DEMAND MODELS FOR TRANSPORTATION MODES A FOCUS ON THE MEASUREMENT OF - - PowerPoint PPT Presentation

▶
demand models for transportation modes
SMART_READER_LITE
LIVE PREVIEW

DEMAND MODELS FOR TRANSPORTATION MODES A FOCUS ON THE MEASUREMENT OF - - PowerPoint PPT Presentation

May 09, 2023 424 likes •884 views

DEMAND MODELS FOR TRANSPORTATION MODES A FOCUS ON THE MEASUREMENT OF LATENT CONSTRUCTS AFFECTING DECISIONS Aurlie Glerum Ricardo Hurtubia My Hang Nguyen Bilge Atasoy TLA/ToL joint seminar Michel Bierlaire KTH Royal Institute of Technology

slide-1
SLIDE 1

Aurélie Glerum Ricardo Hurtubia My Hang Nguyen Bilge Atasoy Michel Bierlaire

DEMAND MODELS FOR TRANSPORTATION MODES

A FOCUS ON THE MEASUREMENT OF LATENT CONSTRUCTS AFFECTING DECISIONS

TLA/ToL joint seminar KTH Royal Institute of Technology Friday 12th October 2012

slide-2
SLIDE 2

Introduction & motivation Methodology The data

  • Vehicle choice case study
  • Mode choice case study

Incorporation of measurements into HCM

  • Vehicle choice case study (ICLV example)
  • Mode choice case study (ICLC example)

Conclusion

OUTLINE

2

slide-3
SLIDE 3

Recent developments in demand modeling for transportation

  • Hybrid choice model (HCM) framework (Walker, 2001; Ben-Akiva et al., 2002)

Comprehensive framework that allows to incorporate unobservable factors as explanatory variables of choice.

  • Choice of transportion mode, car, etc.
  • Influenced by economic factors:
  • Often also involve more subjective factors:
  • HCM framework incorporates these subjective factors.

3

INTRODUCTION & MOTIVATION

Discrete choice model (DCM) Latent variable model (LVM)

  • r

Latent class model (LCM) +

  • Price
  • Trip duration
  • Etc.
  • Attitudes
  • Perceptions
  • Lifestyles
  • Habits
slide-4
SLIDE 4

Figure extracted from Walker and Ben-Akiva, 2002.

INTRODUCTION & MOTIVATION

4

Hybrid choice model (HCM): DCM with latent constructs.

slide-5
SLIDE 5

Hybrid choice model (HCM): DCM with latent constructs.

Latent construct can be… either a latent class model

  • Unobservable construct is discrete
  • Useful for segmentation according to lifestyle

Figure extracted from Walker and Ben-Akiva, 2002.

INTRODUCTION & MOTIVATION

5

slide-6
SLIDE 6

Hybrid choice model (HCM): DCM with latent constructs.

Latent construct can be… or a latent variable model

  • Unobservable construct is continuous
  • Useful to analyze the impact of changes in prices across individuals  pricing

Figure extracted from Walker and Ben-Akiva, 2002.

INTRODUCTION & MOTIVATION

6

slide-7
SLIDE 7

INTRODUCTION & MOTIVATION

7

Important issues in the use of HCMs: 1. Measurement of latent variable / latent class How to obtain the most realistic and accurate measure of an attitude / perception / lifestyle? Opinion statements: usual way in the literature 2. Integration of the measurement into the choice model How to incorporate this information in the choice modeling framework?

slide-8
SLIDE 8

INTRODUCTION & MOTIVATION

8

Important issues in the use of HCMs: 1. Measurement of latent variable / latent class How to obtain the most realistic and accurate measure of an attitude / perception / lifestyle? Opinion statements: usual way in the literature 2. Integration of the measurement into the choice model How to incorporate this information in the choice modeling framework?

Focus of this research: measurement model

slide-9
SLIDE 9

METHODOLOGY

9

Integration of the measurement into the choice model

Latent construct Utilities Choice indicators Explanatory variables

Disturbances Disturbances Disturbances

Measurement indicators

slide-10
SLIDE 10

METHODOLOGY

10

Integration of the measurement into the choice model:

  • Structural equation model (SEM) framework used to characterize latent

construct and relate it to its measurement indicators

(e.g. Bollen, 1989; Hancock and Mueller, 2006; Bartholomew et al., 2011).

Latent construct Utilities Choice indicators Explanatory variables

Disturbances Disturbances Disturbances

Measurement indicators

slide-11
SLIDE 11

METHODOLOGY

11

Integration of the measurement into the choice model

  • In transportation applications:
  • Heterogeneity of latent construct (e.g. attitude) captured among population
  • But: also need to capture heterogeneity in reporting indicators of latent

construct

Latent construct Utilities Choice indicators Explanatory variables

Disturbances Disturbances Disturbances

Measurement indicators

slide-12
SLIDE 12

METHODOLOGY

12

Integration of the measurement into the choice model

  • In transportation applications:
  • Heterogeneity of latent construct (e.g. attitude) captured among population
  • But: also need to capture heterogeneity in reporting indicators of latent

construct

Latent construct Utilities Choice indicators Explanatory variables

Disturbances Disturbances Disturbances

Measurement indicators

slide-13
SLIDE 13

METHODOLOGY

13

Integration of the measurement into the choice model

  • In transportation applications:
  • Heterogeneity of latent construct (e.g. attitude) captured among population
  • But: also need to capture heterogeneity in reporting indicators of latent

construct

Latent construct Utilities Choice indicators Explanatory variables

Disturbances Disturbances Disturbances

Measurement indicators

Focus of this presentation

slide-14
SLIDE 14

METHODOLOGY

14

Model specification Likelihood function given by: with Integrated choice and latent variable model Integrated choice and latent class model

∏

=

=

N n in n in

X I y f L

1

) , , , ; | , (

ω

σ λ β α

∫

⋅ ⋅ =

*

* * * *

) , ; | ( ) ; , | ( ) ; , | ( ) , , , ; | , (

n in

X n n n n in n y n in in in n in

dX X X f X X I f X X y P X I y f

ω ω

σ λ α β σ λ β α

   = =

  • therwise

max if 1

jn j in in

U U y

in

y S s n in n in in in n in

X s P s X I P s X y P X I y P       ⋅ ⋅ = ∑

∈

) , ; | ( ) ; , | ( ) ; , | ( ) , , , ; | , (

ω ω

σ λ α β σ λ β α

slide-15
SLIDE 15

METHODOLOGY

15

Model specification Likelihood function given by: with Integrated choice and latent variable model Integrated choice and latent class model

∏

=

=

N n in n in

X I y f L

1

) , , , ; | , (

ω

σ λ β α

∫

⋅ ⋅ =

*

* * * *

) , ; | ( ) ; , | ( ) ; , | ( ) , , , ; | , (

n in

X n n n n in n y n in in in n in

dX X X f X X I f X X y P X I y f

ω ω

σ λ α β σ λ β α

   = =

  • therwise

max if 1

jn j in in

U U y

in

y S s n in n in in in n in

X s P s X I P s X y P X I y P       ⋅ ⋅ = ∑

∈

) , ; | ( ) ; , | ( ) ; , | ( ) , , , ; | , (

ω ω

σ λ α β σ λ β α

Few examples that incorporate socio-economic information into the measurement model

slide-16
SLIDE 16

THE DATA

16

Two case studies: 1. Integrated choice and latent variable model (ICLV): analysis of the impact of pro-convenience attitude on choice of car. 2. Integrated choice and latent class model (ICLC): analysis of the transportation mode choices for individuals segmented according to dependent / independent classes.

Car purchase choice case study Mode choice case study

slide-17
SLIDE 17

THE DATA

17

Choice Gasoline / diesel New alternative: Electric Competitors Renault Renault

VEHICLE CHOICE CASE STUDY Stated preferences (SP) survey:

  • Car purchase choice study
  • Conducted in Switzerland in 2011 among

individuals who bought a new car recently or intend to buy one soon.

  • Conducted with Renault Suisse SA.
  • Customized choice situations
  • 693 questionnaires obtained
slide-18
SLIDE 18

THE DATA

18

VEHICLE CHOICE CASE STUDY

  • Environmental concern
  • Attitude towards new technologies
  • Perception of the reliability of an

electric vehicle

  • Perception of leasing
  • Attitude towards design

Ratings

  • Total disagreement (1)
  • Disagreement (2)
  • Neutral opinion (3)
  • Agreement (4)
  • Total agreement (5)
  • I don’t know (6)

An electric car is a 100% ecological solution. A control screen is essential in my use of a car. Electric cars are not as secure as gasoline cars. Leasing is an optimal contract which allows me to change car frequently. Design is a secondary element when purchasing a car, which is above all a practical transport mode.

Opinion statements related to five themes

slide-19
SLIDE 19

THE DATA

19

Revealed preferences (RP) survey

  • Mode choice study
  • Conducted between 2009-2010 in low-density

areas of Switzerland

  • Conducted with PostBus (major bus company

in Switzerland, operates in low-density areas)

  • Info on all trips performed by inhabitants in
  • ne day:
  • Transport mode
  • Trip duration
  • Cost of trip
  • Activity at destination
  • Etc.
  • 1763 valid questionnaires collected

Choice

MODE CHOICE CASE STUDY

slide-20
SLIDE 20

THE DATA

20

Opinion statements related to four themes

  • Environment
  • Mobility
  • Residential choice
  • Lifestyle

Ratings

  • Total disagreement (1)
  • Disagreement (2)
  • Neutral opinion (3)
  • Agreement (4)
  • Total agreement (5)
  • I don’t know (6)

MODE CHOICE CASE STUDY

The price of gasoline should be increased in order to reduce traffic congestion and air pollution. Taking the bus helps making a town more comfortable and welcoming. Accessibility and mobility conditions are important in the choice of an accommodation. I always plan my activities a long time in advance.

slide-21
SLIDE 21

Role of indicators of latent construct:

  • Measure a latent variable
  • Enhance a latent class model

Issue: biases in the measurement of indicators due to heterogeneity of response behavior By introducing socio-economic information into the measurement component of the HCM, the bias is reduced. Two examples:

  • Car choice case study (ICLM): capture exaggeration effects in responses

to indicators.

  • Transportation mode choice case study (ICLC): capture bias in responses

to indicators due to various socio-economic characteristics.

21

INCORPORATION OF MEASUREMENTS INTO HCM

slide-22
SLIDE 22

INCORPORATION OF MEASUREMENTS INTO HCM

22

Motivation for integration of explanatory factors of measurement indicators:

  • Dispersion effects:
  • Exaggeration effects in experiments on survey design in social science

literature (Schuman and Presser, 1996)

  • Some individuals tend to report responses at extremities of scale of

agreement though their commitment to the opinion statement is not strong.

  • Socio-economic characteristics might explain different response behaviors

Need to account for heterogeneity of response behavior

slide-23
SLIDE 23

INCORPORATION OF MEASUREMENTS INTO HCM

23

1. Integrated choice and latent variable model (ICLV): analysis of the impact of pro-convenience attitude on choice of car.

Vehicle choice case study

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

Choice Gasoline / diesel New alternative: Electric Competitors Renault Renault

slide-24
SLIDE 24

INCORPORATION OF MEASUREMENTS INTO HCM

24

Battery lease Incentive Gasoline / electricity costs Purchase price Utility Explanatory variables Competitors gasoline, Renault gasoline, Renault electric Choice Socio-economic characteristics Pro-convenience attitude Indicators Design is secondary, a car is a practical transport mode. Spaciousness / capacity more important than look. New propulsion technology more important than look. Explanatory variables

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

Gender Number of people in household Age Retired Owner Electric car model Explanatory variables Exaggeration effects

Latent variable model Discrete choice model

slide-25
SLIDE 25

25

Definition of index:

  • Definition of degree of extremity

with

  • En : number of occurrences of ‘total disagreement’ and ‘total

agreement’ for individual n over all R opinion questions of the survey

INCORPORATION OF MEASUREMENTS INTO HCM

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

∑

=

=

R r rn n

J E

1

   = = =

  • therwise

5

  • r

1 if 1

rn rn rn

I I J

slide-26
SLIDE 26

26

Definition of scale parameter:

  • Measurement model:
  • Scale that captures heterogeneity in response behavior:

INCORPORATION OF MEASUREMENTS INTO HCM

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

) ( ) 1 ( 1

n E E

E I I

Ext n n n

ν θ θ ν

σ σ ⋅ − + ⋅ =

< <

γ

θ θ

⋅ ⋅ − + ⋅ =

< < n E E

E I I

n n

) 1 ( 1

n n n

X m I υ α + = ) ; (

* *

) , ( ~

n

Logistic

n υ

σ υ

slide-27
SLIDE 27

27

Definition of scale parameter:

  • Measurement model:
  • Scale that captures heterogeneity in response behavior:

INCORPORATION OF MEASUREMENTS INTO HCM

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

) ( ) 1 ( 1

n E E

E I I

Ext n n n

ν θ θ ν

σ σ ⋅ − + ⋅ =

< <

γ

θ θ

⋅ ⋅ − + ⋅ =

< < n E E

E I I

n n

) 1 ( 1

n n n

X m I υ α + = ) ; (

* *

) , ( ~

n

Logistic

n υ

σ υ

Define threshold θ above which individuals show extreme behavior Statistical analyses show that highest fit for θ = 7.

slide-28
SLIDE 28

28

Definition of scale parameter:

  • Measurement model:
  • Scale that captures heterogeneity in response behavior:

INCORPORATION OF MEASUREMENTS INTO HCM

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

) ( ) 1 ( 1

n E E

E I I

Ext n n n

ν θ θ ν

σ σ ⋅ − + ⋅ =

< <

γ

θ θ

⋅ ⋅ − + ⋅ =

< < n E E

E I I

n n

) 1 ( 1

n n n

X m I υ α + = ) ; (

* *

) , ( ~

n

Logistic

n υ

σ υ

Group-specific scale

slide-29
SLIDE 29

29

Definition of scale parameter:

  • Measurement model:
  • Scale that captures heterogeneity in response behavior:

INCORPORATION OF MEASUREMENTS INTO HCM

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

) ( ) 1 ( 1

n E E

E I I

Ext n n n

ν θ θ ν

σ σ ⋅ − + ⋅ =

< <

γ

θ θ

⋅ ⋅ − + ⋅ =

< < n E E

E I I

n n

) 1 ( 1

n n n

X m I υ α + = ) ; (

* *

) , ( ~

n

Logistic

n υ

σ υ

Progressive scale:

  • The higher the degree of extremity, the higher the scale.
  • γ parameter to estimate
slide-30
SLIDE 30

Results for the latent variable model Simultaneous estimation of the HCM using the extended version of Biogeme (Bierlaire and Fetiarison, 2009)

30

INCORPORATION OF MEASUREMENTS INTO HCM

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

slide-31
SLIDE 31

Results from the latent variable model

31

INCORPORATION OF MEASUREMENTS INTO HCM

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

γ σ

θ θ υ

⋅ ⋅ − + ⋅ =

< < n E E

E I I

n n n

) 1 ( 1

slide-32
SLIDE 32

Results from the latent variable model We observe dispersion effects, since for the ‘extreme’ group we have:

32

INCORPORATION OF MEASUREMENTS INTO HCM

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

γ σ

θ θ υ

⋅ ⋅ − + ⋅ =

< < n E E

E I I

n n n

) 1 ( 1 42 . 1 7 = ⋅ = γ συn

slide-33
SLIDE 33

Results from the choice model

33

INCORPORATION OF MEASUREMENTS INTO HCM

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

Pro-convenience attitude significantly affects car choice.

slide-34
SLIDE 34

Improvement of fit over model without dispersion effects

34

INCORPORATION OF MEASUREMENTS INTO HCM

VEHICLE CHOICE CASE STUDY (ICLV EXAMPLE)

slide-35
SLIDE 35

INCORPORATION OF MEASUREMENTS INTO HCM

35

2. Integrated choice and latent class model (ICLC): analysis of the transportation mode choices for individuals segmented according to dependent / independent classes.

Mode choice case study

Choice

MODE CHOICE CASE STUDY (ICLC EXAMPLE)

slide-36
SLIDE 36

36

INCORPORATION OF MEASUREMENTS INTO HCM

MODE CHOICE CASE STUDY (ICLC EXAMPLE)

Latent classes

High income Utility Family Single Independent Hard to take PT when I travel with my children. Dependent

Latent class model Class-specific choice model

Indicators Number of children Number of cars Number of bikes Student Travel time Travel cost Trip purpose Distance French part vs German part Urban vs rural Explanatory variables Explanatory variables With my car, I can go where I want when I want. I would like to spend more time with my family and friends. Children Number of cars Full time job Couples + children Couples - children Single parents Explanatory variables

Private motorized modes (PMM), Public transportation (PT), Soft modes (SM)

Choice

slide-37
SLIDE 37

Class-specific measurement equations:

37

INCORPORATION OF MEASUREMENTS INTO HCM

MODE CHOICE CASE STUDY (ICLC EXAMPLE)

slide-38
SLIDE 38

Class-specific measurement equations: Class-specific parameters

38

INCORPORATION OF MEASUREMENTS INTO HCM

MODE CHOICE CASE STUDY (ICLC EXAMPLE)

slide-39
SLIDE 39

Class-specific measurement equations: Class-specific parameters Socio-economic information as explanatory variables of response to indicators

39

INCORPORATION OF MEASUREMENTS INTO HCM

MODE CHOICE CASE STUDY (ICLC EXAMPLE)

slide-40
SLIDE 40

Estimation results for ICLC

  • Increase of the significance of the parameters of the latent class

model.

40

INCORPORATION OF MEASUREMENTS INTO HCM

MODE CHOICE CASE STUDY (ICLC EXAMPLE)

slide-41
SLIDE 41

Estimation results for LCCM

  • Increase of the significance of the parameters of the latent class

model.

  • Income parameter has become more important.

41

INCORPORATION OF MEASUREMENTS INTO HCM

MODE CHOICE CASE STUDY (ICLC EXAMPLE)

slide-42
SLIDE 42

Model application: computation of VOT

  • VOTs comparable with literature on transport economics (Jara-Diaz, 2007),

where VOT can be compared to wage rate.

  • Individuals in the independent class have higher incomes (> 8000 CHF),

hence a higher value of time.

42

INCORPORATION OF MEASUREMENTS INTO HCM

MODE CHOICE CASE STUDY (ICLC EXAMPLE)

slide-43
SLIDE 43

CONCLUSION

Main findings:

  • Heterogeneity of response behavior exists and can be captured by

individual-specific information in measurement model

  • Evidence for the importance of accounting for it:
  • ICLV model of car choice:
  • Significant scale parameter
  • Increases as degree of extremity increases
  • ICLC model of mode choice:
  • Socio-economic characteristics affect response to opinion questions

significantly

  • Parameters of the class membership utility increase in significance
  • VOT are comparable with existing studies

43

slide-44
SLIDE 44

Thanks!

44