Professional Documents
Culture Documents
MMB2020Exam PDF
MMB2020Exam PDF
Laboratoire TRANSP-OR
Name
Signature Section
Question Points
1 20
2 20
3 20
4 20
Total Grade
This exam is written and lasts 3 hours, from 08:15 to 11:15. The only material
that you are allowed to use is the handwritten summary, maximum length of 4
pages (2 double-sided A4 sheets or 4 single-side A4 sheets). The summaries will
be collected at the end along with the exam. Make sure that your name and the
date are mentioned on every page of the exam and on the summary. You shall
answer in English. All answers have to be carefully justified.
2
Question 1
(20 points) Consider the following model (M1 ) for the London Passenger Mode
Choice case study, which involves four alternatives: walking, cycling, public trans-
port (pt), and driving.
The utility functions are defined as Ui,n = Vi,n + εi,n , where i denotes the alter-
iid
native, n the trip, εi,n ∼ EV(0,1) and Vi,n are defined as follows:
costi,n is the travel cost in GBP associated with alternative i for trip n, durationi,n
is the travel time in hours of alternative i for trip n, and interchangesn is the
number of interchanges on the public transport route for trip n. The estimates
for the parameters can be found in Table 1.
1. What does the term εi,n capture in the utility function? [1 point]
3. What statistical test could be used with M1 to verify whether βtime varies
significantly between two alternatives without estimating another model ?
What is the null hypothesis that is tested? [1 point]
3
4. Explain (without calculation) how the Value of Time could be estimated
using the values in Table 1. What are the units? [1 point]
5. What is the perceived cost (in GBP) of making an interchange on the public
transport route? On average, how many minutes of journey time will a trip
with one interchange need to save to be more attractive than an otherwise
identical route with no interchanges? [2 points]
Consider the following proposed models (M2 to M5 ), which are all modifications
of the base model M1 .
M2 :
where distancen is the straight-line distance between the start-point and the end-
point for trip n.
M3 :
where ivttpt,n and ovttpt,n are the in-vehicle travel time (total time on-board bus/
rail/tram services) and the out-of-vehicle travel time (total time walking, waiting,
interchanging etc.) respectively for the public transport route for trip n, so that
durationpt,n = ivttpt,n + ovttpt,n .
4
M4 :
M5 :
0
Vwalk,n = βtime,walk ln(durationwalk,n ) ,
0
Vcycle,n =ASCcycle +βtime,cycle ln(durationcycle,n ),
0
Vpt,n =ASCpt +βtime,pt ln(durationpt,n ) + βcost costpt,n
+βinterchanges interchangesn ,
0
Vdrive,n =ASCdrive +βtime,drive ln(durationdrive,n ) + βcost costdrive,n .
Answer the following questions for each model specification. Note: you can
either order your answers by model (i.e., answer all questions for model M2 ,
followed by M3 , etc.) or by question (i.e., answer question 6 for all models,
followed by question 7, etc.).
6. For each model, identify the behavioural assumptions the modified specifi-
cation captures, compared with the base model M1 . [4 points]
7. Some of these models are not correctly specified. For each model, identify if
there are any normalization or specification issues. If there are issues with
a model, state each issue, and explain how to modify the specification to
fix it. [4 points]
8. State a significance test that could be used to test each model (including
any modifications suggested in Q1.7) against M1 . What is/are the
explicit null hypothesis/hypotheses that is/are tested? How many degrees
of freedom are there for each hypothesis? [6 points]
5
6
7
8
Question 2
(20 points)
You are a student at EPFL and are considering your lunch options. You can
choose between the following destinations d ∈ D ={Parmentier, Esplanade, food
trucks, Holy Cow and Migros}.
Parmentier, Esplanade and food trucks are located on campus, while Holy Cow
and Migros are off campus. Migros and food trucks only allow for take-away,
while all other destinations offer a sitting down option (for simplification, con-
sider that you cannot take-away from the other destinations).
Based on the same problem, the following generative function has been formu-
lated.
C
S D
!µc /µs µ/µc
X X X µ s
G(yd ) = yd s
c=1 s=1 d=1
9
9. What are the 3 conditions that G must verify to be a valid MEV-function?
Verify 2 of them. [3 points]
10. What model can be derived from G? Is it appropriate to capture all corre-
lations present in the problem? [1 point]
10
11
12
Question 3
1. What sampling strategy was used to collect the sample? Justify your answer
[1 point]
2. Fill the gaps (a)-(n) in Tables 2 and 3. Assume that the proportions with
respect to the transportation mode are kept for each age segment between
both tables. Make sure to justify how each value is obtained. [5 points]
Age
Mode Young Adult Retired
(a) (d)
car 1000 5750
(b) (e)
PT 2000 4250
(c) (f)
3000 10000
Age
Mode Young Adult Retired
(g) (j) (m)
car 30
(h) (k) (n)
PT 15
(i) (l)
45 100
We have estimated two models on the sample: an unweighted logit model and a
weighted logit model using Weighted Exogenous Sampling Likelihood (WESML).
13
5. Which estimation should be preferred and why? [1.5 points]
Assume now that the same questionnaire is carried out each weekday of a certain
week during the same time window. The utility specification associated with each
individual n and day d is the following:
where tind and cind represent the travel time and cost associated with alternative
i, individual n and day d, respectively, Yn is a binary variable that takes value 1
if individual n is young and 0 otherwise, An is a binary variable that takes value
1 if individual n is adult and 0 otherwise, and Rn is a binary variable that takes
value 1 if individual n is retired and 0 otherwise. The error terms εind associated
with alternative i, individual n and day d are assumed to be independently and
identically distributed ∼ EV(0,1).
6. How can (1)–(2) be modified to take into account serial correlation with a
fixed effect? Label the resulting utility functions as (3) and (4) for car and
PT, respectively. Make sure to describe any notation you might introduce
and the associated assumptions. Why is the resulting model not consistently
estimated in this case? [2 points]
7. How can (3)-(4) be modified to incorporate the influence of the choice made
the previous day? Label the resulting utility functions as (5) and (6) for car
and PT, respectively. Make sure to describe any notation you might intro-
duce and the associated assumptions. What happens to the observations
collected during the first day? [1.5 points]
Consider again the logit model defined by (1)–(2) and individual n = 1, who is
an adult with the following data for the car alternative:
Answer the following questions. Make sure to simplify the resulting mathematical
expressions as much as possible.
8. Knowing that travel time and travel cost for PT do not change between
d = 1 and d = 5, express the probability associated with car for n = 1 and
d = 5 exclusively as a function of the probability associated with car for
n = 1 and d = 1. [2.5 points]
14
The municipality wants to use the sample to estimate the revenue generated by
the car alternative.
10. Assume that 60% of the cost associated with the car alternative is col-
lected by the municipality in the form of a congestion toll. What is the
expected revenue that will be obtained in one day? Make sure to describe
any notation you might introduce. [1.5 points]
15
16
17
Question 4
(20 points) Consider a stated preference airline itinerary choice dataset that
involves three alternatives: (1) non-stop flight, (2) one-stop flight without airline
change and (3) one-stop flight with airline change. A logit model is developed
based on this dataset. The three utility functions are defined as Ui,n = Vi,n + εi,n ,
iid
where i ∈ {1, 2, 3} denotes the alternative, n the individual, εi,n ∼ EV(0, 1) and
the Vi,n are defined as follows:
where farei,n , timei,n and legroomi,n are the total cost, trip time and legroom
(measured in centimeters) associated with alternative i and individual n, respec-
tively. We call this model the base model from now on.
(a) Rewrite βfare as a transform of ξ ∼ N (0, 1). Make sure to describe any
notation you may use. [1 point]
(b) List the parameters to be estimated in the MXT model. [1 point]
(c) Explain why considering a normal distribution for βfare may be inap-
propriate. Mention another distribution that could be used and justify
your answer. [2 points]
(d) Write the formula of the contribution of observation n to the log-
likelihood function L(in |Xn , θ), where in is the alternative chosen by
individual n, Xn is the list of variables entering the utility functions
and θ is the list of parameters identified in (b). [1 point]
18
where svac and swork denote the two considered latent classes. The resulting
choice model is referred to as the LC1 model. We ask you to perform the
following tasks:
(a) Explain how to compute the probability Pn (i) that individual n chooses
alternative i given the considered class membership model. [2 points]
(b) List the parameters to be estimated in the LC1 model. [1 point]
(c) Name a statistical test that could be used to compare the LC1 model
with the base model. Justify your answer and state the null hypothesis
of the test. [2 points]
3. We now want the class membership model to take into account some so-
cioeconomic characteristics of the individuals with the goal of improving
its predictive power. The probabilities of belonging to each of the latent
classes are therefore given as:
1 1
Pr(n ∈ svac ) = , Pr(n ∈ s work ) = ,
1+e n ω 1 + e−ωn
with
ωn = γ0 + γalone alonen + γage agen + γincome incomen ,
where alonen is a binary variable that equals 1 if individual n travels on her
own and 0 otherwise, and agen and incomen are the age and yearly income
of that same individual, respectively. The resulting model is referred to as
the LC2 model. We ask you to perform the following tasks:
(a) Suppose γalone , γage and γincome are estimated to positive values. Inter-
pret these results. [1 point]
(b) Draw a diagram representing the LC2 model. Make sure to explain
the drawing convention you used, i.e., what the different shapes and
arrows mean. [5 points]
(a) Mention two reasons why attitudinal indicators cannot be used as ex-
planatory variables in the utility function. [1 point]
(b) Update your diagram such that it represents the LC3 model. [1 point]
(c) Propose two additional statements that could have been included in
the survey as psychometric indicators to differentiate vacationers from
workers. Justify your answers. [2 points]
19
20
21
22