CE

E 3

20W

inte

r 20

06

Traffic Forecasting

CEE 320Steve Muench

CE

E 3

20W

inte

r 20

06

Outline

1. Need for Traffic Forecasting

2. Traveler Decisions

3. Trip Generation

4. Mode Choicea. Survey

CE

E 3

20W

inte

r 20

06

Need for Traffic Forecasting

• Impacts of facilities or modes of travel– Lines on existing roads– Roads

– Light rail

– Bus service

• Geometric design• Pavement design

CE

E 3

20W

inte

r 20

06

Traveler Decisions

• Types of decisions– Time (when do you go?)– Destination (where do you go?)

– Mode (how do you get there?)

– Route choice (what route do you choose?)

• Influences– Economic

– Social

CE

E 3

20W

inte

r 20

06

Predicting Travel Decisions

• Collect data on travel behavior– Observation (number of buses, cars, bikes, etc.)– Surveys

• Collect data on what travelers have done

• Collect data on their values and choices (utility)

• Inexact nature of prediction– Incomplete data– Reporting problems

CE

E 3

20W

inte

r 20

06

Trip Generation & Mode Choice

CE

E 3

20W

inte

r 20

06

Trip Generation

• Purpose– Predict how many trips will be made– Predict exactly when a trip will be made

• Approach– Aggregate decision-making units – Categorized trip types– Aggregate trip times (e.g., AM, PM, rush hour)

– Generate Model

CE

E 3

20W

inte

r 20

06

Motivations for Making Trips

• Lifestyle– Residential choice– Work choice

– Recreational choice

– Kids, marriage– Money

• Life stage• Technology

CE

E 3

20W

inte

r 20

06

Reporting of Trips - Issues

• Under-reporting trivial trips• Trip chaining

• Other reasons (passenger in a car for example)

CE

E 3

20W

inte

r 20

06

Trip Generation Models

• Linear (simple)

• Poisson (a bit better)

nnxxxT ββββ ...22110 +++=

nni xxx ββββλ ...ln 22110 +++=

( ) ( )

= −

! tripsofnumber

x

xexP

λλ

CE

E 3

20W

inte

r 20

06

Example

kidsmarriedinternetagegender

autosbus

bicyclesedanvansportssuv

incomeeducationi

*8

*7

*6

*5*4

*#37*36

*35*34*33*32*31

*2*10ln

βββββ

ββ

βββββ

βββλ

+++++

++

+++++

++=

Recreational or pleasure trips measured by λi (Poisson model):

Variable Coefficient Value Product

Constant 0 1 0

Education (undergraduate degree or higher) 0.15 1 0.15

Income 0.00002 45,000 0.9

Whether or not individual owns an SUV 0.1 1 0.1

Whether or not individual owns a sports car 0.05 0 0

Whether or not individual owns a van 0.1 1 0.1

Whether or not individual owns a sedan 0.08 0 0

Whether or not individual uses a bicycle to work 0.02 0 0

Whether or not individual uses the bus to work all the time -0.12 0 0

Number of autos owned in the last ten years 0.06 6 0.36

Gender (female) -0.15 0 0

Age -0.025 40 -1

Internet connection at home -0.06 1 -0.06

Married -0.12 1 -0.12

Number of kids 0.03 2 0.06

Sum = 0.49

λi = 1.632 trips/day

CE

E 3

20W

inte

r 20

06

Example

• Probability of exactly “n” trips using the Poisson model:

• Cumulative probability – Probability of one trip or less: P(0) + P(1) = 0.52– Probability of at least two trips: 1 – (P(0) + P(1)) = 0.48

• Confidence level– We are 52% confident that no more than one recreational or

pleasure trip will be made by the average individual in a day

( ) ( )20.0

!0

0632.10 632.1 =

= −eP ( ) ( )

32.0!1

1632.11 632.1 =

= −eP

CE

E 3

20W

inte

r 20

06

Mode Choice

• Purpose– Predict the mode of travel for each trip

• Approach– Categorized modes (SOV, HOV, bus, bike, etc.)

– Generate Model

CE

E 3

20W

inte

r 20

06

A Mode Choice Model

• Logit Model

• Final form

mkn

kmnmnmk zV εβ += ∑

∑=

s

U

U

mk sk

mk

e

eP

Specifiable part Unspecifiable part

∑=n

kmnmnmk zU β

s = all available alternativesm = alternative being consideredn = traveler characteristick = traveler

CE

E 3

20W

inte

r 20

06

Discrete Choice Example

Regarding the TV sitcom Gilligan’s Island, whom do you prefer?

CE

E 3

20W

inte

r 20

06

Ginger Model

UGinger = 0.0699728 – 0.82331(carg) + 0.90671(mang) + 0.64341(pierceg) – 1.08095(genxg)

carg = Number of working vehicles in household

mang = Male indicator (1 if male, 0 if female)

pierceg = Pierce Brosnan indicator for question #11 (1 if Brosnan chosen, 0 if not)

genxg = generation X indicator (1 if respondent is part of generation X, 0 if not)

CE

E 3

20W

inte

r 20

06

Mary Anne Model

UMary Anne = 1.83275 – 0.11039(privatem) – 0.0483453(agem) – 0.85400(sinm) – 0.16781(housem) + 0.67812(seanm) + 0.64508(collegem) – 0.71374(llm) + 0.65457(boomm)

privatem = number of years spent in a private school (K – 12)

agem = age in years

sinm = single marital status indicator (1 if single, 0 if not)

housem = number of people in household

seanm = Sean Connery indicator for question #11 (1 if Connery chosen, 0 if not)

collegem = college education indicator (1 if college degree, 0 if not)

llm = long & luxurious hair indicator for question #7 (1 if long, 0 if not)

boomm = baby boom indicator (1 if respondent is a baby boomer, 0 if not)

CE

E 3

20W

inte

r 20

06

No Preference Model

Uno preference = – 9.02430x10-6(incn) – 0.53362(gunsn) + 1.13655(nojames) + 0.66619(cafn) + 0.96145(ohairn)

incn = household income

gunsn = gun ownership indicator (1 if any guns owned, 0 if no guns owned)

nojames = No preference indicator for question #11 (1 if no preference, 0 if preference for a particular Bond)

cafn = Caffeinated drink indicator for question #5 (1 if tea/coffee/soft drink, 0 if any other)

ohairn = Other hair style indicator for question #7 (1 if other style indicated, 0 if any style indicated)

CE

E 3

20W

inte

r 20

06

Results10. Regarding the TV sitcom “Gilligan’s Island” whom do

your prefer?

29

9085

30

88 89

7

112

87

0

20

40

60

80

100

120

Ginger Mary Ann No Preference

# o

f R

esp

on

dan

ts

Survey

average

Model

Average probabilities of selection for each choice are shown in yellow. These average percentages were converted to a hypothetical number of respondents out of a total of 207.

CE

E 3

20W

inte

r 20

06

My Results

∑=

s

U

U

mk sk

mk

e

eP

8201.13265.02636.01075.1 =++= −−−∑ eeees

U sk

1815.08201.1

1075.1

===−

∑e

e

eP

s

U

U

ginger sk

mk

4221.08201.1

2636.0

===−

∑e

e

eP

s

U

U

annemary sk

mk

3964.08201.1

3265.0

===−

∑e

e

eP

s

U

U

preferenceno sk

mk

Uginger = – 1.1075

Umary anne = – 0.2636

Uno preference = – 0.3265

CE

E 3

20W

inte

r 20

06

Primary References

• Mannering, F.L.; Kilareski, W.P. and Washburn, S.S. (2005). Principles of Highway Engineering and Traffic Analysis, Third Edition. Chapter 8

• Transportation Research Board. (2000). Highway Capacity Manual 2000. National Research Council, Washington, D.C.