Assessing Highway Tolling and Pricing Options and Impacts: Volume 2: Travel Demand Forecasting Tools (2012)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

163 A p p e n d i x A Appendix A A.1. Details of Selected Models Applied for Pricing Studies A.1.1.Four-Step Trip-Based Models A.1.1.1. Orange County, California ORANGE COUNTY TRANSPORTATION AGENCY Orange County, California Major model feature Detailed feature / sub-model Characteristics Spatial scale Regional Demand model structure Aggregate trip-based. Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Mode Choice & Auto Occupancy DA Toll, SR2 Toll and SR3+ Toll are elemental alternatives in a nested logit model. Utility of a toll mode is a function of its travel time, cost, a constant (unobserved attributes) term, and a ‘bonus’ term that increases with the difference between the toll and no toll travel time. Trip Distribution The HBW distribution model uses mode choice logsums as the gravity model impedance. The mode choice utility constants used for trip distribution are not equal to the constants used for mode choice. Willingness to pay / VOT and user segmentation Vehicle classes Auto only. Vehicle occupancy categories SOV Toll, SOV No Toll, HOV Toll HOV No Toll, but same VOT for all classes. Trip purpose segmentation (low/med/high income) $1989 Home based work ($3.1/$8.4/$19.4) Home based other ($1.5/$4.1/$9.7) Non home based work ($6.7/hr) Non home based other ($6.7/hr) Household / person characteristics Household income (low/med/hi) – VOT for trip distribution vary by income group and trip purpose. Network simulation tool Simulation type Static user equilibrium assignment Representation of priced highway facilities Cost function depends on travel time only. Demand – Network Equilibrium Feedback implementation None. Surveys and other data sources for model estimation / calibration / validation Household travel survey N/A Survey of existing toll road users N/A Stated Preference survey N/A Traffic counts N/A

164 A.1.1.2. Wasatch Front, Utah WASATCH FRONT REGIONAL COUNCIL / MOUNTAINLAND ASSOCIATION OF GOVERNMENTS Salt Lake, Utah Major model feature Detailed feature / sub-model Characteristics Spatial scale Regional Demand model structure Aggregate trip-based Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Mode Choice & Auto Occupancy DA Toll, SR2 Toll and SR3+ Toll are elemental alternatives in a nested logit model. Utility of an auto mode, which includes the toll alternatives, is a function of its travel time, cost, a constant (unobserved attributes) term, and CBD and urbanization indicator variables. Trip Distribution The HBW distribution model uses mode choice logsums in a destination choice framework. For all other purposes, toll costs are expressed in minutes using a VOT factor and added to the travel time. The impedance for the gravity models is the harmonic mean of travel time for the free path and travel time for the toll path. Willingness to pay / VOT and user segmentation Vehicle classes VOT is $40/hour for all classes General purpose lane users HOV lane users, short distance HOV lane users, long distance Toll lane users, short distance Toll lane users, long distance Vehicle occupancy categories SOV, HOV2, HOV3+ Trip purpose segmentation (low income / high income) Home based work ($1.34/$11.5) Home based school ($2.2/$4.2) Home based other ($0.8/$5.6) Non home based ($2.8/$5.7) Household / person characteristics Household income (low/high) Network simulation tool Simulation type Static user equilibrium assignment Representation of priced highway facilities Cost function depends on travel time, distance and toll costs. Demand – Network Equilibrium Feedback implementation N/A Surveys and other data sources for model estimation / calibration / validation Household travel survey 1992 Traffic counts N/A

165 A.1.1.3. Dallas – Fort Worth, Texas NORTH-CENTRAL TEXAS COUNCIL OF GOVERNMENTS Dallas – Fort Worth, Texas Major model feature Detailed feature / sub-model Characteristics Spatial scale Regional Demand model structure Aggregate trip-based. Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Mode Choice The only pricing impact is the inclusion of the toll cost in the utility of the auto alternatives. Willingness to pay / VOT and user segmentation Vehicle classes Drive alone, shared ride HOV lane, shared ride non-HOV lane, trucks. Two values of time: $10/hr for autos and $12/hr for trucks ($1999) Trip purpose segmentation Home based work ($5.91/hr)) Home based non work ($4.07/hr) Non home based ($3.30/hr) Network simulation tool Simulation type Static user equilibrium assignment Representation of priced highway facilities Cost function depends on travel time, operating costs and toll costs. Demand – Network Equilibrium Feedback implementation Fixed number of model iterations. Surveys and other data sources for model estimation / calibration / validation Household travel survey 1996 / 4,500 households Traffic counts 1999

166 A.1.1.4. San Francisco Bay Area, California METROPOLITAN TRANSPORTATION COMMISSION (*) San Francisco Bay Area, California Major model feature Detailed feature / sub-model Characteristics Spatial scale Regional Demand model structure Aggregate trip-based. Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Mode Choice & Auto Occupancy DA Toll, SR2 Toll and SR3+ Toll are elemental alternatives in a nested logit model. Trips that use the existing tolled bridges (Golden Gate, Bay, Dumbarton, San Mateo or San Rafael Bridges) are not considered Toll trips. Utility of a toll mode is a function of its travel time, cost or log of cost, household income, zonal characteristics, and a constant (unobserved attributes) term. Toll costs are shared among vehicle occupants in the off-peak period. Willingness to pay / VOT and user segmentation Vehicle classes DA Toll, DA No Toll, SR2 Toll, SR2 No Toll, SR3+ Toll, SR3+ No Toll, Trucks. Trip purpose segmentation VOT in $1990. Home based work ($9.65) Home based school ($0.36) Home based university ($0.67) Home based recreation ($0.79) Home based shop ($6.58) Home based other Non home based ($1.08) Trucks ($25.0) Internal/External ($1.08) Network simulation tool Simulation type Static user equilibrium assignment. Akcelik volume-delay functions. Representation of priced highway facilities Cost function depends on travel time only. Demand – Network Equilibrium Feedback implementation Speed feedback to mode choice. Surveys and other data sources for model estimation / calibration / validation Household travel survey 1990 (10,000 households, trip-based survey) 1996 (15,000 households, activity- based survey) (*) As modified for the I-680 Corridor Value Pricing Study and the FAIR Lanes Study.

167 A.1.1.5. San Diego, California SAN DIEGO ASSOCIATION OF GOVERNMENTS (*) San Diego, California Major model feature Detailed feature / sub-model Characteristics Spatial scale Regional Demand model structure Aggregate trip-based. Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Mode Choice & Auto Occupancy DA Toll, SR2 Toll and SR3+ Toll are elemental alternatives in a nested logit model. Also considers HOV No Toll as elemental alternatives. Utility of a toll mode is a function of its travel time, cost, and a constant (unobserved attributes) term. Trip Distribution The gravity models use generalized cost as the impedance measure, with time valued at $0.35/min ($21/hr) and distance at $0.13/mile for all purposes. Willingness to pay / VOT and user segmentation Vehicle classes SOV, HOV. Vehicle occupancy categories SOV Toll, SOV No Toll, HOV Toll HOV No Toll, but same VOT for all classes. Trip purpose segmentation VOT in $1995 (low/med/high income) Home based work ($1.8/$5.4/$11.2) Home based other ($0.9/$2.7/$5.6) Non home based ($2.7/hr) Household / person characteristics Household income (low/med/hi) Network simulation tool Simulation type Static user equilibrium assignment Representation of priced highway facilities Cost function depends on travel time, travel costs and distance, with time valued at $21/hour for all vehicle classes. Demand – Network Equilibrium Feedback implementation One feedback iteration. Surveys and other data sources for model estimation / calibration / validation Household travel survey Year 1995 / 2,050 households Survey of existing toll road users Years 1997-1999 / 1,500 commuters Traffic counts Year 2000 / express lane counts (*) As modified for the I-5 North Coast Managed Lane Value Pricing Study.

168 A.1.1.6. Minneapolis – Saint Paul, Minnesota TWIN CITIES METROPOLITAN COUNCIL Minneapolis – Saint Paul, Minnesota Major model feature Detailed feature / sub-model Characteristics Spatial scale Regional Demand model structure Aggregate trip-based. Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Mode Choice & Auto Occupancy DA Toll, SR2 Toll no HOV, SR2 Toll HOV, SR3+ Toll no HOV and SR3+ Toll HOV are elemental alternatives in a nested logit model. Utility of a toll mode is a function of its travel time, cost, and a constant (unobserved attributes) term. Trip Distribution Mode choice logsums are used as the accessibility term in destination choice models for all purposes. Willingness to pay / VOT and user segmentation Vehicle classes Auto only. Vehicle occupancy categories SOV Toll, SOV No Toll, HOV Toll HOV No Toll, but same VOT for all classes. Trip purpose segmentation VOT in $2000 Home based work ($12.27/hr) Home based other ($3.67/hr) Non home based work ($1.92/hr) Non home based other ($2.00/hr) Network simulation tool Simulation type Static user equilibrium assignment Representation of priced highway facilities Cost function depends on travel time only. Demand – Network Equilibrium Feedback implementation Travel times fed back to trip distribution; model converges when VMT difference is less than 5%. Surveys and other data sources for model estimation / calibration / validation Household travel survey Year 2001 / 6,200 households

169 A.1.1.7. Denver, Colorado DENVER REGIONAL COUNCIL OF GOVERNMENTS Denver, Colorado Major model feature Detailed feature / sub-model Characteristics Spatial scale Regional Demand model structure Aggregate trip-based. Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Mode Choice & Auto Occupancy The only pricing impact is the inclusion of the toll cost in the utility of the auto alternatives. VOT in the mode choice model are adjusted for vehicle occupancy (SOV values multiplied by average vehicle occupancy). VOT for parking costs are twice as high than for toll or operating costs. Willingness to pay / VOT and user segmentation Vehicle classes DA, SR2 and SR3+ (trucks are pre- loaded). Time of day VOT in network simulation varies by time of day: $8/hr peak and $6/hr off- peak. Trip purpose segmentation $1996 (low/med/high income) for SOV trips (toll costs) Home based work ($4 / $8 / $16) Home based other ($8.8/hr) Non home based ($8.4/hr) Household / person characteristics Household income (low/med/hi) for HBW trips only Network simulation tool Simulation type Static user equilibrium assignment Representation of priced highway facilities Cost function depends on travel time, distance and toll costs. VOT varies by time period. Demand – Network Equilibrium Feedback implementation Convergence is reached when less than 1% of the links show a speed difference of more than 10%. Speeds are fed back to top of model chain. Surveys and other data sources for model estimation / calibration / validation Household travel survey 1997 / 4,100 households Not used for model estimation.

170 Specifically developed for the E-470 toll traffic study. Denver, Colorado Major model feature Detailed feature / sub-model Characteristics Spatial scale Corridor Demand model structure Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Binary pre-route choice Logistic diversion model, with toll probabilities expressed as a function of the natural log of travel time savings and square of toll. Vehicle occupancy categories Toll, No toll Trip purpose segmentation Home based work Airport trips Non home based work Non home based other Network simulation tool Simulation type Static user equilibrium assignment Representation of priced highway facilities Cost function depends on travel time only. Demand – Network Equilibrium Feedback implementation None. Surveys and other data sources for model estimation / calibration / validation Household travel survey N/A Stated Preference survey Year 1991

171 A.1.1.8. Atlanta, Georgia ATLANTA REGIONAL COMMISSION Atlanta, Georgia Major model feature Detailed feature / sub-model Characteristics Spatial scale Regional Demand model structure Aggregate trip-based. Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Binary pre-route choice Logistic diversion model, with probabilities a function of travel time savings and toll cost: Travel time coefficient (min): 0.0875 Cost coefficient ($): 1.121 Willingness to pay / VOT and user segmentation Vehicle classes $2000 SOV, HOV, Commercial. VOT for passenger car is $15/hour, for Commercial vehicles $35/hour, for purposes of expressing toll cost as time equivalents when building paths. Diversion model parameters are the same for all vehicle classes. Trip purpose segmentation None for the diversion model Network simulation tool Simulation type Static user equilibrium assignment Representation of priced highway facilities Cost function depends on travel time and tolls. Demand – Network Equilibrium Feedback implementation Not implemented for pricing studies Surveys and other data sources for model estimation / calibration / validation Household travel survey 2001 / 8,000 households Focus group study 2004 / 113 participants

172 A.1.1.9. Orlando, Florida FLORIDA TURNPIKE ENTERPRISE Orlando, Florida Major model feature Detailed feature / sub-model Characteristics Spatial scale Regional Demand model structure Aggregate trip-based. Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Mode Choice & Auto Occupancy DA Toll, SR2 Toll and SR3+ Toll are elemental alternatives in a nested logit model. Utility of a toll mode is a function of its travel time, cost relative to natural log of household income, trip length (HBW only), and a constant (unobserved attributes) term. For HBW trips, vehicle occupancy affects cost-sharing; costs are divided by ln(occupancy + 1). Vehicle occupancy has no effect on cost- sharing for other purposes. Willingness to pay / VOT and user segmentation Vehicle classes SOV Toll, SOV No Toll, HOV2 Toll, HOV2 No toll, HOV3+ Toll, HOV3+ No Toll. All use same VOT in network simulation. Vehicle occupancy categories SOV, HOV2, HOV3+ Trip purpose segmentation (Range of VOT by income levels) $2000 Home based work peak ($4.5/hr to $9.5/hr) Home based work off peak ($4.0 /hr to $13.5/hr) Home based other peak ($4.0/hr to $7.50/hr) Home based other off peak ($3.0/hr to $8.0/hr) Non home based Visitors ($5.0/hr) Household / person characteristics Household income (continuous) Time of day VOT vary by time period (peak, off- peak). Network simulation tool Simulation type Static user equilibrium assignment. Akcelik volume-delay functions. Representation of priced highway facilities Cost function depends on travel time and toll cost, with parameters that vary by time period: Time: -0.047 cents/min (peak) / -0.06 cents/min (off peak) Cost: -0.006 cents/cents (peak & midday) / -0.003 cents/cents (night). Equivalent VOT: Peak - $4.7/hr Midday - $6.0/hr Night - $12/hr A queuing model was used to estimate delay at toll plazas.

173 FLORIDA TURNPIKE ENTERPRISE Orlando, Florida Major model feature Detailed feature / sub-model Characteristics Demand – Network Equilibrium Feedback implementation Travel times fed back to trip distribution, calculated using method of successive averages. The model executes four feedback iterations. Surveys and other data sources for model estimation / calibration / validation Origin / Destination survey 2000 Stated Preference survey 2000 / 1,044 respondents Speed measurements 2000 Traffic counts 2000

174 A.1.1.10. Seattle, Washington PUGET SOUND REGIONAL COUNCIL Seattle, Washington Major model feature Detailed feature / sub-model Characteristics Spatial scale Regional Demand model structure Aggregate trip-based. Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Mode Choice & Auto Occupancy The only pricing effect is the inclusion of toll costs in the utility function of the auto alternatives. The mode choice model is multinomial logit. Trip Distribution The HBW distribution model uses a composite impedance variable, partially based on mode choice logsums, in a destination choice framework. Willingness to pay / VOT and user segmentation Vehicle classes and time of day (reported VOTs for final assignment only). $2000 Peak: SOV HBW ($10.6/$19.6/$28.6/$38.4), HBO & Carpools ($16.7/hr) Light Trucks ($35.0/hr) Medium Trucks ($35.5/hr) Heavy Trucks ($41.0/hr) Off Peak: SOV HBW ($8.9/$16.4/$23.9/$31.0), HBO & Carpools ($14.0/hr) Light Trucks ($29.3/hr) Medium Trucks ($29.7/hr) Heavy Trucks ($34.3/hr) Vehicle occupancy categories SOV, HOV2, HOV3+ Trip purpose segmentation $2000 Home based work ($4.0/$7.2/$10.8/$13.8) Home based college ($8.4/hr) Home based other ($3.8/hr) Non home based ($5.1/hr) Household / person characteristics Household income (four groups) for HBW trips. Network simulation tool Simulation type Static user equilibrium assignment. HCM 2000 volume-delay functions. Representation of priced highway facilities Cost function depends on travel time and costs, with parameters that vary by vehicle class. Demand – Network Equilibrium Feedback implementation Travel times fed back to trip distribution. Four feedback iterations performed. Surveys and other data sources for model estimation / calibration / validation Household travel survey 1999 Not used for model estimation. Traffic counts Year / vehicle type

175 Model Developed for the Dulles Greenway traffic study. Washington, D.C. Major model feature Detailed feature / sub-model Characteristics Spatial scale Corridor Demand model structure Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Binary pre-route choice Logistic diversion model, with toll probabilities expressed as a function of the natural log of travel time savings, square of toll and a constant term. Vehicle occupancy categories Toll, No toll Trip purpose segmentation Home based work Airport trips Non work Network simulation tool Simulation type Static user equilibrium assignment Representation of priced highway facilities Cost function depends on travel time only. Demand – Network Equilibrium Feedback implementation None. Surveys and other data sources for model estimation / calibration / validation Household travel survey N/A

176 A.1.1.11. Austin, Texas Austin, Texas Major model feature Detailed feature / sub-model Characteristics Spatial scale Corridor Demand model structure Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Binary pre-route choice Logistic diversion model, with toll probabilities expressed as a function of travel time savings, toll cost (relative to natural log of income for HBW), and constants stratified by electronic vs. cash payment. Vehicle occupancy categories Toll, No toll Trip purpose segmentation Home based work Home based school Home based shop Home based other Non home based work Non home based other Trucks Network simulation tool Simulation type Static user equilibrium assignment Representation of priced highway facilities Cost function depends on travel time only. Demand – Network Equilibrium Feedback implementation None. Surveys and other data sources for model estimation / calibration / validation Household travel survey Size / sample, year, structure / questionnaire Stated preference survey N/A

177 A.1.1.12. Houston, Texas HOUSTON – GALVESTON AREA COUNCIL Houston, Texas Major model feature Detailed feature / sub-model Characteristics Spatial scale Regional Demand model structure Aggregate trip-based. Modeled pricing impacts (traveler responses), sub-model structure, form of utility function, incorporation of pricing Mode Choice & Auto Occupancy DA toll, SR2 toll, SR3 toll and SR4+ toll are elemental alternatives in a nested logit model. Utility of a toll mode is a function of its travel time, cost, time savings with respect to the toll free alternative, and a constant (unobserved attributes) term. The toll options are available only if they imply minimum time savings (3 min. for work trips; 2.5 min. for non-work trips). Willingness to pay / VOT and user segmentation Vehicle classes Auto only. Vehicle occupancy categories SOV, HOV2, HOV3+ Trip purpose segmentation (lowest VOT – highest VOT) $1985 Home based work ($2.7/hr - $5.5/hr) Home based other ($1.6/hr - $3.3/hr) Non home based ($4.2/hr) Household / person characteristics Household income – five classes. Network simulation tool Simulation type Static user equilibrium assignment, using 24 hr speed averages instead of free-flow speeds and a modified BPR volume-delay function. Representation of priced highway facilities Cost function depends on travel time only. Demand – Network Equilibrium Feedback implementation None. Surveys and other data sources for model estimation / calibration / validation Household travel survey 1984 / 1,500 households (estimation) 1994 / 2,400 households (calibration & validation). Traffic counts 1995

178 A.1.2.Activity-Based Tour-Based Models A.1.2.1. San Francisco, California SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY (SFCTA) RPM-9 MODEL San Francisco, California Major model feature Detailed feature / sub- model Characteristics Spatial scale Regional (9 counties) Demand model structure Activity-based tour-based microsimulation Modeled pricing impacts (traveler responses), sub- model structure, form of utility function, incorporation of pricing Household auto ownership and tour frequency models Logit models with accessibility indices in the utility functions. Accessibility indices are based on highway and transit travel times. Effect of pricing can be partially captured through impact on travel times; however there is no direct sensitivity to pricing Primary tour destination choice MNL model with mode choice logsum as one of the variables; sensitivity to pricing is ensured through this logsum Tour mode choice Nested logit model segmented by 7 purposes (Work, K-8, High School, College, Other, Work-Based) with 10 modes: 1=SOVFree, 2=1=SOVPay 3=HOV2 Free, 4=HOV2 Pay, 5=HOV 3+ Free, 6=HOV 3+ Pay, 7=Walk, 8=Bike, 9=Walk to transit, 10=Drive to transit. Mode utilities include time and cost variables; binary sub-choice (toll vs. non-toll) is included below auto modes. Stop frequency Chosen within the context of the daily activity pattern model; currently insensitive to pricing Stop location choice Based on the level-of-service of the chosen tour mode, including toll. Sensitive to toll cost if tour mode is auto. Trip mode choice Nested logit segmented by 7 tour purposes, with 14 modes: 1=SOVFree, 2=1=SOVPay 3=HOV2 Free, 4=HOV2 Pay, 5=HOV 3+ Free, 6=HOV 3+ Pay, 7=Walk, 8=Bike, 9=Walk to local, 10=Walk to MUNI Metro, 11=Walk to Premium, 12=Walk to BART, 13=Drive to Premium, 14=Drive to BART. Mode utilities include time and cost variables, and influenced by tour mode choice; binary sub-choice (toll vs. non-toll) is included below auto modes. Time-of-day choice Multinomial logit models with 5 time periods; currently structured before destination choice for all tour purposes except for work; currently insensitive to pricing. Currently re-structuring model to place time-of-day choice between destination choice and mode choice, to enable use of mode choice logsums in time-of-day choice and allow sensitivity to pricing. Also adding half-hourly time-of-day choice model for auto trips (after trip mode choice) to allow sensitivity to pricing in peak period spreading, based on SP data.

179 SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY (SFCTA) RPM-9 MODEL San Francisco, California Major model feature Detailed feature / sub- model Characteristics Willingness to pay / VOT and user segmentation Vehicle classes Vehicle classes in the time-of-day specific assignments and assumed VOT: 1=SOV Free ($15.0/hr) 2=SOV Pay ($15.0/hr) 3=SOV Already Paid ($15.0/hr) 4=HOV2 Free ($30.0/hr) 5=HOV2 Pay ($30.0/hr) 6=HOV2 Already Paid ($30.0/hr) 7=HOV3+ Free ($45.0/hr) 8=HOV3+ Pay ($45.0/hr) 9=HOV3+ Already Paid ($45.0/hr) 10=Externals Free ($15.0/hr) 11=Externals Pay ($15.0/hr) 12=Commercial Vehicles Free ($30.0/hr) 13=Commercial Vehicles Pay ($30.0/hr) Notes: Already Paid refers to area pricing; if a traveler has already paid the area fee, they are free to travel without paying the toll again, and are placed in the ‘Already paid’ bin. A binary logit model is used to split externals and commercial vehicles into free and pay categories. Vehicle occupancy categories SOV, HOV2, HOV3+ in mode choice; SOV, HOV2, HOV3+ in assignment Trip purpose segmentation, VOT in $1989 (currently deterministic for each purpose) There are two different VOTs available in the SFCTA RPM-9 Models. One is the traditional, fixed value-of-time with segmentation by household income. The other algorithm draws a mandatory and non-mandatory value-of-time for each person day from a log-normal distribution. The draw is based on the ratio of the household income to the number of workers in the household (the per worker household income). The models were calibrated and pricing policies are currently being analyzed using the distributed values of time. Work Segmented VOT: $0-$30k = $3.61/hr $30-$630k = $10.86/hr $60k + = $17.87/hr Distributed VOT: 1/2 of the average hourly wage rate, varying according to a lognormal distribution with mu=0 and sigma = 0.25 Grade School Segmented VOT: $0-$30k = $2.40/hr

180 SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY (SFCTA) RPM-9 MODEL San Francisco, California Major model feature Detailed feature / sub- model Characteristics $30-$630k = $7.23/hr $60k + = $12.00/hr Distributed VOT: Either the parents mandatory VOT or $5/hour, whichever is lower. High School Segmented VOT: $0-$30k = $2.40/hr $30-$630k = $7.23/hr $60k + = $12.00/hr Distributed VOT: Either the parents mandatory VOT or $5/hour, whichever is lower. University Segmented VOT: $0-$30k = $2.40/hr $30-$630k = $7.23/hr $60k + = $12.00/hr Distributed VOT: 1/2 of the average hourly wage rate, varying according to a lognormal distribution with mu=0 and sigma = 0.25 Other Segmented VOT: $0-$30k = $2.40/hr $30-$630k = $7.23/hr $60k + = $12.00/hr Distributed VOT: 2/3 of the mandatory VOT Work-Based Segmented VOT: $0-$30k = $2.40/hr $30-$630k = $7.23/hr $60k + = $12.00/hr Distributed VOT: 2/3 of the mandatory VOT Household / person characteristics Household income affects VOT in both distributed and segmented calculations. Many other person and household characteristics, particularly in day-pattern and time-of-day choice models. Network simulation tool Simulation type Static user equilibrium multi-class assignment (Cube software) Representation of priced highway facilities Tolls are skimmed during assignment, and fed to demand models in cost matrices. Tolls are considered in path-building and assignment algorithms by conversion to travel time units based on the average VOT for each vehicle class; Area

181 SAN FRANCISCO COUNTY TRANSPORTATION AUTHORITY (SFCTA) RPM-9 MODEL San Francisco, California Major model feature Detailed feature / sub- model Characteristics pricing is modeled by rules regarding which tours are exposed to costs first, and which tours may be ‘free’ based on whether toll has been paid in previous choice models. Demand – Network Equilibrium Feedback implementation Feedback is implemented for all models. MSA method is applied for link volumes and trip tables in parallel. Work destination choice shadow pricing uses prices computed in previous iterations, and model is started with skims and shadow prices from a previous model run. Early iterations utilize population sampling to cut down run-time. Full equilibrium requires 3-5 iterations. Model is run 5 times (with full feedback for each run) and results are averaged. Surveys and other data sources for model estimation / calibration / validation Household travel survey Originally estimated using 1990 and 1996 BATS travel surveys, with 10,000 households and 3,700 2- day households each, Model updated/re-calibrated based on 2000 BATS travel survey with 15,000 2- day households. Survey of existing toll road users None available. Aggregate data available on toll bridge crossings. Stated Preference survey SP Survey of commute and non-commute auto trips to downtown San Francisco conducted in summer 2007. Data currently being used to compute updated values-of-time, and mode and time-of-day elasticities with respect to pricing. Traffic counts Extensive set of 1,640 traffic counts from different sources updated on a yearly basis. Transit boardings from various sources, BART station-station daily trip tables, on-board bus speed data, etc.

182 A.1.2.2. New York, New York NEW YORK METROPOLITAN TRANSPORTATION COUNCIL New York, New York Major model feature Detailed feature / sub- model Characteristics Spatial scale Regional (28 counties) Demand model structure Activity-based tour-based microsimulation Modeled pricing impacts (traveler responses), sub- model structure, form of utility function, incorporation of pricing Household auto ownership and tour frequency models Logit models with accessibility indices in the utility functions. Accessibility indices are based on highway and transit travel times. Effect of pricing can be partially captured through impact on travel times; however there is no direct sensitivity to pricing Primary tour destination choice MNL model with mode choice logsum as one of the variables; sensitivity to pricing is ensured through this Logsum Tour mode choice Nested logit model segmented by 6 purposes with 11 modes: 1=SOV, 2=HOV2, 3=HOV3, 4=HOV4+, 5=Walk to transit, 6=Drive to transit, 7=Walk to commuter rail, 8=Drive to commuter rail, 9=Taxi, 10=School bus, 11=Non-motorized. Mode utilities include time and cost variables; No binary sub-choice (toll vs. non-toll) is currently included. Stop frequency & location choice Based on the distance measures and person, household, and zonal attributes; currently insensitive to pricing Trip mode choice Derived from the tour mode based on the relative stop location; rule-based and currently insensitive to pricing Time-of-day choice Predetermined diurnal distributions by tour departure time and duration by half-hour intervals; segmented by purpose, mode, and geography; currently insensitive to pricing; a stand-alone subroutine for peak-spreading was developed that restructures these distributions based on tolls and travel-time savings Willingness to pay / VOT and user segmentation Vehicle classes Vehicle classes in the time-of-day specific assignments and assumed VOT: 1=SOV ($15.0/hr) 2=HOV2 ($30.0/hr) 3=HOV3+ ($45.0/hr) 4=Externals ($15.0/hr) 5=Trucks ($120.0/hr) 6=Commercials ($120.0/hr) 7=Taxis ($30.0/hr) Vehicle occupancy categories SOV, HOV2, HOV3, HOV4+ in mode choice; SOV, HOV2, HOV3+ in assignment Trip purpose segmentation, VOT in $1997 (currently deterministic for each purpose) Work ($15.8/hr) School ($6.50/hr) University ($11.7/hr) Maintenance ($12.4/hr) Discretionary ($10.7/hr)

183 NEW YORK METROPOLITAN TRANSPORTATION COUNCIL New York, New York Major model feature Detailed feature / sub- model Characteristics At-work ($40.0/hr) Household / person characteristics Household income (low/med/hi) – included in tour frequency and mode choice utilities; however it does not directly affect VOT. Network simulation tool Simulation type Static user equilibrium multi-class assignment (TransCAD software) Representation of priced highway facilities Tolls converted to travel time units based on the average VOT for each vehicle class; additionally area pricing is modeled by adjusting the cost skim matrix. Demand – Network Equilibrium Feedback implementation Feedback is implemented including destination choice, mode choice, stop frequency & location, and assignment. MSA method is applied for link volumes and trip tables in parallel. Full equilibrium requires 8-9 iterations. Practically acceptable equilibrium is achieved after 3 iterations. Surveys and other data sources for model estimation / calibration / validation Household travel survey 11,000 households, 1996/7, Household Interview Survey with all trips/activities recorded for all household members for 24 hours. Survey of existing toll road users Not used Stated Preference survey Not used Traffic counts Extensive set of 3,000 traffic counts from different sources updated on a yearly basis. Traffic and transit counts by major screenlines were use for the model update and re-calibration in 2002 and 2005.

184 A.1.2.3. Montreal, Quebec MINISTRY OF TRANSPORTATION OF QUEBEC Montreal, Quebec, Canada Major model feature Detailed feature / sub- model Characteristics Spatial scale Regional Demand model structure Tour-based sample-enumeration model with micro- simulation components Modeled pricing impacts (traveler responses), sub- model structure, form of utility function, incorporation of pricing Household auto ownership and tour frequency models Expanded records from the extensive household survey (5% of the regional population). Every tour/trip record has and expansion factor ( ≅ 20) that is calculated for each future year to account for population growth by zone and socio-economic group Primary tour destination choice Fixed in the sample according to the observed destination for each record. The expansion factors for future years account for employment growth by zone and occupation type Tour mode choice Nested logit model segmented by 3 purposes (work, maintenance, discretionary) with 6 modes: 1=Auto driver, 2=Auto passenger, 3=Walk to transit, 4=Drive to transit, 5=School bus, 6=Non-motorized. Mode utilities include time and cost variables with coefficients segmented by income and gender; Stop frequency & location choice Fixed in the sample according to the observed trips in each tour record Trip mode choice Included binary toll vs. non-toll sub-choice for auto modes (SOV and HOV); included additional nested level for transit by the main mode (bus, subway, and rail). Time-of-day choice Fixed in the sample according to the observed trip departure time in each tour record Willingness to pay / VOT and user segmentation Vehicle classes Vehicle classes in the time-of-day specific assignments and assumed VOT: 1=Auto / toll ($8.0/h) 2=Auto / non-toll 3=Commercial / toll ($12.0/h) 4=Commercial / non-toll 5=Light trucks / toll ($24.0/h) 6=Light trucks / non-toll 7=Heavy trucks / toll ($36.0/h) 8=Heavy trucks / non-toll Vehicle occupancy categories Not considered; auto drivers and passengers are not explicitly linked Trip purpose segmentation, VOT in CAD$1998 (currently deterministic for each purpose, gender, income group, and time-of-day) Gender Income TOD Purpose Work Main Disc Male Low Off $7.3 $4.0 $3.0 Peak $10.3 $4.0 $3.0 High Off $10.2 $4.0 $3.0 Peak $10.2 $4.0 $3.0

185 MINISTRY OF TRANSPORTATION OF QUEBEC Montreal, Quebec, Canada Major model feature Detailed feature / sub- model Characteristics Female Low Off $7.3 $6.4 $6.0 Peak $10.3 $6.4 $6.0 High Off $10.6 $7.3 $7.6 Peak $10.6 $7.3 $7.6 Household / person characteristics Additional household variables (car ownership, presence of children) – included in mode choice utilities; however they do not directly affect VOT. Network simulation tool Simulation type Static user equilibrium multi-class assignment (EMME software) Representation of priced highway facilities Tolls converted to travel time units based on the average VOT for each vehicle class. Demand – Network Equilibrium Feedback implementation Feedback is implemented including mode choice, and assignment. MSA method is applied for level-of- service skims. Full equilibrium requires 5-6 iterations. Practically acceptable equilibrium is achieved after 3 iterations. Surveys and other data sources for model estimation / calibration / validation Household travel survey 60,000 households, 1998, Household Phone Interview (Origin-Destination) Survey with all trips/activities recorded for all household members 12 years old and older for 24 hours. Survey of existing toll road users Not used Stated Preference survey Special SP survey to estimate willingness to pay; 1,000 persons driving in the priced corridors with 11 SP scenarios offered to each person. Traffic counts The base year (2000) expansion factors were validated and adjusted to match traffic counts (about 500 locations)

186 A.2. Technical Details of Survey Methods for Pricing Studies The development of models for road pricing analysis requires supporting data collection and travel surveys. In many respects, the types of surveys that are used to evaluate potential or implemented road pricing projects are similar to those used for other transportation planning purposes. There are often some important considerations in the context of a road pricing study that affects both the types and design of special surveys that are required, In theory, any change in transportation service, including changes in road prices, could affect all travel-related decisions, ranging from residential location and auto ownership to activity participation, destination, mode, and route choices. For facility pricing projects, route choice is an obvious choice dimension for which survey data can be used to refine existing models. Time-of-day choice is important for projects that include time- variable pricing, and surveys for these can be designed to support time-of-day or peak- shifting models. Destination choice can be affected by area pricing schemes, but this element is typically already included in regional travel demand forecasting models. A unique choice specifically related to road pricing is whether an individual acquires a transponder to allow participation in electronic toll collection (ETC). The transponder acquisition decision is in some ways analogous to a decision whether to acquire a monthly transit pass because, as with transit passes, most tolled facilities give ETC discounts and those discounts can range from 10% to 50%. The transponder acquisition choice is different, however, because it can be mandatory for access to a facility–some facilities such as the California SR 91 Express Lanes require a transponder–and because there can be non-trivial upfront costs associated with acquiring the transponder. As a result, surveys that support modeling transponder acquisition choices can be important components of the data collection program. A comprehensive household travel survey is generally needed to develop a regional transportation model that can serve as the source for Value of Time (VOT) and other relevant model parameter estimates. However, there is a growing recognition that the household survey data have to be supported by complementary/project-specific Revealed Preference (RP) and/or Stated Preference (SP) surveys. This is especially crucial for start-up projects in regions with no prior experience with highway pricing where the RP survey cannot provide direct information about and SP surveys are typically designed to address willingness-to-pay factors relevant for road pricing (value of time savings, value of reliability). Survey data collection can also support other model development data needs, including HOV/HOT lane usage and payment media choice. The sub-sections that follow describe the types of surveys that can be used to support the development of travel forecasting models for road pricing projects. These surveys can be configured as part of a data collection program that is designed to support the evaluation of proposed road pricing projects as they move through stages from preliminary screening to investment grade analysis to post-implementation refinement. The concluding sub-section presents general recommendations for the design of such a survey and data collection program.

187 A.2.1.Travel Pattern Surveys (“Revealed Preferences”) As for general transportation planning studies, surveys of current travel patterns can be effectively used to support pricing studies. These surveys can be administered as household-based travel/activity diaries or as trip-intercept, origin-destination, and route surveys. A.2.1.1. Household-Based Travel /Activity Surveys The standard regional household travel surveys, with its complete and detailed accounting of all travel within a household on surveyed day (or multiple days), can provide useful information about the general patterns of travel within the region in which they are conducted. All major U.S. metropolitan regions have had household travel surveys conducted to support development of their regional travel forecasting models and, in general, road pricing projects would not themselves require new household surveys. Household travel surveys represent the only holistic framework that allows for an analysis of the entire daily activity pattern of persons and households. This type of survey is necessary for understanding the impact of congestion and pricing on the whole hierarchy of choices, from the short-term trip-level responses to long-term responses that relate to activity pattern (trip frequency) and location of activities. The important principal advantage of a household interview survey is that all related travel dimensions, and generally travel segments of interest, can be analyzed in one coherent framework, while most of the other surveys would focus on a certain trip or fragment of the daily pattern of individuals. Household travel surveys provide general information about the number of trips (or tours) made, time-of-day distribution of those trips, their geographic distribution and the modes used. They also generally include information about the costs of those trips. However, in most regions, these are dominated by vehicle operating costs which are perceived very differently across individuals and in most cases appear to include little more than gasoline costs, and where parking costs may represent the only significant trip-based costs. More importantly, household travel surveys are limited in the sense that they might not provide a sufficient number of observations of actual toll road users, as well as do not provide enough (in a statistical sense) trade-offs in terms of travel time, reliability, and price to reliably estimate VOT, and other parameters. Toll costs are not specifically enumerated in all household surveys, in part because many regions do not currently have any toll facilities and, in any case, these surveys do not typically elicit more detailed information about route choices. Since these surveys do not completely capture the details of travel which might be affected by a road pricing project (e.g., route choice and toll costs), it might be appropriate to selectively sample with a refined household travel survey instrument. Some household travel surveys that have been conducted to support toll facility analyses have included questions designed to collect route choice information. Those survey instruments have included focused questions detailing route segments for trips that are

188 made between origins and destinations served by toll facilities, and so either used those facilities or could have used the facilities. Many, if not most, trips are made between points that are not served by toll facilities, even in regions that have many such facilities. Because reporting the route choice information can add significant respondent burden, it is useful to screen for selected origins and destinations as the data are captured so that this information is collected only for those trips where it is relevant. This screening requires both real-time geocoding of origins and destinations and information about the locations served by toll facilities. The latter can be provided through skim tables from the regional travel forecasting model. Figure 68 below shows a portion of the section of a recent household survey that was designed specifically to collect detailed information about toll route choices, as well as the other “usual” trip details [Resource Systems Group, 2004 ]. For each trip segment, respondents were asked whether they used a toll facility. If they indicated that they had used a toll facility, they were asked where they entered and exited the facility. From that information, along with information about ETC use, toll costs can be calculated directly. This additional information can provide a rich base of information to support the modeling of toll route choices.

189 Figure 68: Example South Florida Household Survey Route Choice Screens

190 A GPS-based supplement is increasingly a feature that is included with household surveys since it can provide detailed route information for all recorded trips. Either vehicle or person-based GPS data collection can be used, but vehicle-based GPS data collection is generally more useful for collecting route information, assuming that tracking routes for transit and pedestrian/bicycle alternatives is not necessary. In theory, the data from these route traces, along with information on ETC use in the vehicles, could be used to support modeling of price response, but to date there do not appear to have been any such applications. If tolled facilities run parallel and very close to non-tolled lanes (e.g. the conversion of an existing HOV lane to a HOT lane), then high resolution GPS location data are needed for a level of accuracy that can identify whether or not a vehicle has used the tolled facilities. Household travel and activity surveys universally include information about the time that trips were made and these provide some information about current time-of-day choices. However, it is very difficult to infer from those observed choices the degree to which individuals have flexibility to shift their trips/activities to other times-of-day. In addition, in regions that have significant peak period traffic congestion, the observed choices may reflect shifts that individuals have already made away from their preferred times in order to avoid that congestion. While it would be possible to include questions in a household or GPS-follow-up survey to collect information on time-of-day flexibility, those questions must be asked selectively to avoid adding significant respondent burden to an already- burdensome survey process. Typical household and activity surveys cost in the range of $100 to $150 per household and sample sizes can vary from several hundred to several thousand depending on the survey’s purpose. A.2.1.2. Origin-Destination Surveys Surveys that collect information about the origins, destinations and other details have been widely used to determine the characteristics of trips that are observed at selected locations [Hagen, 2006 ]. These types of surveys are particularly useful for characterizing the trips that currently travel in particular corridors that are, or might be, served by a toll facility and the trips that cross into or out from a cordon that might be subjected to area pricing. This type of focused information is especially useful in estimating the numbers and types of trips that might be affected by facility or area pricing. Although regional travel forecasting models can also be used to synthetically provide this information, those models are typically not refined sufficiently to estimate these details as precisely as can be done with an origin-destination survey. Significantly higher precision in the origin-destination table can be provided using a matrix estimation procedure, assuming that sufficient count data are available to support that procedure. A combination of origin-destination survey data, synthetic modeling and matrix estimation can be used to provide the additional precision required for road pricing studies.

191 Sampling trips for origin-destination surveys is a special challenge. Roadside interviews typically have high response rates and, generally, result in high quality data for the facilities on which the interviews are conducted. However, these surveys are logistically difficult and at least one state – Florida – does not allow them because of safety and operational concerns. One alternative is to record license plates of a sample of vehicles on a facility, match those plates with addresses using vehicle registration data and use mailout/mailback questionnaires. Response rates for these license plate surveys, however are typically much lower and the small uncontrolled sample problematic in analysis. An alternative approach for origin-destination surveys is to intercept vehicles at locations where they are already stopped and hand drivers mailout/mailback questionnaires. This alternative has been widely used on toll facilities; questionnaires are distributed at toll plazas where vehicles are stopped to pay tolls. However, most modern toll facilities operate with a combination of conventional plaza booths where vehicles stop to pay cash tolls and lanes that allow vehicles with electronic toll collection (ETC) devices to pass through without stopping. While origin-destination survey forms can be distributed to cash customers at these plazas, safety and operational issues generally preclude distributing forms to ETC customers. It would be possible to record license plates of ETC customers, match addresses and mail survey forms to those addresses, but there is generally a more efficient and reliable way to sample ETC customers using ETC data. The agencies responsible for electronic toll collection maintain databases that record vehicle transactions at all of the locations where tolls are collected. Each of these transactions is linked to ETC registration data that includes a mailing address and other contact information. While privacy restrictions generally prevent the agencies from providing these data to third parties, they can generally use the data themselves to contact a sample of customers who use a facility at selected times. For example, the 11-state E-ZPass Consortium in the Northeast has supported numerous travel surveys by sampling E-ZPass transactions at selected facilities on selected days and then mailing survey questionnaires to those customers. Ideally, the transactions are sampled for the same time periods and locations as for cash surveys so that the two samples can be weighted back to a common base. Data for weighting can be provided by the cash/ETC transaction counts at each location, but it is also useful to collect other primary data such as occupancy counts and vehicles’ states of registration from direct observations to adjust for possible differences in response rates (Jacobs and Spitz, 2006). In addition, for toll facilities with multiple interchanges, the ETC-based origin- destination survey data can be re-weighted so that they accurately represent the on-off patterns reflected in ETC transaction data. Figure 69 and Figure 70 show examples of origin-destination survey forms. Figure 69 is a form that was used as part of license plate matching origin-destination survey conducted on major toll-free routes in South Florida for Florida’s Turnpike Enterprise. Since the survey stations were on toll-free routes, a simple toll route choice question is included to identify trips that used toll facilities. Both English and Spanish languages are

192 incorporated in the form and an accompanying web-based questionnaire was offered as an alternative for those who preferred to complete the survey on the web. Figure 69: Example Origin-Destination Survey Form Figure 70 shows one of almost 100 forms that were developed as part of a very large- scale origin-destination survey of the New York metropolitan region’s major bridges and tunnels [Jacobs and Spitz, 2006 ]. That survey effort included distributing mailback forms to cash customers at all of the toll plazas and mailout/mailback sampling of ETC customers. The ETC sample was generated from ETC records of those customers who used the facilities at the same time as the cash form distributions. The survey form collected information both about the sampled trip/direction and about the return or previous trip. Traffic counts and concurrent auto occupancy and vehicle state of registration measurements were used to develop expansion weights applied to the sample records for analysis.

193 Figure 70: Example Toll Facility Intercept/ETC Origin Destination Survey Form Costs for origin-destination surveys vary considerably depending on the administration method and the complexity of the application and sample sizes similarly depend directly on the ways that the data are intended to be used. A.2.2.Stated Preference Surveys For more than 20 years, Stated Preference (SP) surveys have been used to estimate values of travel time and other parameters related to the effects of tolls and road pricing (see, for example, [Adler and Schaevitz, 1989]). SP surveys include a set of hypothetical scenarios in which conditions (e.g., travel times, tolls) are varied and respondents are asked to indicate what they would most likely choose under those specified conditions. The conditions are varied according to an experimental design that optimizes the information about the respondents’ preferences from each of the scenarios which they evaluate.. SP surveys are especially useful in applications where an alternative such as a toll facility does not currently exist, but is being planned for the future. In those types of applications, revealed preference surveys are not useful for estimating price effects because road prices, which are the variables of interest, do not vary (all zero) across trips within the region. While other cost elements, such as operating costs, do vary

194 across trips, those variations are highly correlated with trip lengths and travel times and thus generally do not provide reliable indications of the effects of price on travel choices. In regions that do have existing toll facilities, revealed preference data from household and origin-destination surveys can be used to estimate price effects, but there are also complementary uses for stated preference surveys in these regions. The uses include: • Estimating the effects of prices that are outside the range of existing prices or pricing structures, such as time-varying tolls or that do not currently exist in the region, • Estimating the effects of travel time reliability and other variables that are difficult to measure and/or associate with revealed preference observations, • Determining the effects of new features of facilities such as open road tolling, express lanes or “fair lanes”, • Determining the effects of pricing on choices for alternatives that do not exist or that are not easily captured within revealed preference surveys. Examples would include effects of open road tolling on transponder acquisition or of variable pricing on time-of-day choices. In addition, SP surveys can be used to sharpen the information on price effects that is otherwise provided by revealed preference surveys, where the lack of variation in tolls may be problematic. There are often compelling reasons for doing this. • The confidence interval of the marginal rate of substitution between time and cost (value of time) as estimated using revealed preference data alone is often quite wide. This is a result of the measurement errors associated with inferring travel times for chosen and alternative routes, and of the inherent statistical error associated with estimating the value for a ratio of two random variables, particularly when the two random variables are correlated with each other, as travel cost and travel time tend to be. • Stated preference data can be used to estimate the distribution of preferences across the population. Many differences in preferences, such as different travel time sensitivity across trip purposes and varying degrees of cost sensitivity across income groups, can be explained using systematic effects that in turn can be modeled directly. There are other random effects that cannot be measured or directly represented in a preference function, however, but which can be quantified using methods such as mixed-logit or hierarchical Bayes modeling. These methods can estimate the distribution of cost sensitivity and of values of time across the population and this distributional estimate and can in turn be used to help estimate population responses to different pricing levels. For all of these reasons, SP surveys have been used to assist in the evaluation of most of the major road pricing projects, both within the U.S. and internationally. These surveys can take on many different forms.

195 A.2.2.1. Choice Dimensions and Scenario Design The stated preference surveys that have been conducted to support road pricing projects have most often focused on the choice between tolled and toll-free routes. For conventional toll facility studies, these surveys would typically present two alternatives; a toll-free route with a given travel time and an alternative tolled route with a lower travel time and a toll at some level. Many road pricing projects, however, involve more complex effects beyond simply influencing route choice. Some projects, such as HOT-lanes, affect occupancy and mode and so the stated preference scenarios would include other modes and occupancy levels as available choice alternatives. For projects that have time-varying prices, different travel periods should be included among the stated preference alternatives. For area pricing projects, the scenarios could allow alternative destinations. In some special cases, effects on trip frequency may also be included in the stated preference experiments. For example, a recent study, illustrated in Figure 71, evaluated a proposed new bridge toll on a facility for which the only toll-free alternative was a significantly longer route. In this case, it was possible that discretionary trips could be reduced, so stated preference experiments were constructed to assess the reduction that might result from different toll levels [Falzarano & Szeto, 2003 ]. Figure 71: Example SP Scenario to Measure Possible Trip Frequency Changes

196 Finally, as noted earlier, transponder acquisition choice can be a significant issue in the forecasting of demand for priced facilities since the availability of a transponder can affect both access to a cashless facility and the price charged for using a facility. In addition, both anecdotal and quantitative evidence suggests that individuals who use transponders are simply less price sensitive than those who pay out-of-pocket because the price is less apparent to the latter. Transponder acquisition can be modeled using data from stated preference experiments in which the decision whether to acquire a transponder is included as a choice alternative. These experiments should be designed in a way that reflects the fact that this choice is linked to the likely use of the toll facility, and to the level of discount applied on ETC transactions. Figure 72 shows an example transponder acquisition SP scenario that was included in a survey research program by Resource System Group that preceded construction of the California SR-91 Express Lanes. Figure 72: Example Transponder Acquisition Choice Experiment Figure 73 shows a more typical stated preference scenario for a general road pricing study [Adler, et al, 2007 ]. In this scenario, respondents are presented with tolled and toll-free options, time-of-day and mode alternatives.

197 Figure 73: Example SP Scenario with Typical Pricing Choice Options A.2.2.2. Trip Attributes Relevant for Pricing Studies Travel times and toll prices are the primary attributes in most stated preference experiments done for road pricing. The trade-offs between travel time savings and extra cost associated with tolls, are expressed in terms of Value of Time (VOT). In the presence of road pricing However, there are other attributes that may also be significant in travelers’ choices. Some of the other attributes or features that have been tested in stated preference experiments for road pricing projects include: • Travel time components – time in free flow conditions and time in congested traffic, • Travel time reliability,

198 • Occupancy-based toll levels, • Fair lanes policy, • Commercial vehicle restrictions, • ETC discounts, • Travel time variability, • Driving distance along the route, and • Non-toll “running” costs. This is by no means an inclusive list as individual projects may have unique features for which stated preference surveys can provide information. Examples of this from past studies include testing the willingness of travelers to use toll facilities having a long tunnel section vs. at grade sections, and having a smooth asphalt surface vs. jointed concrete. It is important to recognize that not all of these variables or features are necessarily salient for all travelers. For example, a recent study found that vehicle operating and maintenance costs are not even considered by over ¼ of the travelers surveyed, and those who did consider them, applied significant discounts to their effects relative to tolls [Hensher, 2007 ]. In addition, some attributes such as travel time variability may be important primarily as conditioning variables that affect the actual travel conditions faced at the point in time at which a decision is made. So, for example, travel time variability may be less relevant to spur-of-the-moment route choice decisions because travelers can and do make route choices dynamically, based on actual conditions at a given point in time which often include a fairly accurate estimate of current travel times based on real-time traffic reports or prior direct observations. Variability and reliability issues may be most important for frequent corridor uses in their decision of whether or not to purchase a transponder. A.2.2.3. Choice Context In all stated preference experiments, respondents are asked to respond to choices in some defined context. In some past studies, this has been couched in a very general context, such as assuming travel in general or for some given trip purpose for which respondents are asked to simply choose between different travel time and cost alternatives. The limitation of this approach is that travel circumstances for a given individual and even for a given trip purpose for that individual may vary significantly from day-to-day. Without additional guidance, respondents may respond assuming a memorable, but atypical context or may assume only a typical condition and thus not reflect the variations around that typical experience. Other stated preference experiments are framed around a particular trip, which could be one that was in process at the time the traveler was asked to participate in the survey, or it could be a selected, recently completed trip. In either case, respondents are asked to describe that trip, thus creating a “base case” revealed preference observation, and

199 then to assume that the same travel context would apply for each of the stated preference experiments. Assuming that the trips are sampled in a way that covers all of the likely contexts, this approach ensures that both the typical and less typical travel conditions are accounted for. Having the respondent fully describe the trip also helps ensure that respondents consider all of the important conditions that might influence or constrain their choices. There are at least two important challenges, however, in framing stated preference experiments around a particular trip context. One of the most difficult challenges is to ensure that the scenarios described in the experiments are all realistic alternatives for each respondent’s trip. If, for example, some of the new travel times presented in the experiments are unrealistically short, respondents will discount those alternatives in ways that are not easily predictable, and which in any case will not provide reliable data. Using computer-based instruments or pre-processing respondents’ trip information can be helpful in generating realistic stated preference experiments. Given the wide variety of trips that are made, and the need to vary attributes sufficiently, however, it can be very difficult to ensure that realistic scenarios are created for every case, so it is always important to review the resulting survey data to check for possible outliers. The second challenge in framing stated preference experiments around a particular trip is in understanding how the response for a single trip might translate into longer-term behavior. In focus groups conducted to understand travelers’ response to express lanes and variable pricing, several participants have indicated that they would budget their travel so that they did not spend more than a certain amount per month. If traffic conditions were such that tolls were consistently very high most days, they would be more selective about what threshold they would use to choose between tolled and toll- free lanes. On the other hand, if tolls were only occasionally very high, they would use a somewhat lower threshold. Some stated preference surveys have included questions designed to understand these travel budgeting issues, and to look at the frequency of using priced facilities over a number of trips (see Figure 4). Another common approach for SP surveys supporting the modeling of road pricing is to ask respondents if they have made a recent trip in the relevant corridor, and, if so, to ask for details on the most recent trip and use the information to customize the SP choice context. The use of the “most recent” trip rather than the most “typical” one is meant to avoid bias and replicate a random sample, just as we ask household survey respondents to complete a diary for a specific assigned day and not necessarily a typical day for them. A design issue that commonly arises is the limit on how distant in the past the most recent trip can be in order to qualify for the survey. We do not want to be so restrictive that it is difficult to find respondents, but on the other hand do not want to include trips that are so far back in the past that people have forgotten some of the important details. A typical strategy is to set the limit at 1 or 2 weeks prior to the interview, while a retrospective limit of longer than 1 month is rarely used in practice.

200 A.2.2.4. Instrument Design Stated preference surveys have been conducted using several different types of instruments. One important challenge is that the stated preference experiments generally each involve trade-offs among several variables that vary across two or more travel alternatives. It can be difficult for respondents to keep all of this information in their minds unless it is presented visually and, for this reason, telephone-only instruments are rarely used. However, hybrid instruments can be used where trip context information is collected over the phone and the stated preference experiments are provided separately by mail or over the web. In addition, simplified experiments can be designed that are more amenable to phone-based administration. Figure 74 shows the general form of questions that were used to estimate values of time for a Georgia DOT I-75 value pricing study using a phone interview approach (NuStats, 2006). Notes: Values of [$], [#] and [N] were customized according to previous answers about the specific trip, and varied according to an experimental design. Each respondent was given either Order A or Order B at random to control for possible order-related bias. Figure 74: Example of Stated Preference Design for Phone Interview SP experiments can be designed as printed forms, but there may be several versions required to cover all likely contexts and to cover the required experimental design. More

201 commonly, stated preference experiments for road pricing projects use computer-based instruments that can be more flexibly adapted to varying contexts and to more complex experimental designs. Significant detail can be used from the trip context description to tailor the experiments so that they are realistic for each trip. Furthermore, trip origins and destinations can be geocoded in real-time and those data can be used, also in real- time, to determine travel time and cost ranges based on transportation network model data. Both computer-assisted personal interviews (CAPI) and computer-assisted self interviews (CASI) have been used and web-based administration is increasingly common. See Figure 75 for examples. Figure 75: Example CASI/CAPI-Based SP Questions Another common strategy when face-to-face interviews are infeasible or inefficient is to use a multi-stage CATI and mailout approach. Responses from a previous CATI or mailback survey are used to create a customized, respondent-specific printed SP questionnaire. The questionnaire is then mailed to the respondent, who can then be asked to provide their answers over the telephone, via the internet, or by mail. This strategy is often used to carry out a follow-up survey to a regional household travel survey.

202 A.2.2.5. Sampling Sampling for stated preference surveys can also be conducted in any of several ways. For facility-based studies, some type of intercept sampling is often the only viable alternative. This can be because the population using the facility or corridor is widely dispersed geographically and may, for example, include significant numbers of trips made by individuals who live well outside the region in which the facility is located. Intercept sampling can be conducted using the methods described earlier for origin- destination surveys but it can also be accomplished using intercepts at activity centers in the corridor of interest. For area pricing or cordon pricing, it may be most efficient to intercept people within the potential priced area. For studying corridor-specific projects, it is often effective to use Random Digit Dialing (RDD) or address-based sampling within the residential areas that would be served by the project. For broader regional studies, the options are wider and include more standard phone, mail or web/email recruiting. Stated preference surveys have also been administered along with conventional household travel/activity surveys, usually as an add-on to some fraction of those surveys. In general, the sampling plan for a stated preference survey should result in a representative sample of trips within the area of interest. It is important to sample a sufficient range of travelers and trip types to support the statistical estimation of coefficients of a choice model. By collecting data from the full range of traveler and trip types, it is possible to identify the ways in which different characteristics affect choice behavior. These differences can then be reflected in the structure and coefficients of the resulting choice model. The survey sample that supports mode choice model estimation does not need to be perfectly population proportional as long as: 1) any of the behavioral differences are properly represented in the model and 2) the model is applied for forecasting using appropriate population proportions and/or sample weights. There are no universally-accepted guidelines for the sample size required for stated preference surveys. Sample sizes of 400-1,000 are common for stated preference-based road pricing surveys, but larger or smaller sample sizes may be appropriate, depending on several factors: • The number of scenarios presented to each respondent: ranges are typically from four to 16 scenarios (more scenarios generally means that the respondent sample size can be smaller, although this is not a one-to-one tradeoff—generally the more responses from each individual, the less additional statistical information that each new response provides), • The number of behaviorally-distinct traveler segments: models may be segmented by trip purpose, time period, vehicle occupancy and other dimensions (more segments generally means that the respondent sample size should be larger), • The type and required precision of the estimates: estimating values of time which involves computing the ratio of two coefficients (which themselves are random variables) requires a fairly large sample size to yield a tight confidence interval so “investment grade” studies that rely on high precision in this estimate will require

203 large samples. Conversely, general planning or feasibility studies may require smaller samples. As with the sample size, the costs of stated preference surveys can vary significant depending on several factors, such as the study’s complexity, survey sample size and method of administration. Costs ranging from $30,000 to over $300,000 have been reported, with typical U.S. “investment grade” stated preference studies costing somewhere in the middle of that range. A.2.3.Special Issues & Survey Types A.2.3.1. Surveys of Commercial Vehicles The approaches and methods described in previous sections generally apply equally to both passenger and commercial vehicle surveys. However, there are at least two special issues to be considered in designing and administering stated preference surveys of commercial vehicles. First, the driver of the vehicle may or may not be the person who makes the trip decisions that could be affected by road pricing changes. Independent owner-operators generally make these decisions themselves but other fleet and common carrier drivers most often operate under guidelines established by others. As result, the survey administration plan for stated preference surveys of commercial vehicles should identify ways of ensuring that the survey is completed by the actual decision-makers. The second issue for commercial vehicle stated preference surveys is that values of time are typically much higher (and the tolls are correspondingly higher) than for passenger vehicles. The implied values of time for the stated preference scenarios should be checked to ensure that they encompass an appropriate range for each vehicle type. In addition, there may be a wide variation in values of time across the commercial sector, varying with characteristics such as the type and value of goods carried, the distance traveled, whether the vehicle travels full or empty, and the time of day or night. Capturing these sources of variation in a representative way requires care in sampling strategies. A.2.3.2. Behavioral Experiments and Follow-up Surveys Stated preference surveys ask respondents to put themselves into hypothetical situations (scenarios) and indicate what they would likely do in those situations. A logical extension of that approach is to create real pricing experiments for a sample of individuals. Two current studies in the U.S. illustrate this approach. The Puget Sound Regional Council is conducting an FHWA-sponsored pilot study in which “Travel Choices Meters” were installed in 500 vehicles. The drivers were given $600 to spend (or not spend) over a year under conditions in which the meters deducted from this account in accordance with a simulated variable road pricing program. Figure 76 shows the toll map and schedule provided to respondents, and the in-vehicle unit that was used to show respondents what they were being charged at any

204 moment, as well as the cumulative amount they had been charged for the trip and in total. Drivers’ responses to the pricing were recorded using GPS devices installed in the vehicles, providing GPS traces on 750,000 trips. Preliminary analysis has indicated that such a pricing system could reduce VMT by about 10%. For more information, see www.psrc.org/projects/trafficchoices/. A similar effort is being undertaken by the University of Iowa in which 1,200 cars will be outfitted with a GPS device and mileage- based charges will be levied against an initial budget, although the emphasis in this latter study is more on testing in-vehicle technology than it is on studying behavioral responses to pricing. Figure 76: Traffic Choices Study Toll Map/Schedule and In-Vehicle Price Meter Provided to Respondents Follow-up surveys conducted after a road pricing project is implemented can be especially useful for both documenting the effects of the road pricing project and for determining how the project might be refined to better accomplish its objectives. Significant follow-up survey efforts have accompanied several major road pricing projects. Examples include work related to the SR-91 Express Lanes project [Sullivan, 2000] and the Lee County LeeWay variable bridge pricing project [Burris, et al, 2004]. A unique panel survey in Minneapolis [Zmud et al, 2007 ] included a series of SP questions

205 and other types of questions with the same respondents both before and after the opening of the I-394 MnPASS HOT lane facility. A.2.3.3. Attitudinal / Public Opinion Surveys The surveys described in the previous sections are designed to provide information that can be used to estimate the effects of road pricing changes or new toll road facilities. Surveys and other research methods can also be used to inform decision makers about public attitudes and opinions regarding pricing policies or programs. Several of the FHWA Value Pricing Pilot projects and many of the private toll road ventures have included significant qualitative research components. This qualitative research has consisted of both traditional focus groups and individual depth interviews (IDIs) for selected key stake-holders. Qualitative research provides an opportunity to learn how these projects are perceived, what types of concerns they raise with different segments of the general public, both user and non-users, and how those concerns can be addressed. For example, the early qualitative research on variable pricing suggested that many individuals have a hard time understanding why a policy of increasing peak period prices can improve travel times: they assume that most peak periods travels are people making work trips that cannot be made at other times. They are also concerned about how any additional money that is collected is used; they strongly prefer that it be used to improve travel conditions in the corridor in which the tolls are collected. Quantitative research can be used to determine public opinions about a proposed project or program before it is implemented and/or determine effects after implementation. While opinion polling is a well-established method for a wide range of applications, there is one important caveat regarding these methods for road pricing projects and programs. Qualitative research has indicated that individuals in general have a difficult time understanding details of how modern road pricing approaches work or would work, even in areas with existing conventional tolling and especially in regions that have no existing tolled facilities. For example, in a recent focus group for a proposed express lane project (new ETC-only lanes in the median of an existing interstate highway) participants were asked if they had heard about any proposed projects and what they thought about them. Most said that they had heard about the project it and opposed it. On further prompting, the opposition was determined to stem from their perception that the lanes would have conventional toll plazas, which they knew “wouldn’t work” (the perception came from or was supported by a local newspaper article that had simply described the new lanes as “toll lanes”). The point of this anecdote is that individuals can have a hard time imagining the implications of a new road pricing approach and, unless those implications are appropriately conveyed in the survey questionnaire, their stated opinions will not be fully informed. In that regard, qualitative research can be very valuable in informing the design of a quantitative survey instrument.

206 There have been several public opinion/attitude surveys conducted in association with road pricing projects. The most comprehensive effort was a three-wave panel survey designed to be conducted before and after implementation of the I-394 MnPASS project in the Minnesota Twin Cities region [Douma, et al, 2006]. The survey measured baseline pre-project attitudes and later waves were designed to measure changes in those attitudes after the project was operational. Other surveys have been conducted primarily to inform decision makers of the level of project support. As with the other methods, the costs of this type of research can vary considerably. Qualitative research projects involving focus groups commonly include 6-12 groups at approximately $5,000 per group. Public opinion and attitudinal surveys typically have sample sizes of 800-1,200 completes at $30-$60/complete. A.2.3.4. Related Data Collection Methods Traveler surveys comprise only part of the data collection effort that is required to support reliable modeling of the shifts in travel patterns that might result from road pricing changes. Good traffic count coverage is important for providing model validation data. For road pricing studies, it is also useful to compile traffic counts by time-of-day, season, vehicle type, and, for toll facilities with mixed cash/ETC, transaction type. For toll road facilities, toll plaza, ramp, and ETC data can be used together with matrix estimation procedures to develop accurate on-ramp to off-ramp (facility origin- destination) estimates. Travel speeds and variations in those speeds determine the travel times that drivers face when they decide between competing routes. The speeds used in travel forecasting models are often not calibrated closely enough for pricing analysis to accurately represent the travel time differences between competing routes, particularly as they vary with traffic volumes and travel conditions. Travel surveys can be designed so that they collect reported travel times but those are also not necessarily accurate representations of actual speeds. Special travel time surveys can be conducted using GPS devices, but the most comprehensive data on travel speeds and variations in those speeds is likely to come from other sources such as ETC operations data (for tolled facilities with multiple plazas), traffic operations centers or the commercial traffic monitoring services that now collect data in most major U.S. urban areas. A.2.3.5. Combining RP and SP Data It is critical to use both RP and SP data in order to get an accurate model of traveler responses to pricing. While SP data is typically necessary to estimate distributions across the population and provide detailed market segmentation information, it also has potential shortcomings. Research from the limited number of managed lane systems already in place has indicated that willingness to pay estimates estimated from RP data tend to be roughly twice as high as those derived from SP data, in the same context. Reasons hypothesized for this are that travelers have inaccurate perceptions of the time that they actually save on the systems, as well as evidence of possible “protest” responses against pricing options that may outlined in the SP exercises. Carefully

207 pooled analysis of both types of data will be necessary to take advantage of the strengths and avoid the shortcomings of each. From the formal statistical point of view, combined model estimation that is based on both RP and SP data, benefits synergistically from the different nature of RP and SP data. The data are not just blended together, but are used in the best possible statistical manner in which the SP-related parameters are properly scaled by the observed data from RP. This allows for elimination (or mitigation) of many systematic biases inherent to pure SP surveys. One strong data set in this regard is the data from the MnPASS system in Minneapolis. As yet the trip data from the panel surveys have not been combined with objective data on HOT lane travel times and prices and general lane travel times in order to derive RP models. Another important data set for pooled analysis is the data from the Traffic Choices pricing experiment in the Seattle region mentioned above. That experiment combines some of the best elements of SP research—an experimental design of prices that vary by time and space—with critical elements of RP data—objective GPS measurement of travel speeds, times and prices, along with responses that involve actual payment of money. A.2.4.Choice Attributes to Support Advanced Models A.2.4.1. Measuring Distributions of Willingness to Pay Tolls It is vital to collect data that will allow us to measure the distribution of the value of time across the relevant traveling population. For example, a recent Stated Preference study of users of the new MnPASS HOT lane facility [Zmud, et al., 2007], used a survey method to explicitly estimate each respondent’s individual willingness to pay and associated Value of Time (VOT), and found much more variation across the sample than could be captured through observable segmentation variables alone. The observed distribution resembled the log-normal distribution seen in Figure 77 and compared to a symmetric normal distribution. Legend: - Normal - Log-normal Figure 77: Possible Value-of-Time Distributions

208 [Ben-Akiva, et al, 1993] pioneered an econometric approach to directly estimate the parameters of such log-normal VOT distributions from typical SP and RP data sets. This research was done as part of a study for Cofiroute of proposed tolls on the French national motorways, and the resulting distributed models were applied using a customized multi-user-class static assignment routine. Now, 15 years later, there is a variety of approaches and software products available for estimating and applying such models estimated using a mixed logit approach and applied using stochastic microsimulation. In particular, distributed VOT was incorporated in the San-Francisco County Transportation Authority (SFCTA) model (see Figure 78) that is currently being applied in the comprehensive congestion pricing study in the Bay Area. Work Value-of-Time Lognormal Probability Density Functions 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 $0 $3 $6 $9 $12 $15 $18 $21 $24 $27 $30 $33 $36 $39 $42 $45 $48 Value of Time ytilibab orP HH Income: $20K HH Income: $50K HH Income: $75K HH Income: $125K Figure 78: Distributed VOT in the SFCTA model A.2.4.2. Measuring the Value of Reliability Willingness to pay for reductions in the day-to-day variability of travel time is referred to as Value of Reliability (VOR). [Small, et al, 2005] presented an interesting and operational approach for actual estimation of the Value of Reliability (VOR) in a consistent way with VOT by splitting their impacts on traveler choice. The adopted quantitative measure of variability was the upper tail of the distribution of travel times, such as the difference between the 80th and 50th percentile travel times (see Figure 79).

209 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 25 30 35 40 45 50 55 60 Travel time, min ytili ba b or p evital u m u C Difference between the 80th and 50th percentile Figure 79: Travel Time Reliability Measure The authors argue that this measure is better than symmetric standard deviation, since in most situations being “late” is more crucial than being “earlier”, and many regular travelers will tend to leave build a “safety margin” into their departure times that will leave them an acceptably small chance of arriving late (i.e. planning for the 90th percentile travel time would mean arriving late 10 percent of the time). Reliability as defined above proved to be valued by travelers as highly as the median travel time. A number of recent toll-related SP experiments have used the 90th or 80th percentile travel time directly as an attribute. An example is a survey done in San Francisco to study possible area pricing policies (RSG, 2007), as shown in Figure 80.

210 Figure 80: Experiment Including Reliability as the 90th Percentile Travel Time Making this approach operational within the framework of travel forecasting models requires explicit modeling of travel time distributions, as well as making assumptions about how travelers acquire information about the uncertain situation they are about to experience. Dynamic traffic assignment and microsimulation tools are crucial for the application of models that include travel time variability, since static assignment can only predict average travel times. In general, the following types of reliability measures in Table 35 can be used in the model estimation and application. Note that supporting speed/travel time surveys are typically needed for the supply side of estimation. It is also important that there are several simplified ways to account for reliability by means of operational proxies, for example, perceived highway time differentiated by congestion levels. While the direct measures of reliability can be incorporated only in a framework of an Activity-Based microsimulation model, operational proxies can be incorporated in aggregate 4-step models.

211 Table 35: Reliability Measures Measure of reliability Demand side Supply side Estimation Application Derived reliability measures based on the observed or generated travel time distribution statistics Impact of percentile-based or other measure estimated along with average travel time and cost Repeated observations for the same trip and individual (RP, SP) Dynamic / learning framework or multiple network simulations Direct reliability measured based on explicit formulations like “probability of certain delay” Impact of probability of delay estimated along with average travel time and cost Direct question (SP or enhanced RP) Dynamic / learning framework or multiple network simulations Direct reliability proxies measured by the variation of travel times Impact of percentile-based or probability of delay estimated along with average travel time and cost Observed (RP), modeled (RP), or assumed (SP) variation of travel time. Modeled travel time variation as function of facility type, volume, etc (auxiliary regressions). Indirect observed or modeled reliability proxies like V/C (RP) or Perceived highway time differentiation by congestion levels V/C-based measures or perceived travel time (speed/delay/LOS-specific) estimated along with average travel time and cost Network skims for V/C and/or perceived time components Network skims for V/C and/or perceived time components A.2.4.3. Measuring the Choice of Departure Time as Affected by Pricing One response to pricing that is directly related to travel time reliability and has been very difficult to capture with either RP or SP methods is the shift in departure times in response to differences in prices and congestion across the day. Even in the most advanced activity-based models of activity scheduling, it has not been possible to measure the effect of travel times on departure time choice very accurately. A key reason for this is the fact that travel times are endogenous — the more people that choose to travel in a given period, the longer the travel times tend to be. A similar analytical problem may occur with dynamic pricing on managed lanes, where the price is also dependent on the demand (and vice versa). The Seattle Traffic Choices data, as well as RP data from existing priced facilities that use fixed pricing schedules by time of day and week (e.g., the various Orange County toll roads), could prove to be very valuable in this context. An important issue in modeling the effect of time of day pricing on travel demand is the fact that prices that influence one part of a travel tour may also indirectly influence the other trips in the tour as well. For example, when a commuter considers adjusting the time of travel to work in response to pricing, he or she may also consider that they need to spend a certain amount of time at work, and thus adjust the departure time for the trip returning home as well. [Hess, et al, 2007] report on the results of a series of CAPI

212 SP experiments carried out in the Netherlands and the UK, in which the SP choice screens explicitly captured this “knock on” effect by presenting the times and cost for both the trip to work and the trip back home. A.2.5.Typology of Available Surveys for Pricing Analysis and Modeling In the framework of a different research project – SHRP 2 (second Strategic Highway Research Program) 2 C04 (“Improving Our Understanding of How Highway Congestion and Pricing Affect Travel Demand”) – currently being implemented by the same team in parallel, we have identify available datasets collected in different regions that are useful to support pricing analysis and modeling. The following represent the major types of surveys that are used to support the analysis and modeling of transportation pricing and congestion, where the categories represent a combination of survey general methods and the purposes for which the surveys are done: • Type 1: General Comprehensive Household Interview Surveys - Revealed Preference (RP) • Type 2: Stated Preference (SP) Follow-on and Linked to Household Interview Surveys • Type 3: Managed Lane Studies - Combined Revealed Preference and Stated Preference (RP/SP) • Type 4: Regional Pricing Options and Area-Based Pricing - Combined Revealed Preference and Stated Preference (RP/SP) • Type 5: New Facilities with Tolling - Combined Revealed Preference and Stated Preference (RP/SP) • Type 6: Time of Day Tolling - Combined Revealed Preference and Stated Preference (RP/SP) • Type 7: Existing Facilities: Adding Tolls - Combined Revealed Preference and Stated Preference (RP/SP) Even in each of these categories, there are variations in their structure, scope, and design. In view of the objectives of the study, there are advantages and disadvantages to be carefully evaluated. No one single survey type can provide a full basis for a comprehensive analysis of all impacts of congestion and pricing on travel behavior. Altogether, however, we anticipate that the set of existing, as well as planned concurrent surveys, can provide good coverage of the main travel dimensions of interest, and good empirical foundation for this research project Household Interview Surveys represent the only holistic framework that allows for an analysis of the entire daily activity pattern of persons and households. There are several valuable Household Interview Surveys implemented in such over-congested regions as New York, 1996/7 and San Francisco 2000. They provide general observed patterns of behavior under congestion conditions. By comparison to other regions like Mid-Ohio or Atlanta for which comprehensive household interview surveys are also

213 available, the impact of congestion and pricing on activity patterns and lifestyle can be estimated. Household Interview Surveys with a subsequent SP follow up survey represent probably the most promising approach for data collection for the current study. This type of data collection was implemented in Seattle [Cambridge Systematics, et al 2007] and Montreal, is ongoing in Chicago and will be available in early 2008), and is planned for the new major data collection effort in NY (2008-2009). In this case the SP participants are recruited from the households that have already been surveyed. The SP design is built on the relevant reported trip that was implemented, for one or more of the priced facilities (either existing or planned). Thus, in the model estimation, not only are the characteristics of this particular trip available, but also the whole context of the person daily pattern is known The last characteristics, means that additional important situational variables are available, like the number and timing of the other trips and activities undertaken by the person on the same day. A wide range of more specific surveys focused on toll road users and relevant trips is available. Most of these focused surveys had a Combined Revealed Preference/Stated Preference form. The RP/SP combination ensures behavioral realism of the trip selected for the subsequent analysis as well as all associated household, person, and travel characteristics. The SP side, however opens a way to explore a wide range of responses to non-existing alternatives that can include priced and free highway facilities, improved transit, shifting the trip to a different time-of-day period, etc. Different methods of combined RP/SP estimation have been applied for models developed for SR-91, I-15, MnPass, and A-25 (in Montreal, QC). Some of the existing surveys have the form of a multi-day repeated observations and/or multi-wave panel like SR-91 and I-15 data sets processed and used by Small and Brownstone [Brownstone & Small, 2005; Small, Winston & Yan, 2005]. Repeated observations for the same trip of the same person can provide the basis for a direct measurement of reliability. Managed Lane Studies are extremely useful for understanding the trade-offs between travel time savings, reliability, and price since in this case managed lanes and free lanes always constitute two explicit and available route options with monitored level-of-service characteristics. The behavioral framework delimiting possible responses includes route and pre-route choice, and may also include time-of-day choice, occupancy choice (if relevant) as well as some other choices. A different emphasis and range of responses is pertinent to Regional Pricing Options and Area-Based Pricing Studies. In these studies, pricing is primarily considered as a measure of congestion relief by moving travelers from auto modes to transit (and not necessarily providing a free highway alternative like in most Area-Pricing schemes). These studies are especially beneficial for understanding and modeling the impact of congestion and pricing on mode choice. In this cases, travel time, reliability, and price trade-offs made by the travelers are inter-modal. A proper modeling of mode choice in the presence of congestion requires the development of reliability estimates not only for highway modes, but also for different transit modes. Model development and

214 application studies for area pricing in New York and San Francisco are now underway, both of which have clearly shown the importance of an inter-modal view on reliability and the differentiation of transit modes by reliability and other level-of-service characteristics [Resource System Group, 2007]. Some specific features are associated with studies of New Facilities with Tolling. Several such facilities have recently been proposed and/or built in the state of Texas, and some in the state of Florida as well. In general, Greenfield toll facilities are considered as the most complicated for modeling and predicting of traveler responses. The available RP/SP studies (and especially if both “before and after” statistics are available) are beneficial for a better understanding of the sources of patronage of the toll facility. In a certain sense, the entire demand for a new facility can be considered as “induced”, although trips may be diverted from other routes over a fairly wide geographical area. It is of particular importance, to understand and capture differential elasticities for route, mode, and time-of-day switches underlying the demand for new facility. Another important aspect of the dynamics of congestion and pricing is associated with Time of Day Tolling studies. These studies are normally associated with congestion pricing schemes and specifically address/target time-of-day choice (in combination with mode choice in urban areas). Pricing schemes with differential-by-time-of-day tolls, as well as real-time variable tolls, provide an ideal basis for understanding and modeling time-of-day related responses to congestion and pricing. A special case is provided by studies of Adding Tolls to Existing Facilities. There can be different facility types and regional frameworks that create different set of possible behavioral responses. There are however, several unique aspect associated with this type of surveys. They allow for capturing psychological effects like resistance to newly-introduced tolls, ramp-up period associated with certain learning and adaptation of the regional travelers to the new travel conditions created by the tolls, etc. Table 36 contains an inventory of survey datasets that have been collected over the past several years and which have supported or could support analysis of road pricing projects. The datasets have been divided into seven types, ranging from general purpose household interview surveys to those that have been purpose-designed to support specific road pricing applications. While this list is long, it is by no means exhaustive of all of the relevant work in these categories and is intended primarily to be illustrative of the types of efforts that have been undertaken to support these applications. One of the essential tasks of the SHRP project that has to be strongly coordinated with the course of the current NCHRP project is to select several available datasets for pilot studies and estimation of advanced models for pricing. Our intention is to select datasets from the same regions / studies where the NCHRP pilot studies are going to be implemented.

215 gniledoM dna sisylanA gnicirP daoR rof stesataD yevruS gnitsixE :63 elbaT

216

217

218

219 A.3. Travel Time Reliability Measures A.3.1.Highway Utility Components Highway travel utility is a basic expression combining various LOS attributes as perceived by the highway user. It is directly used in the highway trip route choice, for example between the Managed Lanes and General-Purpose Lanes on the same facility. It also constitutes an essential component in mode and time-of-day (TOD) choice utilities. The form of highway utility function is also important for modeling other (upper-level) travel choices since it serves as the basis for calculating accessibility measures. Consequently, it is essential to explore the highway travel utility and its components first, having in mind a simplified framework of route choice in the highway network, since the complexity builds up as additional choice dimensions are considered. In most travel demand models, including those developed for practical and research purposes, highway utility takes the following simple form: Equation 37: CbTaU ×+×= ; where: T = travel time, C = travel cost, 0<a = coefficient for travel time, 0<b = coefficient for travel cost, ba = Value of Time (VOT). Coefficients for travel time and cost normally take negative values, reflecting the fact that travel, in itself is an onerous function necessary only for visiting the activity location. Thus, the travel utility is frequently referred to as “disutility” of travel. There are some research works where the negative character of travel utility was questioned in some contexts. In particular, a positive travel utility was associated with long recreational trips on weekends [Stefan et al., 2007]. Also, an interesting effect was observed for commuting trips where commuters seem to prefer (or expect) to traveling for some minimum time (e.g., 20-30 minutes) and are not interested in reducing it below this threshold [Redmond and Mokhtarian, 2001]. The representation of highway travel utility as a linear function of two variables with constant coefficients is an extremely simplified one. A great deal of the SHRP 2 C04 project effort is devoted to elaboration of this basic form in the following ways: • Investigation of the highway user perception of travel time by congestion levels. This means that a simple generic coefficient for travel time could be replaced with the coefficients differentiated by congestion levels. • Inclusion and estimation of additional utility components of which travel time reliability has been identified as the most important component. Reliability is seen as an additional and non-duplicating term along with the average travel time and cost.

220 • Testing more complicated functional forms that are non-linear in time and cost, as well as accounting for randomly distributed coefficients or VOT (in addition to any explicit segmentation accounting for the observed user heterogeneity). With these enhancements, VOT is not assumed as a constant, but as a varying parameter depending on the absolute values of time and cost as well as reliability. As a working model we adopt the following general expression for the highway travel utility that will be explored component-by-component in the current research: Equation 38: ( )[ ] ( )[ ]∑ ∑∑ = == +×+×= 5 1 3 1 3 1k n nn m mmmkkk RcCbTaU φϕ , where: 1=k - uncongested highway travel time component, 2=k - congested highway travel time component (extra delay), 3=k - parking search time, 4=k - walk access/egress time (e.g. from the parking lot to trip destination, 5=k extra time associated with carpooling (picking-up/dropping/off passengers), kT = (average) travel time by component, 1=m - highway toll value, 2=m - parking cost, 3=m - vehicle maintenance and operating cost, mC = travel cost value by component, 1=n - disutility of time variation (1st measure of reliability), 2=n - schedule delay cost (2nd measure of reliability), 3=n - utility of (lost) activity participation (3rd measure of reliability), nR - reliability measures by component. nmk cba ,, -=- coefficients to be estimated, ( ) ( )...,... mk φϕ -=- functions for non-linear transformation of time and cost variables. This formulation makes it more difficult to calculate VOT although it is still possible in the same way that a Value of Reliability (VOR) can be calculated for the 1st type of reliability measure (assuming that this reliability measure is in min). VOR essentially represents travelers’ willingness to pay for reduction in travel time variability in the same way as VOT represents their willingness to pay for (average) travel time savings. More specifically, the VOT (in the context of willingness to pay tolls for saving time in congestion conditions) can be calculated by the following general formula: Equation 39: ( ) ( )( )111 222 1 2 12 , Cb Ta CU TUCTVOT φ ϕ ′ ′ = ∂∂ ∂∂ = . A similar calculation can be implemented for VOR. With non-linear transformation functions, VOT and VOR are no longer constant values. They now depend on the absolute values of time and cost variables at which the derivatives of the transformation functions are taken. The innovative components that relate to perceived highway time, travel time reliability, and non-linear transformations are discussed in the subsequent sections. It should be noted that some components, specifically perceived travel time and three reliability measures, might be

221 correlated statistically (and also conceptually duplicative at least to some extent). Thus, it is highly improbable that the entire formula (Equation 38) would ever be applied. It rather serves as a conceptual framework in which particular structures can be derived and tested statistically against each other. A.3.2.Perceived Highway Time Perceived transit time has been long recognized and used in travel models. For example, in most mode choice models and transit assignment algorithms, out-of-vehicle transit time components like wait and walk are weighted compared to in-vehicle travel time. It is not unusual to apply weights in the range of 2.0 - 4.0 reflecting that the travelers’ perception of out-vehicle time is different and it is perceived as more onerous compared to in-vehicle time. Contrary to the transit modeling practice, practically all travel models include a generic highway time term, i.e., the same coefficient is applied for each minute of highway time regardless of the travel conditions. However, there is some compelling statistical evidence that highway users perceive travel time differently by congestion levels. For example, driving in free-flow conditions might be very different from driving in heavily congested (stop-and-go) conditions. It is intuitive and behaviorally appealing that highway users driving in congested conditions might perceive the longer travel time as an additional delay or penalty on the top of free-flow (or some expected reasonable) time. In the segmentation of travel time coefficients by congestion levels, the time spent in congestion conditions is expected to have a larger disutility. A larger disutility associated with congestion would have at least two behavioral interpretations: • Negative psychological perception that is similar to the weight for walking to or waiting for transit service, • Simplified operational proxy for reliability that should be explored in combination with the explicit reliability measures. There are several research works reporting statistical evidence of quite high perceptional weights that highway users put on travel time in congested conditions [NCHRP Report 431, 1999; Axhausen et al., 2006; Levinson et al., 2004; MRC & PB, 2008; Wardman et al., 2008]. Also, there have been multiple indications in recent analyses of travel surveys that a perception of the time saved by respondents in the Revealed Preference (RP) survey, is about double the actual measured time saved [Small et al., 2005; Sullivan, 2000]. In the RP framework, this might well be a manifestation that travelers operate with perceived travel times, where time spent traveling through congested segments is psychologically doubled. Two examples of estimated perceptions of travel time are discussed below in order to illustrate the magnitude of the weights as well as possible approaches to differentiate travel time by congestion levels. It should be noted that in both cases the approaches are very simple on the supply side. The network simulation can be implemented, and required LOS skims can be generated by static assignment, though DTA could offer additional improvements. This technique can be easily applied with both ABMs and 4-step models. First example was documented in [NCHRP Report 431, 1999]. The travel time was broken into two parts: • Time in uncongested conditions (LOS A-D),

222 • Time in congested conditions (LOS E-F that is close to the “stop-and-go” condition). The choice framework included route choice only presented in the SP survey context. Travel time and cost variables were not estimated but stated in the SP questionnaires. Highway utility expression included total time, cost, and percentage of congested time. Using the previously introduced notation, the adopted utility specification can be written in the following way: Equation 40: ( ) 21 2 21 TT T cCbTTaU + ×+×++×= . This is different from the suggested formula (Equation 38), but can be transformed into an equivalent formula with certain assumptions (fixed total travel time). The estimation results confirmed a very high significance of the additional term of percentage of congested time. The authors translated it into a recommended mark-up value of 2.5 to VOT savings under congested conditions compared to uncongested conditions. More detailed estimation results are summarized in Table 37. By virtue of the specified utility function, the cost of shifting 1 min from uncongested to congested time is dependent on the total travel time. For an average time of 30 min, the VOT equivalent of the additional perceived burden associated with congestion itself is about $15/hour, which is roughly equal to the average commuting VOT applied in most models. Table 37: Cost of Shifting 1 Minute from Uncongested to Congested Time Total travel time, min Cost of shifting 1 min from uncongested to congested time, $ Equivalent in VOT $/hour 10 0.77 46.2 15 0.51 30.6 20 0.30 18.0 30 0.26 15.6 45 0.17 10.2 60 0.13 7.8 The second example is taken from the recently completed travel demand model for the Ottawa- Gatineau, Canada, region [MRC & PB, 2008]. The model framework, choice context, and utility formulation were different from those used in the [NCHRP Report 431, 1999] study. However, the bottom-line results look similar in many respects. In this study, a mode choice model was estimated for 5 travel purposes and 2 time-of-day periods (AM and PM) based on the RP data from the large household travel survey (5% of the population that corresponds to 23,870 households in the sample). Travel time and cost variables were provided from static assignment equilibrium skims from the modeled network. The highway utility included travel cost with one generic coefficient and travel time broken into the following two components (note that this breakdown of travel time is different from the one adopted for [NCHRP Report 431, 1999]): • Free-flow (minimal) time, • Extra delay, calculated as congested time minus free-flow time for the entire origin- destination path.

223 The highway utility function had the following form: Equation 41: ( )∑ ×+×+×+×= s ss hdCbTaTaU 2211 , where: s = additional mode-specific constants and household/zonal variables, sh = values of additional variables, sd = estimated coefficients. The estimation results are shown in Table 38, as translated into VOT terms. They confirm that for several segments, specifically AM and PM work trips, as well as PM discretionary trips, each minute of congestion delay is perceived as about twice as onerous as the free-flow (minimal) time component. For other segments, however, statistical tests did not show a significant difference between free-flow and congestion time components, thus two coefficients were pooled together. Table 38: VOT Estimates for Free-Flow Time and Congestion Delay Trip purpose VOT, $/hour AM PM Free-flow time Congestion delay Free-flow time Congestion delay Work 22.2 42.7 19.4 40.0 University 10.0 10.0 11.0 11.0 School 5.1 5.1 5.1 5.1 Maintenance 10.7 10.7 12.1 12.1 Discretionary 9.0 9.0 11.4 29.3 The third example is taken from the research work of Mark Wardman et al, 2008 where they provided new evidence on the variation in the valuation of motorists’ travel time savings across a finer gradation of time types, than has been hitherto attempted (6 different levels of congestion), by means of analyzing SP data collected from different tolled road context in the UK and US. The summary of the time relativities is presented in Table 39. The study confirms that a reasonable value for the perceived time weight in congested conditions lies in the range 1.3 to 2.0. Table 39: Highway Time Weight by Congestion Levels Travel time conditions UK US Free Flow 1.00 1.00 Busy 1.05 1.03 Light Congestion 1.11 1.06 Heavy Congestion 1.31 1.20 Stop Start 1.20 1.38 Gridlock 1.89 1.79

224 A.3.3.Reliability Approach 1: Time Variability Measures Time variability can be measured by any compact measure associated with travel time distribution (for example any combination of the mean, dispersion, and higher moments). Taking into account such considerations as behavioral realism and simplicity of the model estimation (specifically, formulation of SP alternatives), as well as application, three main forms have been proposed and tested so far (see ITS, 2008 for a good discussion): • Standard Deviation, that is a symmetric measure assuming that being early or late is equally undesirable (probably not a realistic assumption for many trips and underlying activities). • Difference between 80th, 90th, or 95th and 50th percentile travel times that is frequently referred to as buffer time. This is an asymmetric and more behaviorally appealing measure since it specifically targets late arrivals and is less sensitive to early arrivals. • Simplified asymmetric measures in terms of probability of certain delays; delay thresholds such as 15 or 30 min are frequently used in the SP framework. An illustrative example of the Standard Deviation approach is provided in [NCHRP Report 431, 1999] in the context of binary route choice. The following form of utility function was adopted: Equation 42: ( )TSDcCbTaU ×+×+×= , where: )(TSD = standard deviation of travel time. Standard deviation of travel time was calculated based on the set of 5 travel times presented in the SP questionnaire for each highway route alternative. The estimation results showed that highway users assign a very high value on each minute of standard deviation that is comparable with or even higher than the VOT associated with average travel time itself (i.e., ac ≥ ). Also a certain logical variation across trip purposes and income groups was captured as summarized in Table 40 (for one of the several reported model specifications). Table 40: Value of Reliability measured as Standard deviation of Time Trip purpose and income group Value of Reliability $ per min SD $ per hour SD Work trips, higher income 0.258 15.5 Work trips, lower income 0.215 12.9 Non-work trips, higher income 0.210 12.6 Non-work trips, lower income 0.167 10.0 A good example of the second time variability measure was presented in [Small, et al., 2005]. The adopted quantitative measure of variability was the upper tail of the distribution of travel times, such as the difference between the 80th and 50th percentile travel times (see Figure 81). The authors argue that this measure is better than a symmetric standard deviation, since in most situations, being “late” is more crucial than being “early”, and many regular travelers will

225 tend to build a “safety margin” into their departure times that will leave them an acceptably small chance of arriving late (i.e., planning for the 80th percentile travel time would mean arriving late for only 20% of the trips). 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 25 30 35 40 45 50 55 60 Travel time, min C u m u la ti ve p ro b ab ili ty Difference between the 80th and 50th percentile Figure 81: Travel Time Variability Measure The choice context included binary route choice between the Managed (tolled) Lanes and General Purpose (free) lanes on the section of SR-91 in Orange County, CA. The survey included actual users of the facility and the model was estimated on the mix of RP and SP data. The variation of travel times and tolls was significantly enriched by combining RP data from actual choices with SP data from hypothetical situations that were aligned with the pricing experiment. Distribution of travel times was calculated based on the independently observed data. The measures were obtained from field measurements on SR-91 taken at many times of day, on 11 different days. It was assumed that this distribution was known to the travelers based on their past experience. The utility function was specified by the following formula: Equation 43: ( )TRcCbTaU ×+×+×= , where: ( )TR = difference between the 80th and 50th percentile. Reliability, as defined above, proved to be valued by travelers as highly as the median travel time (VOT was roughly equal to VOR, i.e., ca ≈ ). This particular model form, with the condition of equal VOT and VOR, has a very interesting and intuitive interpretation (that itself could be used for a model formulation in a slightly simplified form where it is assumed from the outset that ca = ). Indeed, if we assume that willingness to pay for saving 1 min of average travel time (the 50th percentile) is equal to willingness to pay for 1 min of reduction of the

226 difference between the 80th and 50th percentile, then we can combine both terms in the highway utility function since they have the same coefficient. This means that the underlying decision-making variable is the travel time value at the 80th percentile. This variable essentially combines both average travel time and time variation measure. An example in Table 41 illustrates this possible approach. In the example, we assume that the highway user has to choose between two roads for commuting that are characterized by different time distributions. Road 1 is longer but more reliable – the travel time varies from 41 min to 50 min. Road 2 is shorter but travel time is less predictable and varies from 29 min to 52 min. We assume that the highway user is familiar with both roads and makes his/her choice based on a rational consideration of the known distributions. In practical terms, this can be interpreted as a recollection of at least 10 trips on each road in the past, sorted by travel times from the best to worst. Table 41: Illustration of Reliability Impact Percentile Travel time, min Preference Road 1 Road 2 10 41 29 20 42 30 30 43 35 40 44 39 50 45 40 Road 2 by conventional approach 60 46 41 70 47 45 80 48 50 Road 1 by suggested approach 90 49 51 100 50 52 Although Road 2 has a better (lower) average travel time and would be preferred in most conventional modeling procedures, Road 1 has a better 80th percentile measure. In reality, the user would probably prefer Road 1 as the more reliable service. This choice framework with a single measure can be used as a simplified version of the approach. Rather than estimating two separate terms (average travel time and additional time associated with 80th-50th percentile, a single measure of 80th (or any other percentile large than 50th if yields a better statistical fit) could be used. For example, in a similar context, a 90th percentile measure was used in [Brownstone & Small, 2005 ]. This framework is based on a plausible assumption that travelers under congestion conditions, characterized by travel time uncertainty, behave as rational risk-minimizers. They do not base their decisions on the average values. However, they do not adopt the extreme mini-max approach (minimize risk and choose according to the worst possible case) either. The decision point probably lies somewhere between the 80th and 90th percentiles. It is important to note that making this approach operational within the framework of regional travel models requires explicit deriving these measures from simulation of travel time distributions, as well as adopting assumptions regarding the ways in which travelers acquire information about the uncertain situation they are about to experience. DTA and traffic micro-

227 simulation tools are crucial for the application of models that include explicit travel time variability, since static assignment can only predict average travel times. Other approaches for measuring variability of travel time can also be considered. They are similar to the approach described above in conceptual terms, but use a different technique in both the estimation and the application stages. For example, in the travel model developed by for the [Toll T&R Study in Montreal, 2002], probability of delays longer than 15 and 30 min was introduced in the SP questionnaires for trucks. The subsequent estimation of the choice model revealed a very high significance of this variable comparable with the total trip time (in line with the VOR estimation of Small, et al., 2005). Application of this model required special probability-of-delay skims that were calculated based on the observed statistics of delays as a function of the modeled Volume-over-Capacity (V/C) ratio. Although this technique requires a multi-day survey of travel times and speeds, it can be applied in combination with the static assignment method. Many regions with continuous traffic monitoring equipment now have such data available for important highway segments. A problem yet to be resolved, however, is that when calculating the travel time reliability measure over the entire origin-destination path, the highway links cannot be considered independent. Reliability is closely intertwined with VOT. In RP models, if variability is not measured explicitly and included as a variable, this omission will tend to inflate the estimated value of average time savings. In reality, variability in travel time tends to be correlated with the mean travel time, and people are paying for changes in both variables, so omitting one will tend to attribute the total effect to the other. Consequently, an important use SP data sets that include reliability, is to use them in combination with RP data sets for which good objective estimates of travel time variability can be derived. It should be mentioned that the direct using of travel time variability in the behavioral modeling framework is not the most appealing approach, compared to the other two (discussed below). The principal conceptual drawback of this approach is that it does not explicitly consider the nature of underlying activities and mechanisms that create the disutility. Needless to say, the largest part of disutility associated with unreliable travel time is being late (or too early) at the activity location, and consequently, losing some part of the planned activity participation. The practical advantage of the time variability approach, however, is in its relative simplicity and exclusive reliance on the data supplied by the transportation networks. A.3.4.Reliability Approach 2: Schedule Delay Cost This approach has been widely accepted by the research community since its inception [Small, 1982 ]. According to this approach, the impact of travel time (un)reliability is measured by explicit cost associated with the delayed or early arrival at the activity location. This approach considers a single trip at a time and assumes that the preferred arrival time that corresponds to zero schedule cost is known. The essence of the approach is that the trip cost (i.e., disutility) can be calculated as a combination of the following three components: α = value of travel time and cost, β = cost of arriving earlier than the preferred schedule, γ = cost of arriving later than the preferred schedule.

228 By definition, only one of the schedule costs can have a non-zero value in each particular case depending on the actual arrival time versus the preferred one. There can be many analytical forms for the schedule cost as a function of the actual time difference (delay or early arrival). It is logical to assume that both functions should be monotonically increasing with respect to the time difference. It is also expected, in most cases, that the schedule delay function should be steeper than the early arrival function for most activities (being late is more onerous than being earlier). The details, however, depend on the activity type, person characteristics, and situational context. The most frequently used forms include simple linear function (i.e., constant schedule delay cost per minute), non-linear convex function (assuming that large delays are associated with growing cost per minute), and various piece-wise functions accounting for fixed cost associated with any delay along with a variable cost per minute – see Figure 82. Preferred arrival Cost, $ Late arrival, minEarly arrival, min LinearLinear w/fixed Non-linear Figure 82: Schedule Delay Cost An example of a schedule delay model estimated in a highway route choice context with a specially designed SP survey is given in [NCHRP Report 431, 1999]. The utility function was specified in the following way: Equation 44: ( ) ( ) ( )ttTSDcCbTaU ∆+∆+×+×+×= γβ , where: t∆ = difference between actual and preferred arrival time, ( )t∆β = early arrival cost specified as a non-linear convex function, ( )t∆γ = late arrival cost specified as a linear function with a fixed penalty.

229 The estimation results with respect to the schedule delay cost are summarized in Table 42 (for one of the tested model specifications). Interestingly, as reported by the authors, in the presence of explicit schedule delay cost, the travel time variability measure (standard deviation) lost its significance. The authors concluded that in models with a fully specified set of schedule costs, it is unnecessary to include the additional cost of unreliability of travel time. Table 42: Estimation of Schedule Delay Cost Component Marginal values, $ Early arrival (non-linear): - by 5 min 0.028/min - by 10 min 0.078/min - by 15 min 0.128/min Late arrival dummy: - work trips 2.87 - non work trips 1.80 Late arrival (linear) 0.310/min Extra late arrival dummy 0.98 Schedule delay cost should be distinguished from TOD choice and the associated disutility of shifting the planned (preferred) trip departure/arrival time, although in practical estimation analysis, the data might mix these two factors. To clearly distinguish between the planned schedule and schedule delay, the person should explicitly report actual and preferred arrival time for each trip. Schedule delay cost assumes that the person has planned a certain schedule, but in the implementation process on the given day the delay occurs to disturb this plan. TOD choice relates to the stage of schedule planning. The outcome of this process is the preferred arrival time. Comparing schedule delay to time variability as two different measures of time reliability, it should be noted that the schedule delay approach provides a better behavioral insight than travel time variability. It explicitly states the reasons and attempts to quantify the factors of the disutility associated with unreliable travel time, specifically perceived penalties associated with not being at the activity location in time. The schedule delay approach, however, has its own theoretical limitations as identified by the following: • The approach is applied separately for each trip made by a person during the day and it is assumed that the schedule delay cost for each subsequent trip is independent of the previous trip. Technically this approach is based on a fixed departure time and a preferred arrival time for each trip. In general, this is not a realistic assumption, since the activity duration requirements would create a dependence of the departure time for the next trip on the arrival time for the previous trip. • This approach does not consider activity participation explicitly, though it makes a step towards such a consideration compared to the travel time variability approach. • If applied for the evaluation of user benefits from travel time savings, this approach must incorporate TOD choice, i.e., travelers’ reconsideration of departure time in response to the

230 changed congestion. Otherwise, travel time savings can result in early arrival penalties overweighting the value of saved travel time. On the practical side, in order to be implementable, the schedule delay approach imposes several requirements that are not easy to meet, especially with the conventional RP surveys: • For each trip, in addition to the actual arrival time, the preferred arrival time should be identified. While the preferred arrival time is generally known to the traveler (or perceived subconsciously), it is generally not observed by the modeler in RP type of data. To explore this phenomenon and estimate models that address it, the SP framework proved to be very effective, since the preferred arrival time and schedule delays can be stated in the design of alternatives. In some research, simplified assumptions about the preferred arrival time were adopted. For example, in [Tseng & Verhoef, 2008 ], the preferred arrival time was calculated as a weighted average between the actual departure time and would-be arrival time under free-flow traffic conditions. • Application of this model for forecasting would again require input in the form of preferred arrival times. This can be accomplished either by means of external specification of the usual schedules on the activity-supply side (that would probably be possible for work and fixed non-work activities), or by means of a planned schedule model on the demand side. The latter would generate individual schedule plans (departure times) based on the optimal activity durations conditional upon the average travel times. The subsequent simulation (plan implementation) model would incorporate schedule delay cost based on the simulated travel times. A.3.5.Reliability Approach 3: Lost Utility of Activity Participation The third approach is based on a concept of time-dependent utility profile by activity type [Supernak, 1992; Kitamura & Supernak, 1997 ]. Recently this approach was adopted in several research works on DTA formulation integrated with activity scheduling analysis [Kim at al., 2006; Lam & Yin, 2001]. The essence of this approach is that each individual has a certain temporal utility profile for each activity that is characterized by function U(t). The utility profile can either be estimated as a parametric or a non-parametric function of time and time can be modeled in either continuous or discrete form. The utility profile represents an instant utility of participation in the activity at the given point of time (or during the discrete time unit that starts at the given point of time). The total utility of participation in the activity can be calculated by integrating the utility profile from the arrival time (τ ) to departure time (pi ): Equation 45: ( ) ( )∫= pi τ piτ dttUU , . Simple utility profiles are independent of the activity duration. In this case, it is assumed that the marginal utility of each activity at each point of time is independent of the time already spent on this activity. This might be too simplifying an assumption, at least for certain activity types like household maintenance needs where the activity loses its value after the errands have been completed. More complicated utility profiles can be specified as two-dimensional functions U(t,d) where d denotes the activity duration until moment t. In this case, the total utility of activity participation can be written as

231 Equation 46: ( ) ( )∫ −= pi τ τpiτ dtttUU ,, . Hypothetical, but typical temporal utility profiles specified in a discrete space with an hourly resolution are shown in Figure 83. The work activity profile is adjusted to reflect the fixed schedule requirements (higher utility to be present at 8.00 AM and 5:00 PM points). The shopping activity profile is much more uniform, with an additionally assumed convenience to undertake this activity after usual work hours. 0 20 40 60 80 100 6 8 10 12 14 16 18 20 22 Time of day Work Shopping Figure 83: Example of Utility Profiles The concept of utility profiles is instrumental in understanding how individuals construct their daily activity schedules. According to this concept, each individual maximizes a total daily utility of activity participation. If we consider a predetermined sequence of activity episodes, it can be said that individuals switch from activity to activity when the time profile of the second activity exceeds the time profile of the previous activity. Travel episodes are placed between activity episodes in such a way that the whole individual daily schedule represents a continuous sequence of time intervals as shown in Figure 84.

232 0 24 Activity i=0 Activity i=1 Activity i=2 Trip i=1 Trip i=2 Trip i=3 Activity i=3 Departure Arrival Duration Travel id iT ipi iτ Schedule{ }ipiθ = Figure 84: Consistent Individual Daily Schedule The effect of unreliability of travel times can be directly measured by comparison of the planned and actual total daily utility of the schedule that includes all activity and travel episodes. For simplicity, but without essential loss of generality, we assume that the sequence of activity episodes and trip departure times are fixed. We will also assume that travel time delay never exceeds the planned duration of the subsequent activity; thus, activities cannot be cancelled as a result of unreliable travel time. Thus, unreliability affects only travel times and arrival times. In this context, the reliability measure can be expressed as the loss of activity participation in the following way: Equation 47: ( )∑ −= i A i P i UUL , where: L = total user loss (disutility) over the whole schedule, P iU = utility of the trip and subsequent activity with preferred arrival time, A iU = utility of the trip and subsequent activity with actual arrival time, where the planned and actual utilities can be written as follows: Equation 48: ( ) ( )∫++×+×= 1i P i dttUCbTaU i P i P i P i P i pi τ τ , Equation 49: ( ) ( )∫++×+×= 1i A i dttUCbTaU i A i A i A i A i pi τ τ ,

233 where: Equation 50: i P i P iT piτ −= ; i A i A iT piτ −= . By substituting expression (Equation 50) into formulas (Equation 48) and (Equation 49), and then, substituting formulas (Equation 48) and (Equation 49) into the basic expression (Equation 47) we obtain: Equation 51: ( ) ( ) ( )∑ ∫+−×+−×= i i A i P i A i P i A i P i dttUCCbaL τ τ ττ , where the last term (integral) represents the loss of activity participation, while the first two terms represent extra travel time and cost. A logical relationship between temporal activity profiles of utilities and schedule delay cost was explored by [Tseng & Verhoef, 2008 ] that led to an insightful general framework. It can be shown that these two approaches are not independent. The schedule delay cost functions can always be consistently derived from the temporal utility profiles; thus, the schedule delay approach can be thought of as a particular transformation of the temporal utility profile approach. Interestingly, the opposite is true, i.e., temporal utility profiles could be fully restored from the schedule delay cost functions only under some specific assumptions. To illustrate the relationship between temporal utility profile and schedule delay cost, consider two adjacent activities in the daily schedule with a trip between them as shown in Figure 85. In this fragment of the daily schedule, we assume that the temporal utility profile of the 1st activity is monotonically decreasing, while the 2nd one is monotonically increasing. We also number the trip as the 2nd one (according to the activity at trip destination), to be consistent with the natural numbering shown in Figure 84. With an (ideal) zero trip time between the activities, the rational individual would switch from the 1st activity to the 2nd activity at the intercept point to ensure a maximum total utility. With a non-zero trip time, the optimal strategy would be to depart at such time that the departure-time utility of the first activity would be equal to the arrival-time utility of the second activity.

234 1210 7 6 5 2 ( )t1β ( )tU2 ( )tU1 ( )t1γ 2pi 2τ 2T 2T 2T 12t2pi 2pi 2τ2τ Early arrival Late departure Optimal departure and arrival 1 3 4 8 9 11 Figure 85: Example of Two Adjacent Activities We can distinguish between three possible cases as shown in the figure: 2122 τpi ≤≤ t = optimal departure before the intercept point and arrival after it, 1222 t<< τpi = arrival earlier than the intercept point, 2212 τpi <<t = departure later than the intercept point. It is natural to specify schedule delay cost function in such a way that it should be equal to zero when the travel time is equal zero and the trip is perfectly timed at the intercept point. It is also natural to refer to the intercept point as the ideal preferred arrival time. In the general case, with non-zero travel time and a not perfectly timed trip, there are two ways to constructively derive the total trip cost from the temporal utility functions with the cost components interpreted as schedule delay cost. In both ways we calculate trip cost as a sum of the following three components: Equation 52: ( ) ( ) ( ) ( )222222222222 ,,,, τpiγτpiβτpiατpi ++=C , where: ( )222 ,τpiα = travel cost, ( )222 ,τpiβ = cost of arriving earlier, ( )222 ,τpiγ = cost of departing/arriving late. The first way is to derive trip cost components as follows:

235 Equation 53: ( ) ( ) ( )[ ]∫∫ == 2 2 2 2 )(,max, 122222 τ pi τ pi ατpiα dttUtUdtt , (loss of maximum activity utility when traveling), Equation 54: ( ) ( ) ( )[ ]∫∫ −== 12 2 12 2 )(, 212222 tt dttUtUdtt ττ βτpiβ , (non-optimal 2nd activity if arrived early 122 t<τ ), Equation 55: ( ) ( ) ( )[ ]∫∫ −== 2 12 2 12 )(, 122222 pipi γτpiγ tt dttUtUdtt , (non-optimal 1st activity if departed late 122 t>pi ). The second way uses a different structural allocation of the same total cost: Equation 56: ( ) ( ) ∫∫ == 2 2 2 2 )(, 12222 τ pi τ pi ατpiα dttUdtt , (loss of 1st activity utility when traveling), Equation 57: ( ) ( ) ( )[ ]∫∫ −== 12 2 12 2 )(, 212222 tt dttUtUdtt ττ βτpiβ , (non-optimal 2nd activity if arrived early 122 t<τ ), Equation 58: ( ) ( ) ( )[ ]∫∫ −== 2 12 2 12 )(, 122222 ττ γτpiγ tt dttUtUdtt , (loss of added 2nd activity in travel and late departure 122 t>τ ). To verify that both ways produces the same total cost and also highlight the differences between them, we summarize all components in Table 43. Also, all cost components are related to the areas 1-12 of integration under the temporal utility curves shown in Figure 85.

236 Table 43: Trip Cost Components Case Component Areas of Integration in Figure 85 1st way 2nd way 2122 τpi ≤≤ t : departure earlier the intercept and arrival later the intercept ( )222 ,τpiα 5,6,7,8 5,6,8 ( )222 ,τpiβ ( )222 ,τpiγ 7 1222 t<< τpi : arrival earlier than the intercept ( )222 ,τpiα 1,2 1,2 ( )222 ,τpiβ 3,5 3,5 ( )222 ,τpiγ 2212 τpi <<t : departure later than the intercept ( )222 ,τpiα 11,12 12 ( )222 ,τpiβ ( )222 ,τpiγ 7,9 7,9,11 In either way of derivation, the schedule delay cost is associated with functions that represent a difference between the temporal utility profiles. The cost of early arrival is associated with the extra utility of the first activity (when it is higher than the utility of second activity). In the same vein, the cost of late arrival is associated with the extra utility of the second activity (when it is higher than the utility of first activity). In other words, schedule-related cost corresponds to participation in non-optimal activity because of the not-optimally-timed trip. This was formalized in the expressions (Equation 54, Equation 55, Equation 57, Equation 58) in the following straightforward way: Equation 59: ( ) ( ) ( )tUtUt 212 −=β , Equation 60: ( ) ( ) ( )tUtUt 122 −=γ . The only difference between the two methods of derivation is in the formulation of the travel cost function and the area of integration for the schedule delay cost for a late arrival. The first way is probably more natural and appealing. In this case, travel cost is associated with the lost (maximum) utility of activity participation when traveling, while the schedule-related cost components are associated with participation in non-optimal activity. However, it operates with both departure and arrival times. Regrouping the cost in the second way allows for expression of both schedule-related cost components in terms of arrival time only. The essence of the second approach is that the extra utility of the second activity over the first activity at the time of traveling (areas 7 and 11 in Figure 85) is transferred from the travel cost component to the schedule delay (late arrival) component. In the second method, the travel cost component might not look behaviorally intuitive since it is associated with the utility of first activity only. However, it should be mentioned that activity utilities are set in an arbitrary scale and only the difference between them is important. Essentially, one of the activities could be chosen as a reference point with zero utility. Thus, the difference between the two approaches is purely definitional. The second method of derivation is more convenient to operate with schedule delay functions depending on the arrival time only. Additionally, the difference between the two approaches is only important when

237 schedule delay cost components are analytically derived from the estimated temporal utilities by formulas (Equation 53-Equation 55) and (Equation 56-Equation 58). If the schedule delay cost components as specified by formula (Equation 52) are estimated directly, the difference is irrelevant since the same explanatory variables can enter any component. However, if the schedule-related cost functions are estimated based on the arrival time only, the second approach would still be more consistent with this method of specifying the schedule delay cost function. With the assumptions on the form of the temporal utility functions, as shown in Figure 85 for a case of two adjacent activities in a fixed order, and with a known intercept (preferred arrival time), it is also possible to restore temporal utility profiles from the known travel cost and schedule delay functions in the following way: Equation 61: ( ) ( )ttU α=1 , Equation 62: ( ) ( ) ( ) ( ) ( ) ( ) > = < + − = 12 12 12 22 2 22 2 for for for , , tt tt tt tt t tt tU γα α βα . Thus, for a simple case under the assumptions explained above, there is no essential difference between the schedule-delay-cost approach and temporal-utility-profile approach. They just represent different ways of grouping the same utility/cost components. The direct analogy does not hold however, when more than two activities are considered (and not necessarily in a fixed order) or when the underlying utility profiles are more complicated and the preferred arrival times cannot be established for each trip (pair of adjacent activities) independently. In this case, utility profiles still provide a comprehensive framework for calculation of the loss of activity participation, while schedule delay cost components are bound to a particular order of activities and trips with predetermined preferred arrival time. With certain additional simplifying assumptions the analogy between the schedule-delay-cost approach and temporal-utility-profile approach can remain valid for multiple activities/trips. Consider a situation where the sequence of activities is fixed and the daily schedule can be broken into fragments where only two activities are feasible with the preferred arrival time defined for each fragment. For example, if we have three activities in the daily pattern “home- work-home” with two trips between them, the first fragment would include (following the numbering convention in Figure 84) the 0th and 1st activities (home and work) and the second fragment would include the 1st and 2nd activities (work and home). The first fragment would include the outbound work commuting leg, while the second fragment would include the inbound work commuting leg. Then, schedule delay cost can be derived from the utility profiles independently for each trip within each correspondent fragment by formulas (Equation 57 and Equation 58). Also, the utility profiles can be restored from the schedule delay cost of the 1st trip, for the 0th and 1st activities and from the schedule delay cost of the 2nd trip, for the 1st and 2nd activities by formulas (Equation 61 and Equation 62). Then, if needed, the utility profiles in one of the fragments can be shifted to ensure continuity of the entire utility profile for the 2nd activity (work) across both fragments. This technique can be applied recursively to any number of

238 activities. It is however, extremely “fragile” and fails if one of the simplifying assumption does not hold. Thus, the concept of temporal utility profiles, that considers travel time unreliability effects as the loss of activity-participation utility, is probably the most holistic among the three possible approaches outlined above. It offers more complete behavioral insight than the travel time variability and schedule delay approaches. It also provides a better platform for the calculation of User Benefits from travel time savings and reliability improvements, including small and discontinuous savings. This concept, however, also has limitations. On the theoretical side, it is based on a very strong assumption that the temporal utility profiles can be measured independently for each activity, and, as a result, the daily schedule utility represents just the sum of them. In reality, the utility of one activity can be a strong function of participation and duration of the other activities. This is quite obvious with several episodes of the same or similar activity types. There are multiple effects related to saturation, satiation, and time-space/budget constraints that make the utility profiles interdependent across activity episodes. From this perspective, a microeconomic framework that distinguishes between direct and indirect utility functions holds promise. However, such a framework has not yet been operationalized in travel demand modeling. For practical applications, this approach requires estimation of the temporal utility profiles on the demand side. This is a realistic task using econometric methods, although it might result in quite complicated structures and would require a large (household type) survey. Conceivably, application of such a model would require an explicit modeling of a planned daily schedule for each individual, taking into account expected average travel times with a subsequent network simulation, and calculation of the utility loss because of the actual travel times that are different from the expected travel times.

239 A.4. Advanced Time-of-Day Models with Enhanced Temporal Resolution The model of this type that is the found in current practice was first estimated and applied as part of the Columbus ABM [Vovsha & Bradley, 2005 ]. Since then, the approach has been employed for other ABMs in Atlanta, San-Francisco Bay Area, Denver, Sacramento, San-Diego, and Phoenix. The model is essentially a discrete choice construct that operates with tour departure-from-home and arrival-back-home time combinations as alternatives. The proposed utility structure based on “continuous shift” variables, represents an analytical hybrid that combines the advantages of a discrete choice structure (flexible in specification and easy to estimate and apply) with advantages of a duration model (parsimonious structure with a few parameters that support any level of temporal resolution including continuous time). The hybrid model originally applied in Columbus had a temporal resolution of 1 hour that is expressed in 190 hour-by-hour departure-arrival time alternatives. The subsequent modifications in Atlanta and Denver used a finer temporal resolution of 30 min that can be achieved with only minor complications. The model is applied sequentially for all tours in the individual Daily Activity Pattern (DAP) according to the predetermined priority of each activity type. The enhanced temporal resolution allows for applying direct availability rules for each subsequently scheduled tour to be placed in the residual time window left after scheduling the tours of higher priority. This conditionality ensures a full consistency for the individual entire-day activity and travel schedule as an outcome of the model. This formulation for the variables is not restrictive since most of the household, person, and zonal characteristics in the TOD model are naturally generic across time alternatives. However, network level-of-service variables vary by time-of-day, and are specified as alternative-specific (based on the departure and arrival time of each alternative). Using generic coefficients and variables associated with either departure period, arrival period, or duration, creates a compact structure of the choice model where the number of alternatives can be arbitrarily large (depending on the chosen time unit scale) but the number of coefficients to estimate is limited to a reasonable number. Duration variables can be interpreted as “continuous shift” factors that parameterize the termination rate in such a way that if the coefficient multiplied by the variable is positive, it means that the termination rate is getting lower and the whole distribution is shifted to the longer durations. Negative values work in the opposite direction, collapsing the distribution towards shorter durations. For a practical implementation of the proposed model, the utility functions for all (multiple) alternatives should be specified in a parsimonious way. In the ABM structure, the tour- scheduling model is placed after the destination choice and before mode choice. Thus, the destination of the tour and all related destination and origin-destination attributes are assumed known and can be used as variables in the model estimation. The choice alternatives are formulated as tour departure from home-arrival at home hour combinations ( hg, ), while mode choice log-sums and bias constants are related to multi-hour departure-arrival periods ( ts, ). Tour duration is calculated as the difference between the arrival

240 and departure hours ( gh − ), and incorporates both the activity duration and travel time to and from the main tour activity including intermediate stops. The tour TOD choice utility has the following general form: Equation 63: ( ) ( )+++= ∑− m mhtgsghhggh VDVVV ,,lnµ , where: hg VV , = departure and arrival time specific components, ghD − = duration-specific components, m = entire-tour modes, stmV = tour mode utility by mode m, leaving home in period s and returning home in period t, µ = mode choice Logsum coefficient. For model estimation the following practical rules can be used to set the alternative departure- arrival time combinations: • Each reported/modeled departure/arrival time is rounded to the nearest half-hour. So, the hour “17” includes all times from 16:45 (4:45 PM) to 17:14 (5:14 PM). • Any times before 5 (5 AM) are shifted to 5, and any times after 23 (11 PM) are shifted to 23. This results in a shift for typically relatively few cases, and limits the number of half- hours in the model to 38. • Every possible combination of the 38 departure half-hours with the 38 arrival half-hours where the arrival half-hour is the same or later than the departure hour is an alternative. This gives 38 × 39 / 2 = 741 choice alternatives. To specify the model as parsimoniously as possible, departure/arrival constants are only applied for seven TOD periods that can be specified, for example, as follows: • 5 to 6 (early morning), • 6 to 9 (AM peak), • 9 to 12 (early midday), • 12 to 15 (late midday), • 15 to 19 (PM peak), • 19 to 21 (evening), • 21 to 23 (late night). The network simulations to obtain travel time and cost skims can be implemented for even broader periods, for example: • AM peak, • Early midday, • Late midday, • PM peak, • Night (including early morning, evening, and late night).

241 The mode-choice log-sums will be used for all relevant combinations of the five time periods above. This structure, however, is only limited by the number of traffic and transit assignments implemented at each global iteration. It could include more TOD periods for network simulation with ultimately approaching a resolution of dynamic traffic assignment. In particular, peak hour 7-8 AM can be singled out of the AM period and distinguished from the AM shoulders (6-7, 8-9). This would lead to a network simulation system with six TOD periods, which is manageable. The predetermined hierarchy of tours by travel purpose and activity setting (individual/joint) is assumed in the scheduling procedure. This hierarchy is based on the general principle on which the most activity-based tour-based models are built. According to this principle, people first make decisions regarding their mandatory activities (work/university/school). Then, conditional upon scheduling the mandatory activities, they schedule joint non-mandatory activities – maintenance and discretionary – of which maintenance (shopping, escorting other persons, and various other household maintenance activities) is generally considered of higher priority compared to discretionary activities (leisure and eating-out). Finally, having scheduled mandatory and joint activities, each household member schedules individual activities within the residual time window remaining after making any mandatory and joint tours. When a person undertakes several activities (tours) of the same priority in the course of the day, those tours are prioritized in a chronological order, i.e. the earlier tour is scheduled first, while the later tour is scheduled next conditional upon the departure/arrival time combination of the first tour, and also forcing the second tour to be scheduled after the first tour (even if there is an available residual window before the first tour). By using the rules described above, all tours of each surveyed individual can be unambiguously ordered by scheduling priority. The residual time window and set of available TOD alternatives are defined for each subsequent tour conditional upon scheduling of the previously processed tours.

242 A.5. Explicit Modeling of Joint Travel An explicit modeling of joint travel constitutes one of the primary advantages of the ABM paradigm. In the basic ABM structure, only joint travel for non-mandatory activities is modeled explicitly in the form of fully joint tours (where all members of the travel party travel together from the very beginning to the end and participate in the same activity). This typically accounts for more than 50% of joint travel. Other types of joint travel like carpooling of workers and escorting children can also be considered as optional extensions. Each fully-joint tour is considered as a unit of modeling with a group-wise decision making regarding the primary destination, mode, frequency and location of stops etc. Formally, modeling joint activities involves two linked stages: • Generation stage that generates the number of joint tours by purpose/activity type made by the entire household. This is the Joint Tour Frequency Model. • Participation stage at which the decision whether to participate or not in each joint tour is made for each household member and tour. This is the Joint Tour Participation Model. For analytical convenience this model is broken into two sub-models: 1) travel party composition, and 2) person participation choice. A.5.1.Household Generation of Joint Tours For this sub-model, the number of travel purposes is limited to 4-5 non-mandatory activities (shopping, maintenance, discretionary, eating-out, visiting relatives and friends) and the observed maximum total number of fully joint tours implemented by a household during a regular workday is limited to 2-3. A simultaneous frequency-choice model can be formulated that would cover all possible frequencies and purpose combinations [Vovsha et al, 2003 ]. A structure adopted in the Columbus model (and subsequently applied in the Atlanta, San Francisco Bay Area, San Diego, and Phoenix ABMs with minor modifications) included 5 purposes and maximum of 2 joint tours that resulted in 21 alternatives – see Figure 86.

243 Frequency of fully-joint tours generated by shared activity 1 tour 2 toursNo tours Sh o pp in g Ea tin g o u t M a in te n a n ce Vi si tin g Sh op / sh op Sh op / e a ti Sh op / m ai n Sh op / v is i Ea ti / d is c M a in / e at i Ea ti / v is i M a in / m ai n M ai n / v is i Ea ti / E at i Travel party composition for shared activity Adults Children Adults+Children By purpose Person participation in shared activity Yes No By purpose & travel party X X D is cr et io n ar y Sh op / d is c M a in / d is c Vi si / V is i Vi si / D is c Di sc / D is c Figure 86: Model Structure for Joint Non-Mandatory Activity Application experience of this sequential structure (tour frequency, party composition by tours, person participation by persons) in Columbus has shown that is performs well in practical terms. A.5.2. Travel Party Composition Travel party composition is defined in terms of person categories participating in each tour (adults and children). It results in a trinary choice model with the following alternatives as shown in Figure 86 above: • Adult party (including adult household members only), • Children’s party (including household children only), • Mixed party where at least one adult and at least one child participate. The statistical analysis and model estimation has shown a strong linkage between trip purpose and typical party compositions [Vovsha, et al, 2003; MORPC Final Report, 2005]. The essence of the joint party composition model is to narrow down the set of possible person participation choices modeled by the subsequent sub-model. A.5.3. Joint Tour Participation at Person Level Frequency choice and travel party composition models discussed above generally fall quite readily into the standard discrete choice structure. Regarding the person participation model, two alternative ways to formulate the choice model have been found. The complexity of the person participation model stems from the combinatorial variety of households (especially relatively large households with, say, two workers and four children), as well as from the

244 necessity to link participation models across household members and tours in order to logically limit the participation of each household member in joint tours. Consider a realistic example of a household having two types of persons - two adult workers and two school children of pre-driving age. Consider a joint tour with the chosen mixed travel party. For simplicity of presentation also assume that only one adult is enough to form the party. The first approach constitutes entire-party choice. This approach is based on explicitly listing all possible person combinations for the travel party formation. Then, the following party- formation tree can be depicted – see Figure 87 (left side): Worker/Children Carpool 1st Worker 2nd Worker 1st Child 2nd Child Both Children One Child Two Children 1st Child 2nd Child Both Children One Child Two Children Worker/Children Carpool 1st Worker 2nd Worker 1st Child 2nd Child Restart Yes No Yes No Yes No Yes No Restart 2nd Child Yes No Entire-party choice Binary person participation Figure 87: Travel Party-Formation Trees In this case, six travel parties can be formed in the household. The following problematic features of the first approach have can be identified: • The total number of alternatives in the choice set may reach hundreds if more dimensions are added to the person segmentation and/or a larger household is considered, • The alternatives have a differential degree of similarity, thus a complicated nested structure should be applied; however, it is not clear how to organize all nesting levels in view of the multiple possible dimensions. The second approach is based on participation choice being modeled for each person sequentially. In this alternative approach, only a binary choice model is calibrated for each activity, party composition and person type. Quantitatively different alternatives by party size are not distinguished. Thus, using the previous example, there will be two different utilities for each worker (assuming that male and female differences are important for this joint travel category) and one utility for the child. Then a sequence of binary choices is applied assuming a single possible participation for each person – see Figure 87 (right side).

245 The following problematic features of the second approach are seen: • Participation probabilities might not be independent (some particular person types or household members may tend to cooperate more), thus, this fine effect would be lost, • There is an uncontrolled party size, including a non-zero probability for failure to form a party if all persons have chosen not to participate. Comparing pros and cons of both approaches, we have found that the second one is more practical and operational in both model estimation and application. This approach makes travel party size automatically linked to the household size and composition. For example, the more children in the household, the more likely a bigger travel party will occur for the relevant joint travel where children are in the party composition. The case of a failure to form a travel party in model application is resolved by re-starting the Monte-Carlo simulation until the suitable travel party has been formed. This version of the model was included in the Columbus, Atlanta, and San Francisco Bay Area Models.

246 A.6. Evaluation of Pricing Projects: Example Application To show how Cost Benefit Analysis (CBA) techniques can be applied and highlight welfare calculations, an example application is provided. While the example is on a relatively small scale (addition of a single link in a toy network), it provides useful insights. A.6.1.Methodology The methods used here consider multiple alternatives for travel between a single origin and two destinations. The alternatives include the choice of destination, mode (auto, bus, or walk), time-of-day (AM peak, PM peak, and off-peak), and route. Figure 88 details the layout of this idealized network. Figure 88: Idealized Network Using a nested logit specification [Ben-Akiva and Lerman, 1985 ], so that clusters of similar options exhibit correlated error terms, and making some assumptions about cost and time sensitivity, as well as scale terms and nesting (inclusive value) parameters, one can estimate flows for each alternative. There are four distinct choice dimensions being modeled here, so the nesting structure exhibits three embedded nests. At the lowest level is route choice, followed by time-of-day, mode, and destination choices at the higher levels (Figure 89). Reasonable Origin Destination A Destination B Local Trip Modes: Walk, Bus, Auto 3 TOD Periods Distance = 1 mi Route 2 Modes: Bus, Auto 3 TOD Periods Distance = 8 mi Route 1 Modes: Auto 3 TOD Periods Distance = 8 mi

247 behavioral parameter values were selected to characterize preferences. Figure 89 shows the overall nesting structure of the model, and the associated scale parameters. Figure 89: Nested Logit Model Structure and Parameters Two destination options (A and B) are available for each user. Destination A represents a location close to the origin (1 mile), while destination B lies much farther away (8 miles). However, the assumed attractiveness of destination A is much less than that of B (10 versus 200 − much like a local versus regional activity center). Further, the free-flow speed to A via automobile is only 10 mph, as compared to 60 mph to B. The two routes to destination B (existing and new) are assumed to be identical in their physical characteristics, and the Bureau of Public Roads (BPR) link performance function (Equation 1) was used to compute travel times as a function of free-flow times, capacities, and volumes, with alpha (α) and beta (β) parameters of 0.85 and 5.5, respectively (as suggested by Martin and McGuckin, 1998 ): Equation 64: ( )( ), 1l free l l lt t v c βα= + , A M µ=1.6 µ=1.6 µ=1.6 µ=1.4 µ=1.2 O ff-P eak PM D estin atio n A W alk B u s A uto A M Off-P eak PM AM Off-P eak PM AM µ=1.6 µ=1.6 µ=1.4 O ff-P eak PM D estin atio n B B u s A uto A M Off-P eak PM R o ute 2 µ=1.8 R o ute 1 R o ute 2 µ=1.8 R o ute 1 R o ute 2 µ=1.8 R o ute 1

248 where tl is travel time on link l, tfree,l is free-flow travel time on link l¸vl is demand for link l, and cl is link l’s capacity flow volume. For destination A, capacity is assumed to be unlimited, which is reasonable when such trips use a local street network with multiple paths (and relatively low demand, as compared to supply). In the second level of the nest, three mode alternatives are available, though the walk mode is only available to destination A. Walk speed is assumed to be 4.47 mph, and bus speed is assumed to be the same as the auto mode (In the case of travel to destination B, buses are assumed to travel on route 2 − the tolled route, though bus passengers do not pay the toll.). However, a flat 15 min penalty is added to bus times to represent its added wait, access, and egress times. Furthermore, the bus fare is set at $0.50 per trip, buses on the network are assumed to be equivalent to 2.0 passenger cars (as suggested by the Highway Capacity Manual [TRB 2001]), and buses are assumed to ride “full”, at 20 persons of capacity. For the auto mode, a fixed operating cost of $0.20/mile is assumed. Last, in calculation of utilities for each alternative, alternative specific constants (ASCs) are assumed for each mode: 0.0 for auto, -1.1 for bus, and -1.3 for walk. These values were selected to represent reasonable preference structures and are simply for illustrative purposes. The last two levels of the nesting structure are for time-of-day (TOD) and route choices (though choice of route is only available to those driving to destination B). Three TOD alternatives are available and link capacities to destination B are assumed to vary by the number of hours in each time period (which assumes uniform assessment of all traffic within each period). AM peak is assumed to last 3 hours (6-9 am) and PM peak is assumed to last 4 hours (3-7 pm). Instead of giving the off-peak (OP) period the remaining 17 hours of the day, it is assumed that most OP travel will occur between the AM and PM peaks; thus, the OP period lasts 6 hrs. If 2,000 passenger cars per hour per lane (pcphpl) is assumed for freeway capacity, and each route to B has two lanes, then capacities on both routes are the same for each TOD: 12,000 passenger cars equivalents in the AM, 24,000 in the OP, and 16,000 in the PM. In computing utilities, ASCs for each TOD alternative are assumed to be 0.0, -0.3, and 0.2 for the AM peak, OP, and PM peak, to reflect relative preference for travel during the PM and then AM periods, respectively. Several other assumptions are needed here as well. The total number of system users is assumed to be 125,000, segmented into two groups. Low value of travel time (VOTT) users make up half of the population (with a $6/hour VOTT), and high VOTT users make up the other half (with a $12/hour VOTT). Finally, it is important to discuss the scale parameters (also known as inclusive value coefficients) in each level of the nested model. While scaling parameters need not be the same for two different nests at the same level in the nesting structure, all were assumed to be the same here for simplicity. For example, the scale parameters across TODs for walk mode to A are assumed to be the same as the scaling parameter across TODs for bus mode to B. Consistent with McFadden’s random-utility theory the scale parameters for the route choice, TOD, mode and destination choice nests were assumed to be 1.8, 1.6, 1.4, and 1.2, respectively. In contrast to most nested logit specifications (where the top level nest enjoys a 1.0 scale factor), the top level scaling parameter is assumed here to be 1.2. The reason for this is that the coefficient on cost in the utility equations is set equal to -1 (as will be shown below). In this way all top-level utility values are in terms of dollars already. An equivalent formulation emerges when setting the top-

249 level scaling parameter to 1.0 (as is customary) and adjusting other parameters accordingly. Such formulations require subsequent conversion of utility values to dollars, however. As shown in Figure 89 with a scale parameter (µ1) of 1.8 in the lowest nest (driving to destination B via route 1 or route 2), 1.6 (µ2) in the next lowest nest (AM versus PM versus OP TOD), 1.4 (µ3) in the second highest level nest (walk versus bus versus auto), and 1.2 (µ4) in the upper level nest (destination A versus destination B), equilibrium destination, mode, TOD, and route shares, and travel times and tolls were estimated for a variety of pricing scenarios. The associated equations, for generalized trip costs, systematic utilities, inclusive values (scaled Logsums) of the nested choices and choice probabilities are as follows: Equation 65: dmprdmprdmpridmpri OCtVOTTGC ++⋅= τ, Equation 66: [ ] dmpripmBddmpri GCASCASCAttrAttrV ,, )ln()ln( −++−= Equation 67: ( ) ( )][ 2,,11,,1 1 , expexpln1 routedmpiroutedmpidmpi VV µµµ +=Γ Equation 68: ( ) ( ) ( )][ PMdmiOPdmiAMdmidmi ,,2,,2,,2 2 , expexpexpln1 Γ+Γ+Γ=Γ µµµ µ Equation 69: ( ) ( ) ( )][ AutodiBusdiWalkdidi ,,3,,3,,3 3 , expexpexpln1 Γ+Γ+Γ=Γ µµµ µ Equation 70: ( ) ( )∑ ∈ Γ Γ = Dl li di di ,4 ,4 , exp exp Pr µ µ Equation 71: ( ) ( )∑ ∈ Γ Γ = Ml dli dmi didmi ,3 ,3 ,, exp exp PrPr µ µ Equation 72: ( ) ( )∑ ∈ Γ Γ = Pl dmli dmpi dmidmpi ,2 ,2 ,, exp exp PrPr µ µ Equation 73: ( ) ( )∑ ∈ Γ Γ = Rl dmpli dmpri dmpidmpri ,1 ,1 ,, exp exp PrPr µ µ Here GC is the generalized cost, V stands for systematic utility of the alternative (as measured in dollars), Γ denotes the inclusive value or expected maximum utility for an upper level alternative, Pr(⋅) represents the probability of a particular choice, i denotes user group (either low or high VOT), d stands for the destination of interest (either A or B), m represents the mode of interest (walk, bus, or auto), p denotes the TOD (AM, OP, or PM), r is the route (either 1 or 2), D is the set of destination alternatives, M is the set of mode alternatives, P is the set of TOD alternatives, and R is the set of route alternatives. Here, VOT denotes the value of travel time for the associated traveler group, µ1, µ2, µ3, and µ4 serve as the scaling parameters for the

250 route, TOD, mode, and destination nests, respectively, τ represents out-of-pocket charges (for toll or bus fare) and has no coefficient (so that utilities are in dollars), OC is the out-of-pocket operating expenses (set to zero for bus and walk modes), t denotes the travel time, ASC represents the alternative specific constants for mode and TOD alternatives, and Attr is the attractiveness value of each destination. Two routes exist only if auto mode and destination B are chosen. In the other cases, route 2 can simply be assumed to have some arbitrarily large disutility (or travel cost) associated with it such that route 2 is not chosen. Since utility is unobserved, forcing the cost coefficient to equal one necessitates the use of two (non-unitary) scale factors (one for each nest). This offers greater transparency in dimensioning, but is in some contrast to most NL specifications (where µ is set equal to 1 in the upper [or lower] nest). Estimates of the Consumer Surplus (CS) of each tolled scenario were computed as well. In general, the CS can be measured between any two scenarios, but we will look primarily at the CS measured in reference to the base scenario − where only one of the two routes to destination B is available. In other words, the base scenario is a “do nothing” scenario where no new roads are built to destination B. The CS computation is as follows: Equation 74: ( ) ( )Γ−Γ= ∑∑ ∈∈ Dk ki Dk kiiCS 0 ,4 1 ,4 4 explnexpln1 µµ µ Equations from Equation 65 through Equation 74 were applied for both traveler types, recognizing the distinctive values of time for each. A.6.2. Application Results An assortment of tolled and non-tolled scenarios was investigated. Each scenario was run to find equilibrium travel times and tolls on all network links. A base scenario is developed so that only one of the two routes to destination B exists. In addition, another non-tolled scenario is constructed such that both routes to destination B exist, but neither is tolled (i.e., build a new road without tolls). Six distinctive tolled scenarios were also considered, for a total of eight scenarios (These scenarios are in no way exhaustive and simply serve to illustrate key policy cases.). The simplest of these involve the building of a new road with a flat toll assessed (both $0.05 per mile and $0.10 per mile tolls are considered here). Optimal toll levels were sought, to maximize expected net benefits, across all 125,000 travelers (relative to the non-tolled scenario). Moreover, this scenario was extended to the case where optimal toll levels are assessed on both routes to destination B. Finally, revenue maximizing tolls were considered on the new route as well as on both routes to destination B. Unfortunately, throughput maximization cannot be undertaken here since a maximum flow is not defined by the BPR function. The BPR function suggests that demand equals flow, and, since demand is unbounded to the right (i.e., demand can grow toward positive infinity), flow is also unbounded to the right. The results of these applications emerge from relatively straightforward network equilibration and optimization procedures, and are discussed below.

251 A.6.2.1. Traveler Choices and Network Effects Under the above assumptions, equilibrium base conditions (where only one route to destination B exists) result in volume–to-capacity (V/C) ratios for peak and off-peak periods of 1.08 (for both AM and PM peaks) and 0.98, respectively, to destination B. This results in 19 minute and 15 minute peak and off-peak travel times to destination B, which are quite high relative to its 8- minute free-flow travel time. Of course, what is of interest is how this compares to scenarios in which a second route (to destination B) is added (essentially doubling corridor capacity). In each case of an added route to destination B, substantial delay reductions emerge. When the new route is not tolled, V/C ratios, and thus travel times, to B are lower. Travel times are just under 14 minutes in the peak periods and 10 minutes in the off-peak period (Table 1), saving travelers about 5 minutes per trip in all TODs. If a flat toll of $0.40 (equivalent to 5¢ per mile) is assessed on the new route to destination B, lower V/C ratios are experienced on the tolled route (in comparison to the non-tolled case); and V/C ratios are higher on the non-tolled route (as compared to the non-tolled case), but lower than the base (no-build) scenario. If a flat toll of $0.80 (equivalent to 10¢ per mile) is assessed on the new route to B, similar results emerge, but with more significant differences. Thus, in comparison to the non-tolled scenario, travel times to destination B in these two tolled scenarios fall by about 2 minutes, to 3.5 minutes per trip in peak periods (for the 5¢ and 10¢ per mile settings, as shown in Table 44), however, traffic shifts to the non-tolled route, where travel times rise.

252 Table 44: Travel Times, Tolls, V/C Ratios, and VMT across Scenarios Parameter Link TOD Base (1 Link to Dest. B) Build 2nd Link (No Toll) delloT skniL htoB delloT 2 kniL 5 cent/mi Toll 10 cent/mi Toll Welfare Maximizing Toll Revenue Maximizing Toll Welfare Maximizing Toll Revenue Maximizing Toll Travel Time (min) Link 1 AM 19.12 13.69 14.26 14.94 14.43 15.55 9.40 8.44 MID 14.20 9.69 10.11 10.67 10.16 11.14 8.70 8.02 PM 19.26 13.81 14.39 15.07 14.56 15.68 9.43 8.45 Link 2 AM N/A 13.72 11.85 10.28 9.44 8.52 9.20 8.77 MID N/A 9.71 8.51 8.06 9.00 8.03 8.67 8.03 PM N/A 13.84 11.97 10.38 9.45 8.54 9.21 8.81 Toll ($) Link 1 AM $0.00 $0.00 $0.00 $0.00 $0.00 $0.00 $1.13 $1.70 MID $0.00 $0.00 $0.00 $0.00 $0.00 $0.00 $0.58 $1.36 PM $0.00 $0.00 $0.00 $0.00 $0.00 $0.00 $1.14 $1.71 Link 2 AM N/A $0.00 $0.40 $0.80 $0.89 $1.33 $1.20 $1.63 MID N/A $0.00 $0.40 $0.80 $0.26 $0.99 $0.60 $1.36 PM N/A $0.00 $0.40 $0.80 $0.91 $1.35 $1.21 $1.64 V/C Ratio Link 1 AM N/A 0.97 0.99 1.00 0.99 1.02 0.75 0.61 MID N/A 0.78 0.81 0.84 0.81 0.87 0.66 0.36 PM N/A 0.97 0.99 1.01 0.99 1.02 0.75 0.61 Link 2 AM 1.08 0.96 0.89 0.80 0.73 0.60 0.70 0.62 MID 0.98 0.77 0.62 0.42 0.70 0.35 0.65 0.36 PM 1.09 0.96 0.89 0.81 0.73 0.60 0.70 0.63 VMT (1,000 veh- mi/day) All Auto Links AM 125 195 190 185 176 168 152 136 MID 201 303 280 249 297 242 259 148 PM 168 261 255 248 235 225 204 183 Total 494 759 726 682 708 636 615 467

253 In the case of welfare maximizing tolls, two scenarios were investigated: one where only the new route to destination B is tolled and one where both the new and old routes to destination B are tolled. I(Note: Welfare maximizing tolls refer to toll levels that result in the maximum social welfare, which includes traveler perceived costs and benefits, along with generated revenues.) In the case of one tolled route, the welfare- maximizing toll on that route (at system equilibrium) is found to be $0.89 in the AM peak, $0.91 in the PM peak, and $0.26 in the off-peak period (Table 44). Both peak periods’ optimal tolls are higher than the flat tolls considered above, while the off-peak period toll is somewhat less, since it is less attractive to travelers (and thus was assigned a negative ASC). These tolls result in travel times on the new route that are almost the same for peak and off-peak periods (just 9 minutes during the off-peak and about 9.5 minutes during the two peaks), in clear contrast to the flat tolls discussed above. If welfare maximizing tolls are charged on both routes to destination B, tolls rise (to about $1.10 to $1.20 in the peak periods and about $0.60 in the off-peak period, as shown in Table 44). While tolls may be high, travelers enjoy significant travel time benefits when driving to destination B. No travelers destined for B experience more than a 9.5 minute travel time. If road managers instead wish to maximize revenue on the new route, optimal tolls will be $1.33 in the AM peak, $0.99 in the off-peak, and $1.35 in the PM peak (Table 44). If one maximizes revenues by tolling both routes to destination B, the lowest travel times emerge, since fewer travelers choose destination B, due to tolls on the order of $1.60 to $1.70 in the peak periods and $1.35 in the off-peak period (as shown in Table 44), much higher than in any of the other scenarios. These higher tolls result in a substantial VMT reduction. In fact, maximizing revenue on both routes is the only scenario in which VMT drops relative to the base (No Build) scenario (5.4% less). All other scenarios exhibit a substantial increase in VMT relative to the base, ranging from a 24.6% increase (in the case of welfare maximizing tolls on both routes) to a 53.4% increase when neither route to destination B is tolled. A.6.2.2. Welfare Results Equation 74 specifies equivalent variation or average traveler welfare change as measured relative to the one-route (to destination B) base scenario, in units of dollars per traveler. A positive welfare change means that users benefit (on average) from the policy, whereas a negative welfare change indicates user losses. In addition to traveler welfare impacts, revenues resulting from each tolling scenario must be considered here. Table 45 presents the predicted traveler welfare change and revenue streams for each of the scenarios previously discussed.

254 Table 45: Revenues and Welfare Results by Scenario Measure VOT Build 2nd Link (No Toll) One Link Tolled Both Links Tolled 5 cent/mi Toll 10 cent/mi Toll Maximum Welfare Toll Maximum Revenue Toll Maximum Welfare Toll Maximum Revenue Toll Welfare Change from Base ($/traveler/day) Low VOT $0.63 $0.54 $0.45 $0.52 $0.39 $0.24 $0.00 High VOT $0.69 $0.63 $0.53 $0.63 $0.46 $0.48 $0.24 Average $0.66 $0.58 $0.49 $0.57 $0.42 $0.36 $0.12 Daily Welfare Change ($/day) Low VOT $39.6K $33.9K $28.4K $32.3K $24.3K $15.2K $0.13K High VOT $42.9K $39.1K $33.3K $39.1K $28.4K $30.2K $15.1K Total User Benefit ($/day) $82.5K $73.0K $61.7K $71.4K $52.7K $45.4K $15.3K Toll Revenue ($/day) $0 $15.9K $26.0K $23.0K $30.9K $65.9K $81.1K Net Welfare (User Benefit plus Revenue in $/day) $82.5K $88.9K $87.7K $94.4K $83.6K $111.3K $96.4K As shown in Table 45, generated revenues range from $0 in the Build, No Toll route scenario to $81,000 per day in the Revenue Maximizing, Both Routes Tolled scenario. In discussing traveler welfare, it is not so surprising that in all of the scenarios with the new route to B, welfare change estimates are positive (even for the Revenue Maximizing, Both Routes Tolled scenario where VMT falls), meaning net benefits exist for all travelers. This is due to the highly congested conditions of the one-route base scenario, and the simple result that doubling capacity to destination B allows for great congestion relief. The greatest welfare improvements for travelers emerge in the no-toll scenario ($0.63 and $0.69 per traveler per day for low- and high-VOT travelers, respectively, and $82,500 total per day). However, when toll revenues are considered in addition to traveler welfare, this no-toll scenario offers the lowest net welfare overall. Even when tolls are set to maximize revenues on one or both routes, net welfare is greater than the no-toll scenario. These welfare benefits are useful to highlight for all stakeholders. Of course, the greatest net benefits emerge when all “goods” are priced optimally − so that tolls are set to maximize welfare on both routes (net welfare of $111,300 per day). If a no-toll route to destination B must be provided, the best option emerges from the welfare maximizing scenario with a single route tolled (net welfare of $94,400 per day). Clearly, there are benefits for both low-and high-VOT travelers and high-VOT travelers benefit more, but the disparity between the two traveler types is larger when both routes are tolled (differences of $6,800 in traveler benefits with one route tolled and $15,000 with both tolled). A similar result is found when tolls are set to maximize revenues. The difference between low- and high-VOT travelers when one route is tolled is $4,100 per day ($24,300 versus $28,400) while the difference when both routes are tolled is $15,000 per day ($100 versus $15,100), again supporting the notion that the impacts are more evenly distributed when one route is left non-tolled. In fact, if equity is measured as the difference in welfare between low- and high-VOT travelers, the least equitable scenario occurs with welfare maximizing tolls on both routes (though the equity is almost the same for the revenue maximizing tolls on both routes). The availability of substitute travel options may be essential in maximizing user benefits

255 under tolling (and other) policies while wooing supporters across all demographic classes. Of course, such welfare calculations do not account for the costs of construction of the new facility nor do they account for the operation of tolling technology needed for scenarios with tolling implemented. As a point of comparison, Rule-of-Half (RoH) results can also be computed − relative to the base scenario and relative to the Build, No Toll route scenario. It is important to recall that, in general, one cannot use the RoH when new alternatives are added, since the price associated with zero demand for the new alternative in the base scenario (i.e., where the demand curve intercepts the price axis) is unknown. Thus, price changes cannot be measured. However, this idealized example (with perfect route substitution) allows one to assume that link capacity to destination B is simply doubled (instead of an entirely new link being added to the network). Alternatively, one can view the situation as one where both links are present in the base case, neither tolled, and the capacity on the new/second link is negligible, so no travelers use that link until it is expanded. Either way, however, the RoH approach (relative to the No Build scenario) neglects the fact that a new alternative is being added New alternative convey a variety of unobserved benefits in individual utility perceptions, offering subtle but often substantial benefits. Under this approach, the RoH can be used to approximate welfare changes relative to the No Build or base scenario. When the Build, No Toll scenario is used as the base, the RoH can be used in the standard fashion. Table 46 shows the results of the RoH analysis in terms of daily welfare change from the base. Welfare changes resulting from the logsum analysis are also provided. When the single non-tolled route scenario is considered the base, the RoH calculation produces very different results than the logsum approach. In each scenario, the RoH estimates are lower than logsum estimates; they range from about 33% lower (for maximum- revenue tolls on a single link) to over 100% lower (for maximum revenue tolls on both links, where welfare estimates become negative under the RoH). This is due mostly to the fact that the addition of a new alternative provides the opportunity for a new choice, with a random utility component (the Gumbel error term, reflecting unobserved factors). Thus, even if the added alternative did not appear to offer generalized travel cost benefits, it would still offer benefits. As mentioned earlier (under the RoH discussion), such benefits are neglected in the RoH framework. In addition, RoH estimates perform best for policies resulting in small travel cost and time changes. Here, the new link offers extensive congestion relief to the corridor, so the RoH’s linear-demand assumption is problematic. However, when the Build, No Toll route scenario serves as the base case, the RoH estimates lie very close to the Logsum-based estimates, differing by no more than 2.4%.

256 Table 46: Traveler Welfare Changes Using RoH versus Using Logsum Measure by VOT Build 2nd Link (No Toll) One Link Tolled Both Links Tolled 5 cent/mi Toll 10 cent/mi Toll Maximum Welfare Toll Maximum Revenue Toll Maximum Welfare Toll Maximum Revenue Toll Single Non-Tolled Link Scenario as Base RoH Approach Low VOT $23.2K $19.9K $17.0K $19.8K $15.8K $2.3K -$10.3K High VOT $31.0K $27.7K $23.2K $27.7K $19.4K $20.1K $7.0K Total $54.3K $47.6K $40.2K $47.5K $35.2K $22.4K -$3.4K Logsum Approach Low VOT $39.6K $33.9K $28.4K $32.3K $24.3K $15.2K $0.1K High VOT $42.9K $39.1K $33.3K $39.1K $28.4K $30.2K $15.1K Total $82.5K $73.0K $61.7K $71.5K $52.7K $45.4K $15.2K Build 2nd Link (No Toll) Scenario as Base RoH Approach Low VOT N/A -$5.7K -$11.3K -$7.3K -$15.6K -$24.0K -$38.1K High VOT N/A -$3.8K -$9.5K -$3.7K -$14.3K -$12.5K -$27.5K Total N/A -$9.5K -$20.8K -$11.0K -$29.9K -$36.5K -$65.6K Logsum Approach Low VOT N/A -$5.7K -$11.2K -$7.3K -$15.3K -$24.4K -$39.5K High VOT N/A -$3.8K -$9.5K -$3.7K -$14.5K -$12.7K -$27.8K Total N/A -$9.5K -$20.7K -$11.0K -$29.8K -$37.1K -$67.2K Note: Revenues are not added to these estimates of traveler welfare changes A.6.2.3. Accounting for Highway Cost In order to more fully evaluate the scenarios as investment alternatives, it is necessary to recognize the costs associated with building and operating a new roadway. Litman’s, 2006 review of the literature suggests that freeways in urban areas cost on the order of $5 million to $10 million per lane-mile, which includes land acquisition, pavement, and intersection reconstruction. (Note that the costs of building a new road in a non-urban area would be substantially lower). Assuming the cost is $5 million per lane-mile, an 8- mile, 2-lane freeway will cost $80 million. If one also assumes that routine annual maintenance costs of highways are $14,000 per lane-mile (assumed from a range from $13,100 to $14,600 as suggested by FDOT, 2003 ) and toll road management costs are $50,000 per lane-mile per year, a single toll facility will cost $1.02 million per year, and two toll facilities will cost $1.82 million per year. Total operating expenses and number of toll road lane-miles for NTTA (NTTA, 2003), New Jersey Turnpike Authority (NJTA, 2003), and San Joaquin Hills Transportation Corridor Agency [SJHTCA, 2003] were used to find average management costs for toll roads. All three were on the order of $100,000 per lane-mile, but these systems are mature, and rely on past technology. With new, paperless systems, management costs of $50,000 per lane-mile were assumed to be reasonable here. Finally, if it is assumed that calculated revenues are for weekdays only and weekend days generate only half that of weekdays, daily revenues can be multiplied by 313 to find yearly revenue streams in each scenario. Given these assumptions, it is possible to perform a cost-benefit analysis of traveler welfare, system expenditures and toll revenues. (For purposes of policymaking, more comprehensive analysis may also be pursued, including estimation of bus service subsidies, emissions effects, and crash costs.)

257 The analysis of financing the new road is performed in two ways. First, it is assumed that all toll revenues go toward the construction and management costs of the new road, after discounting future revenues at 5% per year. Note that a rate of 7% may be more appropriate for the facility investigated here (as per OMB, 2003 suggestions), but this example is for illustrative purposes only. In the scenario where the new road is built without tolls, there are no revenues, but one can still compute costs and traveler benefits, which results in a net benefit of about $0.52 per traveler per day (one-way), or $20.4 million per year (if costs are financed via a 5-percent 30-year loan. Table 47 presents the results of this first step of the analysis, including total and net annual revenues (after covering construction loan costs and toll road management), and time period it takes to fully recover construction and management costs (assuming an annualized payback). When the new road is built but not tolled, a repayment period clearly cannot be computed (since there are no toll revenues), and in the case of a flat 5¢/mile toll, toll revenues are not enough to cover all costs when future revenues are discounted at 5%. In each of the other scenarios, the repayment period is rather modest (about 20 years or less), with the minimum payback duration (less than 4 years) resulting from tolling of both routes. Of course, once the costs of building, maintaining, and managing the new road have been recovered, future revenues can go toward any number of things, including credits to travelers, other infrastructure improvements, or the improvement of transit services. Table 47: Repayment Period Results for New Road Investment Measure Build 2nd Link (No Toll) One Link Tolled Both Links Tolled 5 cent/mi Toll 10 cent/mi Toll Maximum Welfare Toll Maximum Revenue Toll Maximum Welfare Toll Maximum Revenue Toll Toll Revenues1 $0 $4.98M $8.15M $7.18M $9.68M $20.62M $25.38M Maint. Cost 2 $0.224M $0.224M $0.224M $0.224M $0.224M $0.224M $0.224M Manage. Cost 3 $0 $0.8M $0.8M $0.8M $0.8M $1.6M $1.6M Net Revenue 4 $0 $3.96M $7.12M $6.16M $8.66M $18.80M $23.56M Repay. Time 5 N/A N/A 16.9 yrs 21.5 yrs 12.7 yrs 4.9 yrs 3.8 yrs 1Revenue generated for a year assumes 261 weekdays and 104 weekend days per year, where weekend-day revenues are one half those of regular weekdays. Values are shown in millions of $/year. 2These are roadway maintenance costs for the new highway in millions of $/year. 3These are tollway management costs in millions of $/year. 4Net revenue is the difference between total revenue and the sum of maintenance and toll management costs, shown in millions of $/year. 5Repayment time is the time (in years) it takes to pay off an $80 million loan using all of the net revenues generated by the scenario. Here, a discount rate for future revenues is assumed to be 5%. In the second step of the analysis, a more standard approach to CBA is taken where costs and benefits that accrue over time are discounted to find equivalent NPVs. First, it is assumed that the construction of the new road will be paid for by a 30-year loan with 5% interest rate and fixed yearly payments. This amounts to annual payments of approximately $5.2 million (not including maintenance and management costs, which are subtracted from net revenues before applying them to loan payments). In addition,

258 other costs include the annual maintenance and management costs, and benefits include yearly toll revenues and traveler benefits. Once all annual costs and benefits are computed, the NPV of each can be found (by discounting at the assumed rate of 5% per year) and summed to determine a project’s total NPV. Table 48 shows the results of this analysis. Note that the NPV of all costs includes only construction costs and the NPV of all benefits includes all other benefits and costs per FHWA, 2003 guidance in computing B-C ratios. As shown in Table 48, each scenario enjoys very high NPV values (due to heavy congestion in the base scenario) ranging from $306.6 million in the case of revenue-maximizing tolls on a single route to $427.4 million in the case of welfare-maximizing tolls on both routes. Since the NPV of costs is the same for each alternative, the scenario rankings based on total NPV and B-C ratio are identical. Table 48: Cost-Benefit Analysis Results for New Road Investment Measure Build 2nd Link (No Toll) One Link Tolled Both Links Tolled 5 cent/mi Toll 10 cent/mi Toll Maximum Welfare Toll Maximum Revenue Toll Maximum Welfare Toll Maximum Revenue Toll Construction Costs1 $5.2M $5.2M $5.2M $5.2M $5.2M $5.2M $5.2M Maint. Costs1 $0.22M $0.22M $0.22M $0.22M $0.22M $0.22M $0.22M Manage. Costs1 $0 $0.8M $0.8M $0.8M $0.8M $1.6M $1.6M Revenue1 $0 $4.98M $8.15M $7.18M $9.68M $20.62M $25.38M Traveler Benefits1 $25.8M $22.8M $19.3M $22.4M $16.5M $14.2M $4.8M Yearly Costs2 $5.2M $5.2M $5.2M $5.2M $5.2M $5.2M $5.2M Yearly Benefits2 $25.6M $26.8M $26.4M $28.5M $25.1M $33.0M $28.3M NPV of Costs3 $80.0M $80.0M $80.0M $80.0M $80.0M $80.0M $80.0M NPV of Benefits3 $393.3M $412.0M $406.6M $438.5M $386.6M $507.4M $435.4M Total NPV $313.3M $332.0M $326.6M $358.5M $306.6M $427.4M $355.4M B-C Ratio 4.917 5.150 5.082 5.482 4.832 6.343 5.443 1All values are shown in millions of $/year. Construction costs are computed assuming a 5% interest rate on 30-year loan. Revenues and traveler benefits are computed assuming 261 weekdays and 104 weekend days per year, where weekend-day revenues and traveler benefits are one half those of regular weekdays. 2Yearly costs include only construction costs and yearly benefits include maintenance and management costs, revenue, and traveler benefits. All values are shown in millions of $/year. 3NPVs are calculated using a discount rate of 5% per year. Of course, the NPV of each scenario depends greatly on the assumed discount rate of 5%, though, in this case, the rankings of scenarios will be the same regardless of chosen discount rate (since annual demand, costs and benefits are simply assumed constant over the 30 year period of analysis). However, it is of interest to understand the sensitivity of NPV to the chosen discount rate. The welfare-maximizing tolls on a single route yield a NPV of $358.5 million, when discounting at 5% per year (as shown in Table 48). If the discount rate is changed to 3%, the NPV estimate rises to $457.2 million (28% greater than discounting at 5%). In contrast, if the discount rate is 7%, the NPV estimate falls to just $289.4 million (19% less than the original). Thus, for even small deviations in the discount rate, large fluctuations in the estimated NPV may result.

259 A.6.2.4. Summary The example application provided here illustrates how Logsums can be used as a measure of traveler welfare, and how cost-benefit analysis can be used as a tool for project evaluation in toll road settings. In addition, theRoH estimates of welfare change were illustrated, in order to demonstrate how close and how far they can be from the Logsum measure (when wholly new routes/alternatives are added versus existing routes are tolled). As shown in eight numerical examples, with two distinctive (and latently heterogeneous) traveler types, congestion levels can be largely reduced in the presence of pricing (even with flat tolls), and estimated net welfare effects can be significant (even when tolls are set to maximize revenue). While disparities exist between the welfare benefits of low- and high-VOT travelers, these are lessened when a no-toll or low-tolled option is preserved. The results also show how congestion pricing can provide a means to finance new highways along previously congested corridors. In the congested-corridor context examined here, it was found that all but one pricing policy led to revenues that could fully finance the infrastructure costs within 30 years, with excess revenues. Such excess revenues can be used for any number of things. Of course, the analysis provided here illustrates only key concepts with an idealized set of scenarios. Nonetheless, it shows how a variety of pricing policy options exist for those willing to invest in new transportation infrastructure that offers travel time savings. The tools and techniques highlighted here illustrate practical methods for identifying welfare-enhancing and cost-recovering investment opportunities. These techniques recognize demand elasticity across times of day, destinations, modes and routes, which are standard features of most travelers’ choice sets, but which are too often lacking in most analysts’ toolkits and not applied. A.6.3.Conclusions and Recommendations The topics covered in this appendix seek to aid planners and decision makers in using transportation models to inform the decision-making process, with emphasis on toll road projects. Traveler welfare calculations were discussed at length, with a focus on logsum calculations across discrete alternatives. Such measures of traveler welfare provide the most rigorous estimates when demand estimates are a consequence of discrete choice models, such as with the MNL and NL specifications. However, they are not appropriate in model specifications, where multiple choice dimensions are modeled in a largely sequential and less integrated fashion. In such cases, the RoH may instead provide reasonable estimates of welfare changes, at least when existing road policies are only modified (e.g., tolling is added to existing systems). When choice alternatives are added (such as new roadways), however, the RoH is inappropriate. CBA techniques are invaluable in weighing attributes of different alternatives, and a variety of measures (e.g., B-C ratios, NPVs, and IRRs) can support objective results. The NPV approach may be the most robust of the CBA measures discussed here, since it offers a quantity for direct comparison across potential projects, with obvious dollar– value implications. Of course, it can be difficult to measure all project impacts in monetary terms (as discussed by Small, 1999 ) and each project is unique; thus ultimately, there is no substitute for expert judgment in toll road project evaluation, as a complement to such calculations.

260 As discussed, discount rate selection is a component of all cost-benefit analyses, meriting serious consideration. OMB, 2003 guidelines suggest a real rate of 7% for public projects. However, it is especially important for toll road projects in order to properly evaluate the implications of the chosen discount rate (whatever it is), adjusting it up and down by 2, 3, or even 5 percentage points. Robust investment decisions and tolling policies should rank near the top (of all potential policies) across various discount rates (though NPV generally vary substantially across discount rates). Toll rate selection is a critical component of toll project evaluation. Three methods for rate selection were discussed here: welfare maximization, revenue maximization, and throughput (flow) maximization. While each is distinct and can result in very different toll levels, different stakeholders will prefer different objectives, and it can be valuable to explore the implications of all three approaches. Such investigations allow planners and policymakers to quantify the relative closeness of a project’s/policy’s expected welfare, revenue, and throughput, to optimal levels. More robust policies will perform relatively well across multiple measures. In the end, of course, the pursuit of social welfare maximization is a meaningful goal likely to appeal most to the traveling public and policymakers.

Abbreviations and acronyms used without definitions in TRB publications: AAAE American Association of Airport Executives AASHO American Association of State Highway Officials AASHTO American Association of State Highway and Transportation Officials ACI–NA Airports Council International–North America ACRP Airport Cooperative Research Program ADA Americans with Disabilities Act APTA American Public Transportation Association ASCE American Society of Civil Engineers ASME American Society of Mechanical Engineers ASTM American Society for Testing and Materials ATA American Trucking Associations CTAA Community Transportation Association of America CTBSSP Commercial Truck and Bus Safety Synthesis Program DHS Department of Homeland Security DOE Department of Energy EPA Environmental Protection Agency FAA Federal Aviation Administration FHWA Federal Highway Administration FMCSA Federal Motor Carrier Safety Administration FRA Federal Railroad Administration FTA Federal Transit Administration HMCRP Hazardous Materials Cooperative Research Program IEEE Institute of Electrical and Electronics Engineers ISTEA Intermodal Surface Transportation Efficiency Act of 1991 ITE Institute of Transportation Engineers NASA National Aeronautics and Space Administration NASAO National Association of State Aviation Officials NCFRP National Cooperative Freight Research Program NCHRP National Cooperative Highway Research Program NHTSA National Highway Traffic Safety Administration NTSB National Transportation Safety Board PHMSA Pipeline and Hazardous Materials Safety Administration RITA Research and Innovative Technology Administration SAE Society of Automotive Engineers SAFETEA-LU Safe, Accountable, Flexible, Efficient Transportation Equity Act: A Legacy for Users (2005) TCRP Transit Cooperative Research Program TEA-21 Transportation Equity Act for the 21st Century (1998) TRB Transportation Research Board TSA Transportation Security Administration U.S.DOT United States Department of Transportation