Travel Demand Modeling Moshe Ben-Akiva 1.201 / 11.545 / ESD.210 Transportation Systems Analysis: Demand & Economics Fall 2008Review ● Discrete Choice Framework – A decision maker n selects one and only one alternative i from a choice set Cn={1,…,Jn} – Random Utility Model where Uin = Vin (attributes of i, characteristics of n, β) +εin ● Discrete Choice Models – Multinomial Logit – Nested Logit • Correlated Alternatives • Multidimensional Choice Next… Travel Demand Modeling 2Outline ● Introduction ● Approaches – Trip – Tour – Activity ● Emerging Approaches 3Long Term Choices ● Urban Development • Firm location and relocation decisions • Firm investment in information technology ● Mobility and Lifestyle Decisions • Labor force participation • Workplace location • Housing • Automobile ownership • Information technology ownership and access • Activity program 4Activity and Travel Pattern Choices ● Activity sequence and duration ● Priorities for activities ● Tour formation ● Telecommunications options ● Access travel information – Traffic conditions – Route guidance – Parking availability – Public transportation schedules ● Reschedule activities ● Revise travel plans 5Modeling Framework Land Use and Economic Development Transportation System Performance Household & Individual Behavior Lifestyle and Mobility Decisions Activity and Travel Scheduling Implementation and Rescheduling Long Term Short Term 6The Fundamental Modeling Problem ● Adequately represent a decision process that has an inordinate number of feasible outcomes in many dimensions ● Example - Activity Schedule N um ber of activities 10 10 Sequence Tim ing Location M ode R oute Total Num ber of Activity Schedule Alternatives 10 per activity 1000 per activity 5 per activity 10 per activity 10! 100 10,000 50 100 1017 ● Simplify ● Achieve valid results 7Simplifying the Problem ● Discrete time intervals ● Individuals defined by socioeconomic variables ● Divide space into zones ● Categories of activities ● Depiction of travel patterns � trips, tours, activity schedules 8Approaches to Modeling Travel ● Trip-based ● Integrated trip-based ● Tour-based ● Activity schedule 9Representing Activity/Travel Behavior Schedule Tours Trips Space D S H Space Space H D S H D S H H H H D S H W W W W H HH Time Time Time H: Home W: Work S: Shop D: Dinner out 10Trip-Based: The 4-Step Model Trip Purpose Home-based work (HBW) Home-based shop (HBS) Home-based other (HBO) Non-home-based (NHB) Behavioral Steps 1. Trip Generation (Frequency) 2. Trip Distribution (Destination) 3. Modal Split (Mode) 4. Assignment (Route) 11The 4-Step Model: Trip Generation ● Trip Production • Household Size, Household Structure, Income, Car Ownership, Residential Density, Accessibility ● Trip Attractions • Land-use and Employment by Category (e.g. Industrial, Commercial, Services), Accessibility ● Cross Classification, Regression, Growth Factor 12The 4-Step Model: Trip Distribution ● Trip matrix Generations 1 2 3 : i : I 1 T11 T21 T31 : Ti1 : TI1 2 T12 T22 T32 : Ti2 : TI2 3 T13 T23 T33 : Ti3 : TI3 Attractions … j … … T1j … … T2j … … T3j … : … Tij … : … TIj … J T1J T2J T3J : TiJ : TIJ ij j ∑T O1 O2 O3 : Oi : OI ij i ∑T D1 D2 D3 … Dj … DJ i j ∑ ∑ ij T = T 13The 4-Step Model: Trip Distribution ● Gravity Model Oβ ( ) , i = 1 I and j = 1..... JT =α D f C ..... ij i i j j ij ∑Tij = Oi , i = 1..... I j ∑Tij = Dj , j = 1..... J i • Where, -f C ij ) Function of the generalized cost of travel ( = from i to j and -αiand βjare balancing factors ijSolve iteratively for T , αiand βj 14The 4-Step Model: Modal Split ● Logit Vauto P (auto )= eVauto e + eVtransit ● Nested Logit µINM eP(NM ) = eµINM + eµIM 15The 4-Step Model: Assignment ● Route Choice – Deterministic: Shortest Path, Minimum Generalized Cost – Stochastic: Discrete Choice (e.g. Logit) ● Equilibrium – Supply Side – User Equilibrium vs. System Optimal 16Limitations of the Trip-Based Method ● Demand for trip making rather than for activities ● Person-trips as the unit of analysis ● Aggregation errors: – Spatial aggregation – Demographic aggregation – Temporal aggregation ● Sequential nature of the four-step process ● Behavior modeled in earlier steps unaffected by choices modeled in later steps (e.g. no induced travel) ● Limited types of policies that can be analyzed 17Complexity of Work Commute (Boston) Simple Commute (no other activities) Complex Commute (includes non-work activities) home work home work daycare bank 64% Complex 36% Simple 77% Complex 23% Simple 60% Complex 40% Simple Females with Males with All Adults Children Children 18 Source: Ben-Akiva and Bowman, 1998, “Activity Based Travel Demand Model Systems,” in Equilibrium and Advanced Transportation Modeling, Kluwer Academic.Complex Responses to Policies Example: Peak-Period Toll Source: Bowman, 1998, “The Day Activity Schedule Approach to Travel Demand Analysis,” PhD Thesis, MIT 19 ShopShopShopShopWorkWorkWork(Home)SpaceSpaceSpace(a) ChangeMode & Pattern(b) ChangeTime & PatternPotential Responses to Toll(c) Work at HomeSpaceTime= Peak PeriodTimeTimeTimePre-Toll ScheduleFigure by MIT OpenCourseWare.Modeling Travel at the Level of the Individual ● Classic 4-step – Trip Frequency – Destination Choice – Mode Choice – Route Choice ● Beyond 4-step – Time of Day – Integrated Trips – Tours 20--Home based work tripsIntegrated Trip-Based Framework (e.g., MTC, STEP) Auto ownership Home Based Work trips Home Based Other trips Non-Home Based trips 21Highlights of Integrated Trip-Based System ● Key features – Disaggregate choice models – Models are integrated, via conditionality and measures of inclusive value, according to the decision framework ● Key weakness – Modeling of trips rather than explicit tours 22Tour-Based Framework (e.g. Stockholm) Other Tours Business Shopping Personal Business Work Tours Other 23Highlights of Tour-Based System ● Key features – Explicitly chains trips in tours – Validated and widely applied ● Key weaknesses – Lacks an integrated schedule pattern – Doesn’t integrate well the time dimension ● Data requirements – Same as for trip-based models 24Basics of Activity-Based Travel Theory ● Travel demand is derived from demand