MODE CHOICE MODELING FOR LONG-DISTANCE TRAVEL

Rolf Moeckel, Rhett Fussell, Rick Donnelly Parsons Brinckerhoff, Inc. MODE CHOICE MODELING FOR LONG-DISTANCE TRAVEL 15 January 2013 92nd Annual Meeting of the Transportation Research Board, Washington D.C.

2 Agenda 1. Introduction 2. Long-distance travel demand 3. R 3 Logit long-distance mode choice model 4. Parameters 5. Lessons learned

3 INTRODUCTION Mode Choice Modeling for Long-Distance Travel

4 Relevance of long-distance mode choice 2% of trips are long-distance, contributing 31% to VMT High-speed rail projects Costly infrastructure investments for long-distance travel Regional accessibility affects economic prosperity (Krugman) Even if focus is on auto travel, mode choice is relevant Auto occupancy Remove transit users from highway

5 Existing models Few mode long-distance choice models published Most refer to a very specific corridor Most apply econometric parameter estimation

6 LONG-DISTANCE TRAVEL DEMAND Mode Choice Modeling for Long-Distance Travel

7 Relevant publicly available data (U.S.) American Travel Survey (ATR) 1995 National Household Travel Survey (NHTS) 2002 National Household Travel Survey (NHTS) 2009 National Household Travel Survey (NHTS) 2015 Bureau of Transportation Statistics: Air travel data

Model Concept 8

Synthesize NHTS Records 9

10 Expansion of NHTS Dataset Element Calculation Result A Number of NHTS air traveler records 3,302 B Number of yearly air travelers (BTS) 85,191,050 C Number of daily air travel tours = B / 365 233,400 D Tours per NHTS record = C / A 71 Assumption: Each NHTS record represents 71 long-distance travel tours.

11 Disaggregation of Trips trips zone, zone i j = trips state a, state b zone State k a weight zone State l zone, zone b i weight j zone k, zone l weight ( λ p + ( λ) e ) μ p + ( 1 μ) ( e ) ( d ) zone, zone i i j j exp i j = 1 β i, j Population & employment at origin Population & employment at destination Distance

Assignment of Auto Trips 12

13 LONG-DISTANCE MODE CHOICE MODEL Mode Choice Modeling for Long-Distance Travel

Nesting Structure 14

15 Utility for auto modes u i, j, m, p = ivtc tt i, j, m + ovtc autoegr + prkc p 0.5 p occ m j + aocc p dist i, j aoc Travel Time Egress Time Parking costs Auto-operating costs

16 Utility for Transit Modes Travel time from origin to station Access time Transit travel time Number of transfer u i, j, m, p trfc p = ivtc tt fare i, istat, auto istat, jstat, m + tfqc + ovtc trnacc p frqu istat, jstat, m dist i, j m + ivtc tt istat, jstat, m + ovtc trnegr + ntrc trnsf m + ivtc tt istat, jstat, m jstat, j, auto + Fare Frequency Egress time Travel time from station to destination

Access to and Egress from Transit 17

18 Calibration Mode Observed share Mode-specific constants Business Personal Commute Business Personal Commute Auto 68% 75% 92% 0 0 0 Drive-alone 31% 10% 66% 0 0 0 Shared-ride 2 22% 32% 17% 0.11 0.55-0.20 Shared-ride 3 8% 16% 6% -0.08 0.46-0.39 Shared-ride 4 7% 17% 3% -0.02 0.58-0.51 Transit 32% 25% 8% 2.24 0.52 0.30 Bus 13% 7% 2% 0 0 0 Rail 2% 1% 5% -2.39-1.47-0.62 Air 17% 16% 1% 0.88 1.21 1.53

19 HEURISTICALLY DERIVED PARAMETERS Mode Choice Modeling for Long-Distance Travel

20 Econometrics versus Heuristics Most models use econometrically estimated parameters Small sample size limits robustness of estimation Often leads to large constants

21 Reviewed Papers Bhat, C. R. (1995). "A Heteroscedastic Extreme Value Model of Intercity Mode Choice." Transportation Research Part B: Methodological 29(6): 471-483. Grayson, A. (1981). "Disaggregate Model of Mode Choice in Intercity Travel." Transportation Research Record: Journal of the Transportation Research Board 835: 36-42. Koppelman, F. S. and C.-H. Wen (2000). "The paired combinatorial logit model: properties, estimation and application." Transportation Research Part B: Methodological 34: 75-89. Mandel, B., M. Gaudry, et al. (1997). "A disaggregate Box-Cox Logit mode choice model of intercity passenger travel in Germany and its implications for high-speed rail demand forecasts." The Annals of Regional Science 31: 99-120. Ridout, R. and E. J. Miller (1982). A disaggregate logit model of intercity passenger mode choice. Roads and Transportation Association of Canada Annual Conference, Halifax, N.S. Wen, C.-H. and F. S. Koppelman (2001). "The generalized nested logit model." Transportation Research Part B: Methodological 35: 627-641. Wilson, F. R., S. Damodaran, et al. (1990). "Disaggregate mode choice models for intercity passenger travel in Canada." Canadian Journal of Civil Engineering 17: 184-191. Yao, E. and T. Morikawa (2005). "A study of an integrated intercity travel demand model." Transportation Research Part A 39: 367 381.

22 Sample Size 1,624 [W] 14,195 [R] 51,101 [Y] 540,000 [B] 2,720 [G] 2,769 [Bt] 4,234 [WK] [B] Baik et al. (2008) [Bt] Bhat (1995) [G] Grayson (1981) [R] R 3 Logit [W] Wilson et al. (1990) [WK] Wen, Koppelman (2001) [Y] Yao, Morikawa (2005)

23 In-Vehicle Time Parameters -1.69 [M] -0.261 [B] -0.14 [M] -0.108 [RM] -0.04 [RM] -0.008 [K] -0.027 [B] -0.008 [G] [B] Baik et al. (2008) [Bt] Bhat (1995) [G] Grayson (1981) [K] Koppelman, Wen (2000) [M] Mandel et al. (1997) [R] R 3 Logit [RM] Ridout, Miller (1984) [WK] Wen, Koppelman (2001) -0.026 [R] -0.028 [G] -0.018 [R] -0.011 [Bt] -0.003 [WK]

24 Cost Parameters -0.62 [M] -0.04 [M] -0.0309 [B] -0.0224 [Y] -0.0173 [WK] -0.01 [R] -0.0328 [G] -0.0242 [K] -0.012 [R] -0.0318 [Bt] -0.0333 [Y] [B] Baik et al. (2008) [Bt] Bhat (1995) [G] Grayson (1981) [K] Koppelman, Wen (2000) [M] Mandel et al. (1997) [R] R 3 Logit [WK] Wen, Koppelman (2001) [Y] Yao, Morikawa (2005) -0.0111 [G] -0.0094 [B] -0.006 [R]

25 Mode-Specific Constants -2.4 [R] 0.1 [WK] 2.2 [R] 6.3 [WK] 15.4 [W] 18.0 [W] -3.4 [G] -0.5 [Bt] 1.9 [Bt] -11.3 [Y] -0.3 [RM] 0 0.9 [RM] 1.9 [Y] 1.9 [K] [Bt] Bhat (1995) [G] Grayson (1981) [K] Koppelman, Wen (2000) [R] R 3 Logit [RM] Ridout, Miller (1984) [W] Wilson et al. (1990) [WK] Wen, Koppelman (2001) [Y] Yao, Morikawa (2005)

26 Sensitivity Analysis Transit Scenarios Gas Price Scenarios Mode Base 11 buses per day 15 buses per day Tripled AOC Doubled AOC Drive-alone 11.2% -0.3% -0.7% -0.6% -1.2% Shared-ride 18.9% -0.3% -0.8% -1.0% -1.9% Bus 43.0% 3.7% 8.4% 1.3% 2.7% Rail 0.0% 0.0% 0.0% 0.0% 0.0% Air 26.9% -3.1% -7.0% 0.2% 0.5%

27 LESSONS LEARNED Mode Choice Modeling for Long-Distance Travel

28 Lessons Learned Teaming with Amtrak and bus providers helps a lot Pay attention to trip duration Limit party size (R 3 Logit: 7) Model depends on representing travel options correctly, less than on fine-tuned calibration