You are on page 1of 16

Methods of Eliciting Response

• Rating
• on an arbitrary scale (between 1 and 5 or 1 and 10)
• Likert scale: 1-Strongly disagree; 2-Disagree; 3-Neutral; 4-Agree; 5-Strongly agree
• Depending on the model employed, the starting numerical value of the rating scale may be
either 0 or 1. (i.e., [0, 4]/[1, 5] or [0, 9]/[1, 10])
• Ranking
• Choice
• binary choice or group of alternatives
• Choice cum Rating
Rating Experiment
• One of the ways of eliciting
response to each of the
0 options is by asking the
8
respondent to rate on a scale
of 0 to 4 or 0 to 9.
2 • Respondent rates each option
1 based on an existing option
Rate this option based on your current trip which they are currently
Travel Time Travel Cost Waiting Time Transfers experiencing, e.g., the journey
Excellent that they were making.
Good 20 minutes ₹ 15 5 minutes Nil
• Each of the option appears on
Fair a card along with rating scale.
Poor Tick the preferred rating box
• In this example the options 1
Very Poor4 3 2 1 0
to 8 that we designed are
Excellent Good Fair Poor Very Poor printed on cards numbered 1
to 8. The card with number “0”
has the details of the existing
option which the respondent
uses for comparison.
Ranking Experiment

8 • Another way of eliciting


0 response is by asking the
respondent to arrange the
7 option cards in descending
order of their preference.
1
• In this process the respondent
Journey Characteristics by Bus ranks the options, with the
option on the top receiving
Excellent Travel Time: 20 minutes rank 1.
Good Travel Cost: ₹ 15
Fair • The existing option that is
Waiting Time: 5 minutes being currently experienced
Poor
(indicated here by card “0”) by
Very Poor Transfers: Nil the respondent is also
included as one of the options
in this ranking exercise.
Choice Based SP Experiment

8 • In a choice-based experiment,
the respondent choses one
option among the options made
2 available to her/him
1 • For example, the respondent is
Which alternative you choose? provided every time with two
alternatives printed on a card as
 shown
• Each of the designed hypothetical
CAR BUS option is compared with the
existing option currently being
Travel Time 25 minutes Travel Time 20 minutes experienced by the respondent.
Travel Cost ₹ 20 Travel Cost ₹ 15 • The respondent will choose the
attractive option between the
Waiting Time 0 minutes Waiting Time 5 minutes hypothetical and existing, each
time.
Transfers Nil Transfers Nil
Choice cum Rating SP Experiment
• In this kind of experiment, each
hypothetical option is
8
compared with the current
option experienced by the
2 respondent, making it a binary
1 choice experiment
Current Trip BUS
• Instead of choice, the
Travel Time 25 minutes Travel Time 20 minutes likelihood of preferring the
Travel Cost ₹ 20 Travel Cost ₹ 15 hypothetical option is collected
on a qualitative scale as shown
Waiting Time 0 minutes Waiting Time 5 minutes
• For modelling purpose
Transfers Nil Transfers Nil
probability of choosing the
How likely are you to choose Bus in the above context? hypothetical option is
Definitely Probably Can’t say Probably not Not at all interpreted as:
• Definitely: 0.9
• Probably: 0.7
• Can’t say: 0.5
• Probably not: 0.3
• Not at all: 0.1
Sampling Strategy
• Type of sampling
• Random: a sampling technique where every person in the population has an
even chance of being selected in the sample.
• Stratified: The population is divided into subgroups or strata based on a
socioeconomic characteristic (e.g. income) and samples are drawn from each
strata following random sampling method
• Choice based: This is a type of stratified sampling, wherein the stratification is
done based on the choice made. For example, the travellers can be stratified
as Car users, Train users, Bus users and random samples can be drawn from
each of these strata.
• SP studies require smaller sample compared to RP studies
• 75 to 100 samples per segment is the minimum size required
Modelling with Stated Preference Data

• Rating Data
• Modeled using ordered choice models

• Ranking Data
• Modeled using exploded MNL

• Choice and choice cum rating data


• Modeled using MNL
Ordered Choice Model for Rating Data
• The discrete choice models viz., MNL, NL and
RPL that we discussed so far dealt with choices
among options which are unordered.
• In stated preference rating experiment that we
have just learnt about, the options of
improved bus system are rated on a scale of 0
to 4. 
• One of the rated options is shown here
• In view of the rating, the options get ordered
or categorised. Such ordered choices can be
modelled using ordered choice models
Rating Data PERNUM IVTT FARE WT RATING
Obtained from 1
1
20
30
15
15
5
5
4
3
the SP 1 20 25 5 4
Experiment 1
1
30
20
25
15
5
10
2
4
1 30 15 10 1
1 20 25 10 3
1 30 25 10 0
2 20 15 5 4
2 30 15 5 3
2 20 25 5 4
Note: 2 30 25 5 2
PERNUM: Person number 2 20 15 10 1
IVTT: In Vehicle Travel Time 2 30 15 10 1
WT: Waiting Time 2 20 25 10 3
2 30 25 10 0
. . . . .
. . . . .
. . . . .
Theory of Ordered Choice Model
• As in random utility models, let Ui* be the utility as perceived by individual i for an option, i.e.,
Ui* =  ‘ Xi + εi,
Where,  = column vector of parameters
Xi = column vector of explanatory variables
εi = error term
• In this SP experiment, instead choice information, the rating, Ri of an option is obtained from individual i
• As we have only the rating, the underlying utility is translated to a rating. This in statistics is known as
censoring, i.e., censoring of the underlying utility.
Ri = 0 if μ-1 < Ui* < μ0,
Ri = 1 if μ0 < Ui* < μ1,
Ri = 2 if μ1 < Ui* < μ2,
Ri = 3 if μ2 < Ui* < μ3,
Ri = 4 if μ3 < Ui* < μ4
Where, μ-1, μ0, μ1, μ2, μ3, and μ4 are the threshold parameters
In general, Ri = j if μj-1 < Ui* < μj
Calibration of Ordered Choice Model Contd.
Prob [Ri = j] = Prob[μj-1 < Ui* < μj]
= Prob[μj-1 < ( ‘ Xi + εi) < μj]
= Prob[(μj-1 - β′xi) < εi < (μj - β′xi)]
= Prob [εi < (μj - β′xi)] – Prob [εi < (μj-1 - β′xi)],
j = 0, 1, ..., J.

Probability
Prob[Ri = j] = F[μj - β′xi] - F[ μj-1 - β′xi]

• Log likelihood of ordered choice model:


ln[F(μj - β′xi) – F( μj-1 - β′xi)]
where = 1 if Ri = j and 0 otherwise.

μj - β′xi
εi

μj-1 - β′xi
• Maximization is done for estimating β′ and thresholds μ
subject to the constraints μ-1 = -∞, μ0 = 0 and μJ = +∞.

• In the case of ordered logit model, the error term, εi is distributed as per logistic distribution.
• In the case of ordered probit model, the error term, εi is distributed as per normal distribution.
• For Ordered Probit Model, error term is distributed as per standard normal
distribution:

• For Ordered Logit Model, error term is distributed as per logistic distribution
Rating Data PERNUM IVTT FARE WT RATING
Obtained from 1
1
20
30
15
15
5
5
4
3
the SP 1 20 25 5 4
Experiment 1
1
30
20
25
15
5
10
2
4
1 30 15 10 1
1 20 25 10 3
1 30 25 10 0
2 20 15 5 4
2 30 15 5 3
2 20 25 5 4
Note: 2 30 25 5 2
PERNUM: Person number 2 20 15 10 1
IVTT: In Vehicle Travel Time 2 30 15 10 1
WT: Waiting Time 2 20 25 10 3
2 30 25 10 0
. . . . .
. . . . .
. . . . .
Load the file containing the data of rating experiment from the folder of the file
using the menu path Project->Import->Variables
Open Text/Command Document from the menu path File->New and enter the command
ORDERED ;Lhs = RATING ;Rhs = ONE, IVTT, FARE, WT $ for Ordered Probit or
ORDERED; Lhs = RATING ;Rhs = ONE, IVTT, FARE, WT ;Logit $ for Ordered Logit and Run the same
NLOGIT Command for Ordered Choice Model
• For calibrating ordered choice models on the rating data of stated preference
experiment shown in the beginning, following commands are to be written in the
command document of NLOGIT:
• For Ordered Probit write the following:
ORDERED ;Lhs = RATING ;Rhs = ONE, IVTT, FARE, WT $
• For Ordered Logit write the following:
ORDERED; Lhs = RATING ;Rhs = ONE, IVTT, FARE, WT ;Logit $

You might also like