, however, significan differences beween he Unied Saes and oher counries, such as Japan, France and Germany, ha have developed successful high-speed rail projecs. I is sill no sure wheher high-speed rail will ruly allure U.S. ciizens o ge ou of heir vehicles or off air planes and choose high speed rail as an alernaive. The only high-speed rail in he Unied Saes, he Amrak s Acela Express, which serves he Norheas Corridor (NEC) from Boson via New York, Philadelphia, and Balimore, o Washingon, D.C., has aained posiive revenue because of is seadily increasing ridership since i was pu ino service, and i has become he mos shining service of Amrak (See Figure 1). Given hese facs, i is no clear wha facors affec Acela Express s ridership, or which facors conribue o he growh of ridership performance. Before a naionwide high-speed rail projec consrucion, an empirical analysis of he curren high-speed rail ridership becomes necessary in order o undersand he relaive facors ha affec high speed rail projecs success. In his sudy, we will focus on he Acela Express, and ry o find ou how several seleced facors affec he Acela Express s ridership using mulivariable regression analysis. The broader objecive of his research is o provide policy suggesions for building of an efficien high-speed rail nework ha can boh be profiable and solve pracical problems ha he conemporary ransporaion sysem faces. Based on he daily experience, here are many facors affecing rail ridership, such as populaion densiy, levels of Figure 1 Amrak Norheas Corridor Source: Amrak s monhly performance repors 1 Elecronic copy available a: hp://ssrn.com/absrac=1729816

privae vehicle ownership, opography, service frequency, fares, sysem reliabiliy, and cleanliness. Sudies also show ha ridership increases wih increased income, wheher for business, personal, or leisure ravel [3,4]. One noable sudy, Review and Analysis of he Ridership Lieraure by Brian D. Taylor and Camille N.Y. Fink (2003), caegorizes all hese facors ino wo groups: (1) exernal facors, which are largely exogenous o he sysem and is managers, such as service area populaion and employmen; and (2) inernal facors, on he oher hand, are hose over which managers exercise some conrol, such as fares and service level. This sudy also caegorizes sudies of ransi ridership facors ino wo general groups: (1) research ha focuses on raveler aiudes and percepions ends o use he descripive approach; and (2) sudies ha examine he environmenal, sysem, and behavioral characerisics associaed wih ransi ridership, and end o be srucured as causal analyses.[5,6] The significance of his sudy is ha i examines boh principal findings and mehodological weaknesses for he wo analyical approaches respecively. This sudy helps clarify he appropriae mehodology for ridership research. DATA SOURCES AND METHODOLOGY According o he lieraure on ridership sudies, rail ridership aribues o many facors. Some of daa such as rail ridership, icke revenue, fare, are easy o obain, while daa of some oher facors are no easy o obain, such as opography, sysem reliabiliy, cleanliness, ec. Due o he consrains of daa accessibiliy, in his sudy, we inend o make a comprehensive facor analysis which is based on hree inernal facors----fare of boh he Acela Express (high speed rain) and Acela Regional Train (normal rain) and on-ime performance of he Acela Express, and nine exernal facors, including gasoline price and GDP, personal income and facors reflec employmen of differen careers in NEC. Meanwhile, in order o es how seasonal facor affecs he ridership of he Acela Express, welve monhly dummy variables are also included in his analysis. Therefore, he null hypohesis should be: There is no relaionship beween seleced facors and he Acela Express s ridership. Dependen Variable In his sudy, he dependen variable is Acela Express s ridership. In fac, he Acela Express was in service since December, 11, 2000. Unforunaely, we can only obain is monhly ridership performance from is monhly repor ranging from January, 2004 o July, 2009 released on Amrak s websie. In heir repors, hey have separaed saisics of he Acela Express monhly performance. Toally, we obained 79 monhs daa from January, 2003 o July, 2009. Independen Variable Acela Express Average Fare: This variable represens he icke price of he Acela Express. Since he acual icke price of he Acela Express varies in erms of differen seas and deparure ime, here we only obained an average price from he monhly performance repor, using monhly icke revenue divided by monhly ridership. The Acela Regional Train s Average Fare: Acela Regional Train is a normal rain which runs on he same rack beween Washingon. D.C and Boson via New York. The Regional rain arges general passengers who ravel in NEC while he high speed Acela Express arges a premium passenger marke. Compared wih he high speed rain, he regional rain s icke price is much lower. In our analysis, we are going o check wheher he regional rain s price have any influence on he high speed rain s ridership. The daa also comes from Amrak s monhly performance repors. On Time Performance: This is a percenage number indicaes he monhly on ime performance of he Acela Express high speed rain. The daa also comes from Amrak s monhly performance repors. Gasoline Price: Gasoline price are normally reaed as a key facor affec public s decision of choosing oudoor ransporaion ools. I is assumed ha higher gasoline price conribues o less usage of privae vehicle and more usage of public ranspor. This analysis will also es how gasoline price affecs he usage of high speed rain in he Unied Saes. U.S. Regular All Formulaions Reail Gasoline Price from he Energy Informaion Adminisraion (EPA) is used as he gasoline price variable in his analysis. GDP: The monhly naional real GDP daa is represened by Macroeconomic Advisers index of Monhly GDP (MGDP), which is a monhly indicaor of real aggregae oupu ha is concepually consisen wih real Gross Domesic Produc (GDP) in he NIPA s. The consisency is derived from wo sources. Firs, MGDP is calculaed using much of he same underlying monhly source daa ha is used in he calculaion of GDP. Second, he mehod of aggregaion o arrive a MGDP is similar o ha for official GDP. Growh of MGDP a he monhly frequency is deermined primarily by movemens in he underlying monhly source daa, and growh of MGDP a he quarerly frequency is nearly idenical o growh of real GDP. Disposable Personal Income: Prior research indicaes personal income also affecs he decision making of differen ransporaion modes among general public. I is undersandable ha higher disposable personal income may increase people s preference on choosing premium ransporaion mode. In his analysis, we assume he disposable personal income has influence on he ridership of high speed rain in NEC. of differen careers in NEC: As one of he imporan demographic characerisics ha affec regional ransporaion demand paern, employmen is assumed o be an essenial facor ha influences ridership of high speed rain. In order o es wheher employmen of differen careers in he Norheas Corridor Meropolian Areas have differen influence on he usage of high speed rain, we inroduce seven employmen variables in erms of career ype ino our analysis. The seven employmen careers are: professional and business, governmen, informaion, civilian labor force, manufacuring, and oher service employmen. The daa uni of each career employmen is housands of persons, aggregaed by 2 Elecronic copy available a: hp://ssrn.com/absrac=1729816

employmens of he wo major meropolian saisical areas (MSA) in NEC. 1 Monh Dummy Variable: Twelve monhly dummy variables are creaed o es during differen monh, how ridership of high speed rains change. These variables can also help o undersand he characerisic of ridership flucuaion of he Acela Express high speed rain. DESCRIPTIVE ANALYSIS Table 1 shows he descripive saisics of an array of preliminary variables. In his analysis, he primary goal is o find which of hese variables have influence on he ridership of he Acela Express. In order o ge a beer resul, hree major relevan problems mus be solved. The firs problem is mulicollinear problems among hese preliminary variables. In his siuaion of highly correlaed variables exising ogeher, he coefficien esimaes may change erraically in response o small changes in he model or he daa. Table 2 shows he correlaion among preliminary variables. As i clearly shows, he correlaed coefficien among some variable pairs are higher han 0.8. For example, he Acela Express fare (logp) and he Regional Acela fare (logf) has a correlaion coefficien a 0.8860, which indicaes he log ransformaion of he high speed rain s price variable and normal rain s price variable are highly correlaed, When doing regression, i is beer o include only one of he wo correlaed variable so as o avoid mulicollineariy. The second problem is endogeneiy problem of analyzing he relaionship of high speed rain ridership and is fare. The iniial reason ha made us feel here should have an endogeneiy problem is ha, as he scaer plo of boh price and ridership and log ransformaion of price and log ransformaion of ridership indicaes, he price has a posiive relaionship wih he ridership, which is differen from our normal cogniion on ransporaion price (Figure 2). Thus we doub a much complicaed relaionship may exis beween ridership of he Acela Express high speed rain and is price. Also, according o some oher public ransi ridership scholar, he ridership of inerciy passenger rail is affeced by boh rail supply and demand, as well as changes of price. 2 While some of he facor variables are apparen exogenous, ohers are no. Table1 Descripive Saisics of Preliminary Variables Independen Sd. Variable Obs Mean Min Max Dev. (variable name) Average Fare (p) 79 123.6 10.05 106 143 On ime Performance (op) Acela Regional (nr) Gasoline Price (gp) 79 80.99 8.12 61.8 93.2 79 565622.2 58650.1 418351 673628 79 236.3 65.44 145.8 406.2 Real GDP (rgdp) 79 12730.9 524.8 11608.6 13507.6 Unemploymen Rae (unem) Disposable Personal Income (dpi) Pro & Business (bs) Gov (gov) Informaion (info) Leisure & Hosp (lh) Civilian Labor Force (cl) Manufacuring (ma) Oher Services (oher) 79 3.69.86 2.7 6.5 79 9720.5 900.1 8134 11236.3 79 1414.6 46.9 1329 1496 79 1519.7 23.3 1457.1 1579 79 309.8 6.83 291 324 79 684.6 41.8 600 784 79 12195.3 272.3 11745 12839 79 530.8 38.6 455.1 592.1 79 422.9 13.4 394.6 447.3 In fac, like he air ranspor, he Acela Express high speed rain is a premium ransporaion mode which specifically focuses on premium passengers. Is price is no fixed, bu varies a differen ravel ime. For example, according o Amrak s online icke reservaion websie, icke price of he Acela Express is offered a a lower price in non 1 The wo MSA are Washingon-Arlingon-Alexandria, DC-VA-MD-WV (MSA) and New York-Norhern New Jersey-Long Island, NY-NJ-PA (MSA). Daa is from U.S. Deparmen of Labor: Bureau of Labor Saisics, released on hp://research.slouisfed.org/fred2/caegories/27281 2 We grealy appreciae recommendaions of Professor Brian D Taylor in UCLA and Professor Hiroyuki Iseki in NOU. 3

Table 2 Correlaion among preliminary variables logp logf op loggp logbs loggov loginfo logcl loglh logma logoh logp 1.0000 logf 0.8860 1.0000 op 0.5946 0.7107 1.0000 loggp 0.5994 0.7071 0.5684 1.0000 logbs 0.7254 0.8666 0.6041 0.8507 1.0000 loggov 0.6463 0.5917 0.4465 0.3557 0.4444 1.0000 loginfo -0.6283-0.5280-0.5083-0.4149-0.4048-0.5592 1.0000 logcl 0.7974 0.8738 0.6554 0.7344 0.8221 0.5890-0.7261 1.0000 loglh 0.5487 0.6793 0.4891 0.7283 0.8497 0.3766-0.5387 0.8421 1.0000 logma -0.8468-0.8296-0.6729-0.5970-0.6918-0.6400 0.8879-0.9233-0.7115 1.0000 logoher 0.7366 0.8021 0.6430 0.7589 0.8235 0.6552-0.7575 0.9052 0.8598-0.8917 1.0000 rush hour period, bu is normally sold a a higher price during rush hour period. When he ridership goes up, he price may go up and vice versa. As a resul, ridership and price can affec each oher a he same ime and hus may have a simulaneous relaionship which ineviably causes an endogeneiy problem. adjacen observaions are oo similar han hose ha would be expeced under independence. As a resul, auocorrelaion may happen which can make independen variables more significan han hey may really be hrough smaller s.e for he bea coefficien. In summary, in our case of he Acela Express ridership facor research, mulicollineariy, endogeneiy, as well as auocorrelaion may happen simulaneously ha requires a beer soluion o explain. Figure 3 shows he analysis srucure, including regression problems and he relaed soluions in his analysis. Mulicollineariy Corr and VIF Endogeneiy Two Sage LS Regression Auocorrelaion Cochrane-Orcu Procedure Figure 3 Analysis Srucure Figure 2 Scaer Plo of price and ridership The hird problem originaed from he daase iself. Since in our daase, he 79 observaions represens 79 successive monhly daa concerning each variable. These REGRESSION ANALYSIS In he following discussion, we will show hree seps of how our find model has been achieved, wih solving mulicollineariy, endogeneiy and auocorrelaion problems sep by sep. Sep One: Mulicollineariy Wihou considering endogeneiy and auocorrelaion, we begin our analysis by running wo manual sepwise regressions using all of he preliminary variables o deermine which ones have a saisically significan impac on he wo dependen variables ridership and price. As able 3 and 4 exhibi, he regressions respecively shows seven and six variables are saisically significan under 0.05 significan level. 4

Table 3 Resul of Sepwise Regression on Logp (logprice) logma -1.589892** -12.66-1.840229-1.339554 logrider.1613656** 7.52.1185857.2041455 loginfo 2.20797** 7.22 1.597971 2.817969 loglh -.4593681** -5.65 -.6214109 -.2973252 op -.0014974** -2.91 -.0025215 -.0004733 loggp.0370914* 2.16.0028707.0713121 _cons 3.04647* 2.36.4732928 5.619646 Noe: Adj R-squared = 0.9031, F = 0, p <.1, * p <.05, ** p <.01. Table 4 Resul of Sepwise Regression on Logrider logp 1.332894** 5.81.8752803 1.790508 logbs 4.484596** 7.04 3.214228 5.754965 logcl -6.6499265** -5.4-9.082454-4.217398 loginfo -2.482033** -3.29-3.98524 -.978825 mar.1123463** 3.24.0432578.1814347 dec -.0832957* -2.21 -.1584166 -.0081747 aug -.0626187-1.55 -.143036.0177986 oc.0814448* 2.19.0073673.1555222 jun.065393 1.83 -.005777.136563 _cons 50.19376** 4.25 26.60767 73.77985 Noe: Adj R-squared = 0.8496, F = 0, p <.1, * p <.05, ** p <.01. Table 5 Correlaed Coefficiens among Saisical Significan Variables logrider logp logbs logcl loginfo logma loglh loggp op logrider 1.0000 logp 0.7629 1.0000 logbs 0.6909 0.7254 1.0000 logcl 0.5617 0.7974 0.8221 1.0000 loginfo -0.4519-0.6283-0.4048-0.7261 1.0000 logma -0.5938-0.8468-0.6918-0.9233 0.8879 1.0000 loglh 0.4942 0.5487 0.8497 0.8421-0.5387-0.7115 1.0000 loggp 0.5664 0.5994 0.8507 0.7344-0.4149-0.5970 0.7283 1.0000 op 0.5014 0.5946 0.6041 0.6554-0.5083-0.6729 0.4891 0.5684 1.0000 Then we sar o check VIF of he wo regression equaions and variable pairs wih a correlaed coefficien higher han 0.8, using command corr in STATA. Alhough he resuls of VIF of he wo models are 2.87 and 4.32, which are no bad, he correlaion check, as shown in Table 5, found some variables of employmen among differen careers are highly Afer conrolled he correlaion among variables, we go he following resuls wih solving mulicollineariy problem (See able 6 and 7). The Adj R-squared of regression model on price is 0.8496, which has a 0.0535 decrease from he prior model. All independen variables are saisically significan a 0.05 significan level excep he dummy variable of November. On he oher hand, he Adj R-squared of regression model on correlaed. For example, variable logcl and logp and wih logbs, logma wih logp and logma wih logcl, loggp wih logbs. Therefore, we hen ry manual sepwise regressions on ridership as dependen variable and price as dependen variable, inroducing saisical significan variables ha have correlaed coefficien lower han 0.8 wih oher variables sep by sep. ridership is 0.6890 which represens a 69.90% explanaory abiliy of he real fac, has a 0.0854 decrease compared wih he prior model on ridership. The independen variables are almos saisical significan a 0.05 significan level excep he variables of gasoline price (loggp), on ime performance (op), as well as he dummy variable of Sepember. 5

Table 6 Resul of Regression on Logp (logprice) afer eliminaing Mulicollineariy logma -.7894616** -10.53 -.9388884 -.6400348 logrider.1642355** 6.01.1097168.2187542 loglh -.3646891** -3.69 -.5616487 -.1677296 dec.0357103** 2.63.0086588.0627619 loggp.0557288** 2.65.0138546.097603 nov.0246787 1.78 -.0028946.052252 _cons 9.810254** 9.31 7.709495 11.91101 Noe: Adj R-squared = 0.8496, F = 0, p <.1, * p <.05, ** p <.01. Table 7 Resul of Sepwise Regression on Logrider afer eliminaing Mulicollineariy Logp.6943342** 2.89.215847 1.172821 loggp -.1311333-1.49 -.3071941.0449274 op -.00093-0.50 -.0046633.0028033 logbs 3.921168** 4.46 2.169286 5.67305 jul -.1463235** -3.35 -.2335567 -.0590902 aug -.1707968** -3.60 -.2654453 -.0761483 sep -.0293055-0.67 -.1163288.0577178 dec -.1508018** -3.15 -.2461594 -.0554443 _cons -18.60999** -3.42-29.45425-7.76573 Noe: Adj R-squared = 0.6890, F = 0, p <.1, * p <.05, ** p <.01. Sep Two: Endogeneiy The second sep is o ackle endogeneiy. As we menioned earlier, ridership is he resul of a muual effec from rail supply, demand, as well as price. Price can affec he ridership, bu a he same ime, ridership also affec price. As a resul, a normal muliple variable regression becomes insufficien o explain his simulaneous relaionship, since he dependen variable ridership has a endogenous problem wih explanaory variable price. Thus, we inroduce a wo sage leas square regression model. We follow seps lised below and finally ge he resul, as shown in Table 8. OLS regression on price only, wih all exogenous variables which are saisically significan from previous es. Predic he esimaes of price. OLS regression on ridership, wih prediced price and oher exogenous variables. The Adj-R of he firs sage model on price is 0.8029, which is 0.0467 lower han he prior model. All he independen variable is sill saisically significan a 0.05 level excep he variable of employmen of leisure and hospialiy (loglh). The Adj-R of he second sage model on ridership is 0.7404, which is 0.0514 higher han he previous model. Meanwhile, he variable of price, which is a prediced value wih solving endogeneiy, becomes saisically insignifican. The insignifican variables in his model also include on ime performance of he rain (op), ogeher wih gasoline price (loggp). Table 8 Resul of Two-Sage LS Regression on Logrider afer eliminaing Endogeneiy logrider logp.3808711 1.31 -.1932616.9550037 op -.001459-0.79 -.0050921.002174 logbs 4.94085** 5.24 3.076054 6.805645 loggp -.16749-1.95 -.3375815.0026015 mar.1119462** 2.80.0328014.191091 jul -.1547896** -3.63 -.2390823 -.070497 aug -.184476** -3.92 -.2774635 -.0914885 dec -.1514875** -3.29 -.2425227 -.0604523 6

_cons -24.26789** -4.28-35.47994-13.05584 logp loggp.0761247** 3.23.029511.1227383 logma -.8583523** -9.57-1.035683 -.6810217 loglh -.15980091-1.15 -.4334135.1138117 jul -.0411186* -2.42 -.0747251 -.0075122 aug -.0426847* -2.45 -.0771193 -.0082501 nov.033527* 2.08.0016691.065385 _cons 10.83152** 8.46 8.301267 13.36176 Noe: Equaion logrider: Adj R-squared = 0.7404, F = 0. Equaion logp: Adj R-squared = 0.8029, F = 0. p <.1, * p <.05, ** p <.01. Endogenous variables: logrider logp Exogenous variables: Sep Three: Auocorrelaion op logbs loggp mar jul aug dec logma loglh nov Since our daase is based on ime series daa, each observaion which reflecs performance of each successive monh becomes correlaed wih oher adjacen observaion. Consequenly, he observaion makes independen variables more significan han hey may really be hrough smaller s.e for he bea coefficien. Afer Durbin-Wason (DW) es of our previous model confirms ha auocorrelaion indeed exiss (able 9 and 10). In his sudy, we follow Cochrane-Orcu Ieraive Procedure o solve auocorrelaion. The procedure is implemened by he Prais-Winsen ransformaion in STATA. The procedure of Prais-Winsen ransformaion is shown in Figure 4. Regress logp loggp logma loglh jul aug nov =>Save Residuals e ) ( Creae he lag residual variable e ) ( 1 Regress e ρ + µ = e 1 Compare e and e sop if he difference is small enough Regress Y a + X + e new = new =>Save Residuals ( e ) Use ρ o conver Y, X Y ρ X ρ new = Y Y 1 new = X X 1 The final resuls of he Cochrane-Orcu Ieraive Procedure are shown is Table 9 and 10. As we compared he DW value for he price regression on he firs sage, i increased from he original 0.307812 o he ransformed 1.44775. And he DW value for he ridership regression on he Figure 4 he procedure of Prais-Winsen ransformaion second sage increased from original 0.811855 o he ransformed 2.000870, which indicaes he auocorrelaion has already been eliminaed. Table 9 Resul of Regression on Logp afer eliminaing Auocorrelaion loggp.0329455 1.22 -.0208598.0867508 logma -.9939553** -4.56-1.429193 -.5587171 loglh -.3021769* -2.23 -.573126 -.0312279 jul -.0303305** -4.51 -.0437706 -.0168904 aug -.0369788** -5.51 -.050391 -.0235666 nov.0147482* 2.62.003485.0260113 _cons 12.86703** 7.55 9.464997 16.26907 7

Noe: Adj R-squared = 0.4561, F = 0, p <.1, * p <.05, ** p <.01. Durbin-Wason saisic (original) 0.307812 Durbin-Wason saisic (ransformed) 1.444775 Table 10 Resul of Regression on Logrider afer eliminaing Auocorrelaion plogp.4002818 0.78 -.627224 1.427788 op.0018952 1.07 -.001637.0054274 logbs 3.646159** 2.80 1.048342 6.243975 loggp -.0837126-0.73 -.3123146.1448895 mar.0910242** 3.68.0415303.1405181 jul -.1183893** -3.78 -.1809125 -.0558662 aug -.1470117** -4.05 -.2195342 -.0744892 dec -.1374485** -3.75 -.2106515 -.0642455 _cons -15.69346-1.91-32.10835.7214255 Noe: Adj R-squared = 0.5086, F = 0, p <.1, * p <.05, ** p <.01. Durbin-Wason saisic (original) 0.811855 Durbin-Wason saisic (ransformed) 2.000870 Resul Inerpreaion The null hypohesis is: here is no relaionship beween he seleced facor variables (shown in able 1) and he Acela Express s ridership. According o he final resul of our wo-sage leas square model, we found a couple of variables have saisically significan relaionship wih he dependen variable. The firs sage regression (Table 9) idenifies wha exogenous variables have influence on he Acela Express s price. The final resul shows employmen of manufacure in NEC, employmen of leisure and hospialiy, and monhly dummy variables of July, Augus and November are saisically significan. Facors including on-ime performance and gasoline price are no significan. Par of he reason is due o he mulicollineariy among differen variables. The Adj-R is 0.4561, which means his model have 45.61% of explanaion for he real fac. The deailed inerpreaions of each variable are described as follow. Logma: 1% increase in he employmen of manufacure is associaed wih around 1% decrease in he price of he Acela Express high speed rain. Loglh: 1% increase in he employmen of manufacure is associaed wih around 3% increase in he price of he Acela Express high speed rain. Jul: he price of he Acela Express is associaed wih 0.30% decrease when i is July Aug: he price of he Acela Express is associaed wih 0.36% increase when i is Augus Nov: he price of he Acela Express is associaed wih 0.01% increase when i is November The second sage regression (Table 10) idenifies wha exogenous variables affec on he ridership of he Acela Express. The final resul shows employmen of business and professional in NEC, monhly dummy variables of July, Augus and December are saisically significan. Facors such as icke price, on-ime performance, gasoline price are no saisically significan. The Adj-R is 0.5086, which means his model have 50.86% of explanaion for he real fac. The deailed inerpreaions of each variable are described as follow. Logbs: 1% increase in he employmen of business and professionals in he Norheas Corridor region is associaed wih 3.65% increase of he ridership of he Acela Express high speed rain Mar: he ridership of he Acela Express is associaed wih 0.09% increase when i is March Jul: he ridership of he Acela Express is associaed wih 0.12% decrease when i is July Aug: he ridership of he Acela Express is associaed wih 0.15% decrease when i is Augus Dec: he ridership of he Acela Express is associaed wih 0.14% decrease when i is December CONCLUSION The resul of his analysis above shows ha he Amrak s Acela Express sraegy is quie peneraed on a premium passenger marke. As some media repored, since he debu of he high speed rain in NEC, air shule services which also dominaed by premium passenger have suffered a huge slump. In our sudy, hrough a wo-sage leas square model, we have hree major findings: Firs, he sudy demonsraes ha a higher business and professional employmen conribues o higher usage of he Acela Express. This has revealed ha in he Unied Saes, he high speed inerciy rain service is currenly serving only as a premium ransporaion compeing wih air ranspor. 8

Second, boh he high speed rain price and ridership are srongly affeced by seasonal facors. According o he resul, generally speaking, he icke price decreases in he hird quarer and so does he similar rend on he ridership performance. In he fourh quarer, he performance of he high speed rain seems a lile uncerain because of he increasing price and decreasing ridership. One possible explanaion for his phenomenon migh be ha since a business dominance markeing sraegy, he ridership of he Acela Express is naurally associaed wih business performance in he Norheas Corridor. Normally, in he beginning and end of one year, business aciviy is much prosperous han in he middle of ha year. Third, he sudy shows ha curren riders of he high speed rain are no sensiive o price and on-ime performance, which used o be considered as key facors affec ridership. Par of he reason for he insignificance of price is ha premium passengers pay more aenion o ravel ameniy and service qualiy han price. And his also can be found from he goodness fi of he final model, because he model only shows roughly 50% of explanaion for he real fac, which implies ha oher facors also play imporan role on ridership of he high speed rain service. As for he insignificance of on ime performance, wo possible reasons may have influences. The firs reason migh be ha compared wih Europeans and Japanese, American has lower requiremens on he on ime performance of ransporaion. I is known ha Japanese sociey has much sric puncualiy requiremen han American sociey, especially for people who is involved in business field. Facs can be seemed a he on ime performance of Japanese bulle rain Shinkansen. The average on ime performance is almos 99% [7], and he figures are similar in Ave in Spain and French SNCF. The oher reason migh be curren on ime performance saisfied riders of he high speed rain compared wih oher modes. The bad on ime performance of American air shule service has been blamed for a long ime. Compared wih air, an average 81% of he on ime performance migh have already me riders expecaion. In he Unied Saes, because of he dominance of air ranspor and he advanced naional highway sysem, inerciy passenger rain has long been negleced. However, wih he emergence of problems such as congesion, environmen proecion, as well as energy saving, high speed inerciy rain ha has already demonsraed a grea success in oher counries sars o gain new aenion in he Unied Saes. In his sudy, we ried o explore how he curren high speed rain performs, and i has demonsraed ha i serves only for a premium passenger marke. However, as Presiden Obama said in his HSR speech on April 16, 2009, HSR will be a smar naional ransporaion sysem equal o he needs of he 21 s cenury [8], i is necessary o ge he general public ino consideraion in he process of he HSR projecs planning, designing, implemenaion and operaion. Increasing he ridership should no be he ulimae goal. Raher, how o achieve he equilibrium of public ransporaion uiliy maximizaion and appropriae allocaion of public ransporaion fund should be furher explored in order o build a beer high speed rain in he Unied Saes in he near fuure. REFERENCES [1] GAO, High-speed Passenger Rail: Fuure Developmen Will Depend on Addressing Financial and Oher Challenges and Esablishing a Clear Federal Role, CAO-09-317 (Washingon, D.C.: March, 2009) [2] Passenger Rail Working Group. 2007. Vision for he Fuure: U.S. inerciy passenger rail nework hrough 2050. [3] Polzin, S. E., 2004, Relaionship Beween Land Use, Urban Form And Vehicle Miles Of Travel: The Sae Of Knowledge And Implicaions For Transporaion Planning, Tampa: Universiy of Souh Florida, Florida Deparmen of Transporaion, Federal Highway Adminisraion. [4] Ewing, R. and Cervero, R., 2001, Travel and he Buil Environmen: A Synhesis. Transporaion Research Record 1780. Washingon, D.C.: Transporaion Research Board, Naional Research Council. [5] Brian D. Taylor, and Camille N.Y. Fink. 2003. The Facors Influencing Transi Ridership: A Review and Analysis of Ridership Lieraure. UCLA Deparmen of Urban Planning Working Paper. hp:// www.ucc.ne/papers/681.pdf [6] Brian D. Taylor, Douglas Miller, Hiroyuki Iseki and Camille N.Y. Fink. 2003. Analyzing he Deerminnans of Transi Ridership Using a Two-Sage Leas Squares Regression on a Naional Sample of Urbanized Areas. 2004 Annual Meeing of he Transporaion Research Board [7] Cenral Japan Railway Company Daa Book 2009. hp://english.jr-cenral.co.jp/company/company/ohers/d aa-book/_pdf/2009.pdf [8] A Vision for High Speed Rail in America: High Speed Rail Sraegic Plan. April. 2009. U.S. DOT repor. hp://www.fra.do.gov/downloads/rrdev/hsrsraegicplan. pdf 9