A presentation conducted by Mr AHM Mehbub Anwar, University of Wollongong. Presented on Wednesday the 2nd of October 2013. This paper explains how principal-agent theory (PAT) can be used as an analytical tool to understand the traveller-Transport for NSW relationship and minimise the agency uncertainty in the relationship by examining traveller preferences for mode choices. The paper emphasises latent variables and objective attributes together during the choice process within the agency relationship, as a method by which the utility of the principal (traveller) can be maximised and evaluated using a discrete choice experiment, i.e. random parameter logit (RPL) model. The probability of car useis significantly higher than public transport, which indicates that an agency uncertainty exists in the relationship and incorporating traveller preferences in the transport projects may minimise this uncertainty.

1. ENDORSING PARTNERS
Agency in transport service:
Implications of traveller mode choice
objective and latent attributes using
random parameter logit model
The following are confirmed contributors to the business and policy dialogue in Sydney:
•
Rick Sawers (National Australia Bank)
•
Nick Greiner (Chairman (Infrastructure NSW)
Monday, 30th September 2013: Business & policy Dialogue
3rd
Tuesday 1 October to Thursday,
October: Academic and Policy
Dialogue
Presented by: Mr AHM Mehbub Anwar, University of Wollongong
www.isngi.org
www.isngi.org

2. Agency in Transport Service: Implications
of Traveller Mode Choice Objective and
latent Attributes Using Random
Parameter Logit Model
A.H.M. Mehbub Anwar
ISNGI 2013, Wollongong

3. Principal-Agent (Agency) Theory
 Focuses on a relationship between two parties
 A relationship is understood when they involve in an
association wherein
one party (the principal) entrusts/delegates task
and/or work to another party called agent to act on
its behalf (Eisenhardt, 1989; Rungtusanatham et al.,
2007).

4. Principal-Agent (Agency) Theory:
Assumptions
 Potential goal/choice conflicts exist between principal(s)
and agent(s);
 Each party acts in its own self-interest; and
 Informational asymmetry frequently exists between
principals and agents.

5. Principal-Agent (Agency) Problem
 The assumptions reflect the agency problem, in fact.
 This problem is appeared while the agent behaves
opportunistically in such a way that works against the
(goal) welfare of the principal (Barney & Hesterly, 1996).
 The principal can’t monitor agent’s actions

6. Traveller and TfNSW Relationship
 When travellers (principal) entrust their desire for a mode
of transport that is customer-focused (i.e. safe, reliable
and low cost) to the TfNSW (agent), this creates a
metaphorical contract between travellers and the TfNSW
(Transport for NSW), known as an agency contract.
 Due to experiences and skills of TfNSW, TfNSW is
reasonably effective agent to fulfil the goals /
expectations entrusted by travellers.

7. Traveller and TfNSW Relationship
 The tax and travel fares paid by the citizens (travellers)
are the source of funding of TfNSW.
 Maximisation of the travellers’ benefit
 Traveller and TfNSW act most likely in their own selfinterest - thus the contract is often characterised by
agency problem
 Travellers may not trust the quality of services performed
by the TfNSW

8. Traveller Preference, Utility and Agency
Relationship
 Traveller preference and utility are regarded as key
indicators of the traveller-TfNSW relationship
 Utility is considered as a key indicator of traveller
satisfaction/expectation.

9. Hypotheses
 Hypothesis 1 (H1): Traveller preferences influence TfNSW’s
decisions on modal services.
 Hypothesis 2 (H2): Individual specific attributes affect
TfNSW’s planning of modal services.
 Hypothesis 3 (H3): Mode specific attributes and nature of
trips have an effect also on TfNSW’s
decisions on modal service.

10. Data and Methods
• 2008/09 Household Travel Survey
(HTS)- Bureau of transport statistics,
Sydney
• Sydney and Illawarra Statistical
Divisions and the Newcastle SubStatistical Division
• Data from Sydney Statistical Division
(SSD) only (82121 trips)

11. Data and Methods
1
LVs
i) Comfort
ii) Convenience
iii) Safety
iv) Flexibility
v) Reliability
vi) Satisfaction
2
LOS
i) Travel time
ii) Travel cost
iii) Waiting
time
3
SEC
i) Age
ii) Income
iii) Family size
iv) Gender
v) Car ownership
vi) Number of
children
vii) Number of full
time workers
5
TOAs
1Latent
2
3
4
5
variables
Level of service
Socioeconomic characteristics
Trip characteristics
Traditional objective attributes
Fig. 1: List of LV and TOAs
4
TC
i) Trip rate,
ii) Distance
travelled
iii) Trip purpose

12. Data and Methods
Table 1: Description of indicators of LVs
Latent factors
Comfort
Convenience
Safety
Flexibility
Reliability
Satisfaction
Explained by (indicators)
- Enjoy time to read/relax on vehicle
- Stressfulness on vehicle
- Service slower
- Mode availability
- Accessibility (does not go where required)
- Timetable availability
- Safety response for mode used in 1st trip
- Safety response for mode used in 2nd trip
- Safety response for mode used in 3rd trip
- Fixed start and finish times – each day can vary
- Rotating shift
- Roster shift
- Variable hours
- Frequency
- Punctuality
- Faster
- Cleanliness
- Travel time
- Travel cost
- Waiting time
Definitions
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Importance with 1, otherwise 0
Travel time in minutes
Travel cost in Australian dollar
Waiting time in minutes

13. Data and Methods*
 A discrete choice analysis is the most popular method (Train, 2009).
(1) Structural equation model
(SEM)
(2) Discrete choice model
 MIMIC (Multiple Indicators
and Multiple Causes) model
 Software used: AMOS v.19
 Random parameter logit (RPL) model
 Software used: Nlogit v.4
 The indicators of LVs have been evaluated and validated using factor
analytic model (exploratory and confirmatory factor) (for details please
see Anwar et al., 2011)
*Similar
methods have been used in Anwar et al. (in press)

14. Data and Methods
MIMIC Model
Structural equation: ηijl = Σrαjlr * sijr + νijl
(1)
Measurement equation: yijp = Σlγjlp * ηijl + ζijp
(2)
ηijl =
yijp =
αjlr and γjlp =
sijr =
νijl and ζijp=
Latent variables
Indicators
Vector parameter to be estimated
Observed explanatory variables
Error terms
i = individual
j = alternative mode of transport
l = a LV
r = explanatory variables to TOAs
p = an indicator

15. Data and Methods
Comfort
Indicator - y1
Indicator – y2
Specification of MIMIC Model
Indicator – y3
Income
Indicator – y4
Convenience
Age
Example:
Comfort ij = α inc-com,j *Income i +
α age-com,j *Age i + α gencom,j *Gender i + α car-com,j *Car
ownership i + α ftw-com,j *Full
time workers i + α dtcom,j *Distance travelled + α chicom,j *Having children + ν com,ij
Indicator – y5
Gender
Indicator – y6
Having children
Indicator – y7
Car ownership
Safety
Indicator – y8
Indicator – y9
Travel time
Indicator – y10
Travel cost
Waiting time
Indicator – y11
Flexibility
Indicator – y12
Family size
Indicator – y13
Full time worker
Indicator – y14
Indicator – y15
Trip rate
Reliability
Trip purpose
Indicator – y16
Distance travelled
Indicator – y17
Indicator – y18
Indicator – y19
y y1,ij = γ y1,j * Comfort ij + ζ y1,ij
Satisfaction
Indicator – y20
Process of structural and measurement relationship (Anwar et al., in press)

16. Data and Methods
Hybrid discrete choice modelling
(i) sequential (also known as two-step) approach and
(ii) simultaneous approach
Step 1: A MIMIC model (a type of regression model with a latent
dependent variable(s) is estimated; and
Step 2: A choice model with random parameters is estimated with
incorporating the information from the first step.

17. Data and Methods
Hybrid discrete choice modelling
Why sequential approach:
(1) the estimated results were not statistically different from
simultaneous (Raveau et al., 2010);
(2) it is less cumbersome (Johansson et al. 2006)
(3) travel decision itself is sequential

18. Data and Methods
Hybrid random parameter logit model
Uij = Vij + εij,
(3)
Vij = Σkθjk * Xijk + Σlβjl * ηijl
(4)
Specifications of RPL
Uij =
xijβj
Set of explanatory variables observed
by the researcher. For example: SEC,
LOS and TC
= xijβj
Deterministic
component
εij,
+
+
(5)
Variables not observed by the researcher.
(Stochastic influence / error term)
zijηi +
Additional random term.
It models the presence of
correlation or heteroscedasticity
among alternatives
ϕij (6)
Random
component
Problem: When β parameters vary in the population and the researcher is not able to explain it.

19. Data and Methods
Specifications of RPL Model
 Model of logit Pr (j|η) = Lj (η) = (expXjβj+Zjη)/(ΣJexpXJβJ+ZJη)
(7)
 To derive RPL model from eq. 6 ϕ is assumed as IID extreme value η
follows a general distribution, f(η|Ω).
 As η is not given, the (unconditional) choice probability is this logit
formula integrated over all values of η weighted by the density of η is the
RPL model as below:
P(j) = ∫η[(eXjβj+Zjη)/(ΣkeXkβk+Zkη)]f(ηΩ)∂η
(8)
 Estimating β (random parameter) and Ω (non-random parameter).

20. Empirical Results
Table 2: MIMIC model results (α): structural equations (t-values in the parenthesis)
Travel
time
Comfort
Convenience
Flexibility
Safety
Reliability
Satisfaction
Model fit criteria
GFI
AGFI
NFI
CFI
RMSEA
Lower bound
upper bound
Travel
cost
Waiting
time
Age
Income
Family
size
Gender
Car
ownership
No. child
Full
time
Trip rate
Distance
travelled
Trip
purpose
-0.055
(-2.10)
-0.127
(-9.51)
-0.171
(-7.52)
-0.166
(-6.23)
-0.444
(-5.24)
-0.129
(-1.98)
-0.202
(-5.77)
-0.058
(-2.00)
-0.004
(-1.99)
-0.100
(-3.04)
-0.022
(1.87)
-0.155
(-6.66)
-0.175
(-2.00)
-0.222
(-4.35)
-0.067
(2.99)
-0.089
(-1.97)
-0.107
(-3.33)
-0.077
(-2.80)
-0.014
(-11.1)
-0.132
(-2.45)
-0.184
(-4.12)
-0.258
(-3.45)
-0.142
(-4.44)
-0.143
(-11.11)
0.145
(2.72)
0.189
(2.33)
0.082
(-3.50)
-0.136
(-4.49)
0.026
(2.17)
0.028
(4.52)
-0.008
(-3.15)
-0.006
(-3.45)
0.021
(5.10)
0.011
(6.0)
-0.009
(-2.10)
-0.086
(-4.44)
0.054
(3.35)
0.189
(2.85)
-0.106
(-3.13)
-0.08
(-6.85)
0.074
(3.85)
-0.086
(-3.45)
0.221
(5.00)
0.132
(5.63)
-0.011
(-2.50)
-0.087
(-6.78)
0.122
(3.21)
0.102
(6.19)
0.221
(4.21)
0.136
(2.89)
-0.121
(-6.37)
-0.121
(-6.37)
0.013
(4.25)
0.109
(15.25)
0.008
(2.03)
0.058
(4.68)
0.137
(3.43)
0.012
(2.00)
0.012
(2.00)
0.019
(3.17)
0.107
(17.83)
0.111
(4.84)
0.115
(2.05)
0.160
(8.00)
0.168
(6.41)
0.212
(3.45)
0.022
(7.33)
0.063
(1.75)
0.171
(2.00)
0.126
(10.5)
0.126
(5.73)
0.031
(2.58)
0.025
(2.08)
Significant at 90% level of confidence if 1.960 > t ≥ 1.645;
Significant at 95% level of confidence if 2.576 > t ≥ 1.960;
Significant at 99% level of confidence if 2.810 > t ≥ 2.576;
Significant at 99.5% level of confidence if 3.290 > t ≥ 2.810;
Significant at 99.9% level of confidence if t ≥ 3.290.
(Source: Anwar et al., 2011; Anwar et al., in press)
0.927
0.902
0.964
0.911
0.043
0.030 (90% CI of RMSEA)
0.051 (90% CI of RMSEA)
0.071
(3.44)
-0.037
(-3.63)
-0.037
(-3.44)
0.025
(3.13)
0.045
(5.63)

21. Empirical Results
Table 3 Results of random parameter logit models (t-values within the parenthesis)
Attributes
Travel cost (mean)
Travel cost (st.dev.)
Waiting time (mean)
Waiting time (st.dev.)
Age (mean)
Age (st.dev.)
Car ownership (mean)
Car ownership (st.dev.)
Having children (mean)
Having child (st.dev.)
Trip purpose (mean)
Trip purpose (st.dev.)
Comfort (mean)
Comfort (st.dev.)
Convenience (mean)
Convenience (st.dev.)
Safety (mean)
Safety (st.dev.)
Flexibility (mean)
Flexibility (st.dev.)
Reliability (mean)
Reliability (st.dev.)
Satisfaction (mean)
Satisfaction (st.dev.)
TRPL1
TRPL2
Random parameter in utility functions
-3.14 (-2.11)
-3.19 (-2.56)
1.07 (1.99)
1.02 (2.45)
-1.72 (-2.12)
-1.85 (-3.11)
0.08 (3.11)
0.03 (3.41)
-0.22 (-1.89)
0.48 (1.66)
1.84 (3.52)
0.03 (3.51)
-1.78 (-6.44)
0.11 (3.65)
TRPL3
HRPL
-3.20 (-5.55)
1.05 (3.45)
-1.93 (-3.15)
0.004 (2.48)
-0.11 (-1.11)
0.22 (2.01)
1.91 (5.21)
0.02 (4.21)
-1.80 (-5.41)
0.26 (3.11)
0.07(3.44)
0.003 (2.33)
-2.11 (-2.62)
1.06 (4.21)
-1.75 (-3.14)
0.004 (2.99)
-0.09 (-2.84)
0.58 (2.63)
1.89 (4.00)
0.04 (4.44)
-1.77 (-5.02)
0.12 (2.87)
0.06 (2.15)
0.001 (3.63)
3.32 (7.89)
0.12 (5.66)
3.18 (4.66)
0.22 (5.66)
5.18 (11.11)
0.45 (9.84)
0.73 (1.00)
0.30 (2.16)
5.17 (11.10)
0.01 (9.15)
1.23 (2.66)
0.09 (2.99)

22. Empirical Results
Table 3 Results of random parameter logit models (t-values within the parenthesis) (Cont.)
Attributes
TRPL1
TRPL2
TRPL3
Nonrandom parameter in utility functions
Age
-0.08 (-0.99)
Having children under 5 yrs
-0.97 (-3.62)
Car ownership
1.27 (3.91)
Trip purpose
0.97 (2.89)
0.97 (2.91)
Travel time
-1.17 (-7.85) -1.17 (-8.77) -1.19 (-6.42)
Gender
0.29 (1.89)
0.32 (2.13) 0.39 (2.15)
Income
1.32 (1.85)
1.69 (1.11) 1.98 (1.91)
Family size
-0.94 (-0.45) 0.94 (1.01) 0.93 (0.99)
Full time workers of HH
0.97 (0.32)
0.97 (1.45) 0.97 (0.85)
Trip rate
0.91 (1.11)
0.91 (1.00) 0.91 (1.74)
Distance travelled
-0.19 (-1.89) -0.17 (-1.11) -0.78 (-1.01)
Mode constant
Car as a passenger (base)
0
0
0
Car as a driver
-2.22 (-2.45) -2.23 (-2.54) -2.22 (-3.10)
Train
-1.00 (-1.99) -1.17 (-1.98) -2.18 (-3.41)
Bus
-0.11 (-0.52) -0.12 (-1.23) -0.14 (-1.22)
HRPL
-1.11 (-3.63)
0.21 (2.69)
1.50 (0.89)
0.94 (1.00)
0.97 (1.01)
0.91 (1.86)
-0.24 (-1.12)
0
-2.41 (-9.00)
-2.39 (-7.15)
-0.10 (-1.53)

23. Empirical Results
Table 3 Results of random parameter logit models (t-values within the parenthesis) (Cont.)
Attributes
TRPL3
HRPL
-0.12 (-3.62)
-0.54 (-2.96)
-0.08 (-1.98)
0.01 (3.01)
-0.09 (-2.66)
0.01 (4.01)
-0.01 (-3.99)
-0.03 (-3.85)
-0.12 (-2.14)
0.65 (5.14)
-0.17 (-3.01)
0.05 (3.01)
0.09 (3.10)
0.10 (2.89)
0.45 (11.52)
0.05 (2.45)
0.31 (10.20)
0.08 (5.10)
Log likelihood function
McFadden Pseudo R-squared
Akaike Information Criterion (AIC)
Model statistics
-812.41
-768.31
0.21
0.25
0.019
0.018
-715.28
0.27
0.017
-613.37
0.36
0.014
Car as a driver
Car as a passenger
Train
Bus
Modal choice probability
0.713
0.721
0.080
0.075
0.159
0.160
0.048
0.044
0.731
0.055
0.181
0.033
0.785
0.010
0.190
0.015
Travel cost :Income
Waiting time :Income
Age: Income
Car ownership: Income
Having child: income
Purpose: Income
Comfort: Income
Convenience: Income
Safety: Income
Flexibility: Income
Reliability: Income
Satisfaction: Income
TRPL1
TRPL2
Heterogeneity around the mean
-0.11 (-4.21)
-0.10 (-2.98)
-0.54 (-3.56)
-0.54 (-2.56)
-0.11 (-1.89)
0.02 (3.12)
-0.02 (-1.99)

24. Empirical Results
Significant at 90% level of confidence if 1.960 > t ≥ 1.645;
Significant at 95% level of confidence if 2.576 > t ≥ 1.960;
Significant at 99% level of confidence if 2.810 > t ≥ 2.576;
Significant at 99.5% level of confidence if 3.290 > t ≥ 2.810;
Significant at 99.9% level of confidence if t ≥ 3.290.

25. Discussions and Conclusions
• Percentage of car usage is notably high – presence of
agency problem
• Required to understand the lack of awareness about
travellers’ utility
• The HRPL mode is more powerful than the TRPL model
• The LVs dominate the traveller choice process

26. Discussions and Conclusions
• Therefore, traveller choice attributes are the key issues in
the traveller-TfNSW relationship
• The hierarchy of importance of attributes are relevant in the
context of transport policy responses
• This study has clarified the nature of such a policy response
by indicating which attributes of the traveller-TfNSW
relationship are most important to travellers.

27. Discussions and Conclusions
• It is understood that traveller’s preference to mode choice
is a fundamental factor to resolve the agency problem
• Finally, TfNSW needs to be aware of those attributes of
travellers’ choice process that should increase travellers’
utility the most.
• Thus, the maximisation of traveller’s utility helps to rectify
the agency problem.

28. References
•
•
•
•
•
•
•
Anwar, A.H.M.M., Tieu, K., Gibson, P., Berryman, M., & Win, K.T. (2011). Structuring the influence of
latent variables in traveller preference heterogeneity. Proceedings of the 16th International Conference
of Hong Kong Society for Transportation Studies, Hong Kong, 141-148.
Anwar, A.H.M.M., Tieu, K., Gibson, P., Win, K.T. & Berryman J.M. (in press). Analysing the heterogeneity
of traveller mode choice preference using a random parameter logit model from the perspective of
principal-agent theory. International Journal of Logistics Systems and management.
Barney, J.B. & Hesterly, W. (1996). Organizational economics: Understanding the relationship between
organizations and economic analysis. In handbook of organization, C. Stewart, H. Cynthia, and N.
Walter R. (Ed.), London and Thousand Oaks: Sage Publications
Eisenhardt, K.M., (1989). Agency theory: An assessment and review. Academy of Management Review,
14(1), 57-74.
Johansson M.V, Heldt, T., & Johansson, P. (2006). The effects of attitudes and personality traits on
mode choice. Transportation Research Part A: Policy and Practice, 40(6), 507-525.
Raveau, S., Alvarez-Daziano, R., Yanez, M.F., Bolduc, D. and de Dios Ortuzar, J. (2010) ‘Sequential and
simultaneous estimation of hybrid discrete choice models: some new findings’, Transportation
Research Record, No. 2156, pp.131-139.
Rungtusanatham, M., Rabinovich, E., Ashenbaum, B. & Wallin, C. (2007). Vendor-owned inventory
management arrangements in retail: an agency theory perspective. Journal of Business Logistics, 28(1),
111-35.

29. A.H.M. Mehbub Anwar
Kiet Tieu
Peter Gibson
Khin Than Win
Matthew J. Berryman
PhD Student
Professor
Associate Professor
Senior Lecturer
Senior Research Fellow
Email: ahmma324@uowmail.edu.au
Email: ktieu@uow.edu.au
Email: peterg@uow.edu.au
Email: win@uow.edu.au
Email: mberryma@uow.edu.au