Lecture 7 Stated Preference Methods - National Chiao …ocw.nctu.edu.tw/course/ests021/Lect7.pdfof...
-
Upload
nguyenmien -
Category
Documents
-
view
216 -
download
0
Transcript of Lecture 7 Stated Preference Methods - National Chiao …ocw.nctu.edu.tw/course/ests021/Lect7.pdfof...
1
Lecture 7
Stated Preference Methods
Cinzia Cirillo
2 2
Preference data • Revealed Preferences RP Respondents are questioned about what
they actually do. RP data contain information about
current market equilibrium. Historically economists rely on real
market data because a classical concept affirms that only RP data have thus and such properties to estimate demand equations consistent with market behavior.
• Stated preferences SP Respondents are faced to hypothetical
choice situations. SP data provides insights into problems
involving shifts in technological frontiers.
There are many situations in which analysts and researchers have little alternative to take consumers at their world or do nothing.
3 3
Why SP data? • Organizations need to estimate demand for new products with new attributes
or features. By definition, such applications have no RP data on which to rely, managers face the choice of guessing or relying on well-designed and executed SP research.
• Explanatory variables have little variability in the marketplace. Even if products have been in the market for many years, it is not uncommon for there to be little or no variability in key explanatory variables.
• Explanatory variables are highly collinear in the marketplace. Cost and Time correlation. Technology constraints.
• New variables are introduced that now explain choices. As a product categories grow and mature, new product features are introduced and/or new designs supplant obsolete ones.
4 4
• Observational data cannot satisfy model assumptions and/or contain statistical “nasties” which lurk in real data. All models are only as good as their maintained assumptions. RP data may be of little value when used to estimate the parameters assumptions.
• Observational data are time consuming and expensive to collect. Very often RP data are expensive to obtain and may take considerable time to collect. For example panel data involve observations of behavior at multiple points in time for the same or independent samples of individuals.
• The product is not traded in the real market. Many goods are not traded in real economic markets; for example, environmental goods, public goods such as freeways or stadia. Yet society and its organizations often require that they be valued, their costs and benefits calculated.
5 5
RP data typically: • Depict the world as it is now (current
market equilibrium). • Possess inherent relationship
between attributes (technological constraints are fixed).
• Have only existing alternatives as observables.
• Embody market and personal constraints on the decision maker.
• Have high reliability and face validity, • Yield one observation per
respondent at each observation point.
SP data typically: • Describe hypothetical or virtual decision
contexts (flexibility). • Control relationship between attributes,
which permits mapping of utility functions with technologies different from existing ones.
• Can include existing and/or propose and/or generic (unbranded or unlabelled) choice alternatives.
• Cannot easily (in some cases cannot at all) represent changes in market and personal constraints effectively.
• Seem to be reliable when respondents understand, are committed and can respond to tasks.
• Usually yield multiple observations per respondent at each observation point.
6 6
Preferences 1. Discrete choice of one option from a set of competing ones. This response
measures the most preferred option relative to the remaining, but provides no information about the relative preferences among the non-chosen. That is a true nominal scale.
2. ‘Yes, I like this option’ ‘No, I don’t like this option’. This response clearly separates alternatives into liked and not liked options and provides preferences.
3. Complete ranking of options from most to least preferred. This response orders all options on a preference continuum, but provides no information about degree of preference, no order.
4. Rating options on a scale. Expresses degrees of preference for each option by rating them on a scale or responding via other psychometric methods such as magnitude estimation. If the consumers can supply valid and reliable estimates of their degree of preference this response contains information about equality, order and degrees of differences and magnitude.
7 7
Discrete choice of one option from a set of competing ones
Mode for journey to work Consumer chooses
Take bus Take train Take ferry Drive own auto Carpool
X
Auto > bus, train, ferry, carpool and bus = train = ferry = carpool
8 8
‘Yes, I like this option’ ‘No, I don’t like this option’
• Auto > bus, train, ferry • Carpool > bus, train, ferry • Auto = carpool; bus = train = ferry
Mode for journey to work
Consumer will consider (y/n)
Take bus Take train Take ferry Drive own auto Carpool
No No No Yes Yes
9 9
Complete ranking of options from most to least preferred
• Auto > bus, train, ferry, carpool • Carpool > bus, train, ferry • Ferry > bus, train • Train > bus
Mode for journey to work
Ranking by likelihood of use
Take bus Take train Take ferry Drive own auto Carpool
5 4 3 1 2
10 10
Expressing degrees of preference by rating options on a scale
• Auto > bus, train, ferry, carpool • Carpool > bus, train, ferry • Ferry > bus, train • Train = bus
Mode for journey to work
Consumer likelihood to use (y/n)
Take bus Take train Take ferry Drive own auto Carpool
4 4 6
10 7
11
Part II: Experimental Design
12
Definitions • An experiment involves the manipulation of a variable with one or more observations, taken
in response to each manipulated value of the variable. • The manipulated variable is called “factor”, and the values manipulated are called “factor
levels”. • Such variables are also referred to as independent or explanatory variables or “attributes”.
• Factorial designs are designs in which each level of each attribute is combined with every
level of all other attributes. • The complete enumeration is called a “complete factorial” or a “full factorial”. Complete
factorial guarantees that all attribute effects of interest are truly independent.
13
Choice experiments consist of a sample of choice sets selected from the universal set of all possible choice sets that satisfy certain statistical properties.
There are two general types of choice experiments: 1. labelled (alternative-specific) 2. unlabbeled (generic) There are two general ways to design choice experiments for both types: 1. Sequentially design alternatives and then design the choice sets into which
there are placed; 2. Simultaneously design alternatives and assign them to choice sets.
14
Multiple choice experiments The objective of multiple choice experiments is to design alternatives and the choice sets in which they appear, such that the effect can be estimated with reasonable levels of statistical precision. Multiple choice experiments: 1. There are more than two alternatives (two brands and non-choice, eight
brands, etc) and 2. Choice set sizes may vary (some sets with two brands, some with eight, etc. Design issues involve the following types of alternatives: (a) labelled vs. unlabelled; (b) generic vs. alternative-specific; (c) own vs cross-effects.
15
Designs for MNL models • Design an initial set of P total alternatives (profiles) to create choice sets
containing one or more additional alternatives M. • Make M-1 copies of the initial set of P total profiles, and place the M sets of
profiles in M separate urns. Randomly select one of the P profiles from each of the M urns without replacement to construct a choice set of exactly M alternatives, ensuring that none of the M profiles in the set are the same. Continue this process until all P profiles in each urn have assigned to P total choice sets of M alternatives.
16
• Improve the statistical efficiency of the first procedure by creating M different, statistically equivalent designs. In this case each urn contains a different design. When one randomly draws profiles from the M urns to make the P total choice sets, one does not have to eliminate duplicate profiles.
• Further improve design efficiency by first constructing the P total profiles and then constructing the P total choice sets by a method known as shifting, in which modular arithmetic is issued to shift each combination of the initial attribute levels by adding a constant that depends on the number of levels.
• Make P initial profiles and construct all possible pairs of each. There will be exactly P(P-1)/2 pairs. The total number of pairs will increase geometrically with P.
17
Designs for availability problems • Many problems involve sets of alternatives that vary in nature and
composition. In transport, it is rare for commuters to have all transport modes available for their commuters. If IID is satisfied, label specific intercepts for J-1 alternatives can be estimated by designing this type of experiments.
• Each of the J labels can be treated as a two level variable (present/absent). A nearly optimally efficient strategy is to design the choice sets using a 2J
fractional factorial design. • If IID is violated a minimum strategy is to design the smallest orthogonal a 2J
main effects plus its foldover (a mirror image of the original design; replace each 0 with 1 and each 1 with 0).
18
Set United Delta Northwest US Airways Southwest
1 P P P P P
2 P P A P A
3 P A P A A
4 P A A A P
5 A P P A P
6 A P A A A
7 A A P P A
8 A A A P P
19
• Each airlines appears equally often (count the number of A and P in each column).
• The presence/absence of each airline in independent of the presence/absence of other airline.
• If two events are probabilistically independent their joint probabilities should equal the product of their marginals (4x4)/8 = 2. The correlation of the co-occurrances is exactly zero.
Airline A
Airline B Present Absent
Present 2 2
Absent 2 2
20
• The marginal for each airline can be estimated independently of the marginals of every other airline.
• The marginal of each airline is the best estimate of the alternative-specific intercept or constant in MNL model.
• Alternative-specific intercepts can be estimated from several data aggregation levels, and each will yield the same coefficients up to a multiplication by a positive constant.
• The more one aggregate data, the more one hides individual and choice set variation.
• Thus it is particularly dangerous to aggregate data over subjects because consumers typically exhibits heterogeneous preferences.
21
Unlabelled, generic alternatives The choice outcomes are purely generic in the sense that the labels attached to each option convey
no information beyond that provided by the attributes. Options are simply labelled “A” and “B”.
Option A Option B Set Fare Service Time Fare Service Time 1 2 3 4 5 6 7 8
$1.20 $1.20 $1.20 $1.20 $2.20 $2.20 $2.20 $2.20
5 5
15 15 5 5
15 15
10 20 10 20 10 20 10 20
$ 2.00 $ 2.00 $ 3.00 $ 3.00 $ 3.00 $ 3.00 $ 2.00 $ 2.00
15 30 30 15 30 15 15 30
15 30 30 15 15 30 30 15
22
M = total generic choice outcomes A = total attributes L = levels for each attribute The collective design is an LMA factorial, from which one selects the smallest
orthogonal main effects plan. For example, if there are four choice outcomes, and each is described by eight
four level attributes, the collective factorial is 48x4, or 432. The smallest possible main effect plan is determined by the total degrees of freedom required to estimate all implied main effects.
The total degrees of freedom are determined by assuming the separate degree of freedom in each main effect.
Each main effect has exactly L - 1 degree of freedom (= 3 in the present example).
23
There are 32 main effects (4 x 8 attributes); hence there is a total of 32 x 3, or 96 degrees of freedom. The smallest orthogonal main effects plan requires 128 choice sets.
Unbalanced designs are those for which • Attributes have unequal numbers of levels • The numbers of levels are not multiples of one another. Hensher and al. say: For example if three attributes have levels, respectively of 2, 3 and 4 the design
properties will be unbalanced. If the tree-level attribute can be reduced to two or increased to four levels, design properties will be improved.
24
No of options No of attributes No of levels Full factorial Smallest design
2 2 2 2 4 4 4 4 8 8 8 8 16 16
4 4 8 16 4 4 8 16 4 4 8 16 4 8
2 4 2 4 2 4 2 4 2 4 2 4 2 4
28
48
216
432
216
416
232
464
232
432
264
4128
264
4128
16 sets 32 sets 32 sets
128 sets 32 sets 64 sets 64 sets
256 sets 64 sets
128 sets 128 sets 512 sets 128 sets 512 sets
25
Labelled alternatives The design principle for unlabelled alternatives also apply to designs for labelled
alternatives. The key difference is that the label or name of the alternative itself conveys
information to decision makers. This matters in choice decisions because: • Subjects may use labels to infer missing (omitted) information; • These inferences may be (and usually are) correlated with the random
components. The omitted variable bias can be quite serious. For example, significant differences in price effects will occur to the extent that
consumers associate good or bad omitted variables with brands.
26
Good inferences lead to apparently lower price sensitivity, whereas bad inferences lead to higher price sensitivity.
Such apparent effects are driven by failure to include in the task all the relevant information on which consumers base their choices.
Models estimated from such tasks will be of limited value for future forecasting if the covariance structure of the omitted variables changes.
Such changes should be slower in established, mature product markets, but may be rapid in new and emerging markets.
27
Statistical properties of labelled choice experiments Two statistical properties are of interest in labelled and unlabelled choice
experiments: • Identification, that refers to the type of utility and choice process specifications
that can be estimated; • Precision, that refers to the statistical efficiency of the parameters estimated
from the experiment. Specification is, in principle, under the researcher’s control. In practice, an experiment may be too large for practical application. The real issue is precision, that is a function of the number of non-zero attributes
level differences (continuous attributes) or contrasts (qualitative attributes).
28
Difference design • Difference designs requires one to begin with an initial set of profiles. An
additional M choice alternatives can be designed by using an orthogonal difference design.
• Let all attributes be quantitative and let L = 4. Let the levels of each attribute in the difference design be -3 -1 +1 +3.
• If the original price levels are $5, $7, $9, $11, • The price levels of the second alternative would be: • 5±1,3; 7±1,3; 9±1,3; 11±1,3; ($2, $4, $6, $8, $10, $12, $14) • The resulting design will be orthogonal in its attribute level differences, but
will not be orthogonal in the absolute attribute levels.
29
A labeled experiment with constant third option
• All attribute columns of all alternatives are treated as a collective factorial, and a constant, reference alternative is added to each choice set. Given M options, each described by A attributes with L level, the collective factorial is an LMA. One selects the smallest orthogonal design from this factorial that satisfies the desired identification properties. Each choice set is a row in this fractional factorial design matrix to which a constant is added. The constant can be a fixed attribute profile or an option such as “no choice”. The subtraction of a constant from each attribute column leaves design orthogonality unaffected.
30
Constant reference alternative is added to each choice set One selects the smallest orthogonal design from this factorial that satisfies the
desired identification properties. Each choice set is a row in this fractional factorial design matrix to which a constant is added.
This strategy has limitations: 1. A significant number of between-alternative attribute differences will be zero. 2. Some choice sets will contain dominant alternatives 3. Relatively large number of choice sets will be required.
31
Example of a labeled design and resulting attributes differences
26 factorial; six attributes each with 2 levels of variations two zero differences; correlation service frequency- travel time = 0.474
Commuter train City bus Attribute differences
set 1-way Freq Time 1-way Freq Time 1-way Freq Time
1 $1.20 5 10 $2.00 15 15 -0.80 -10 -5
2 $1.20 5 20 $2.00 30 30 -0.80 -25 -10
3 $1.20 15 10 $3.00 30 30 -1.80 -15 -20
4 $1.20 15 20 $3.00 15 15 -1.80 0 +5
5 $2.20 5 10 $3.00 30 15 -0.80 -25 -5
6 $2.20 5 20 $3.00 15 30 -0.80 -10 +5
7 $2.20 15 10 $2.00 15 30 +0.20 0 -5
8 $2.20 15 20 $2.00 30 15 +0.20 -15 -10
32
A labeled experiment with constant third option
Commuter train City bus Option
set 1-way Freq Time 1-way Freq Time Choose another mode of travel to work 1 $1.20 5 10 $2.00 15 15
2 $1.20 5 20 $2.00 30 30
3 $1.20 15 10 $3.00 30 30
4 $1.20 15 20 $3.00 15 15
5 $2.20 5 10 $3.00 30 15
6 $2.20 5 20 $3.00 15 30
7 $2.20 15 10 $2.00 15 30
8 $2.20 15 20 $2.00 30 15
33
Attributes level differences resulting from random design
• Use separate designs to make profiles for train and bus, put the bus and the train profiles in two different urns and generate pairs by randomly selecting a profile from each urn without replacement.
• In this case there are no zero differences and correlation between service frequency and travel time differences is 0.16. This randomly generated design is more efficient that an orthogonal design but this cannot be generilazed.
34
Attributes level differences resulting from random design
23 x 23 factorial; no zero differences; correlation service frequency- travel time = 0.16
Commuter train City bus Attribute differences
set 1-way Freq Time 1-way Freq Time 1-way Freq Time
1 $1.20 5 10 $3.00 15 30 -1.80 -10 -20
2 $1.20 5 20 $2.00 15 30 -0.80 -10 -10
3 $1.20 15 10 $3.00 30 15 -1.80 -15 -5
4 $1.20 15 20 $2.00 30 15 -0.80 -15 +5
5 $2.20 5 10 $2.00 15 15 +0.20 -10 -5
6 $2.20 5 20 $3.00 15 15 -0.80 -10 +5
7 $2.20 15 10 $2.00 30 30 +0.20 -15 -20
8 $2.20 15 20 $3.00 30 30 -0.80 -15 -10
35
Availability designs for labelled alternatives Sometimes we need to generate designs with choice sets of variable size. This
applies to the following situations: Out of stock. How do supply interruptions or difficulties affect choices? Closure or service interruptions. How to travelers change their behavior when a
bridge or a road is closed? New product introductions. How do choices change in response to new entrants
that may or may not be included? Retention/switching. How do choices change in response to systematic changes in
availability? This is very well adapted to study dynamics in behavior.
36
In the case in which presence/absence of options varies but not attributes, designs can be created by treating alternatives as two level factors (present/absent) and selecting orthogonal fractions from the 2J factorial.
Set Option1 Option 2 Option 3 Option 4 Option 5 Option 6
1 2 3 4 5 6 7 8
P P P P A A A A
A A P P A A P P
P A P A P A P A
A A P P P P A A
P A A P P A A P
A P P A P A A P
37
Alternatives vary in availability and attributes
• Two design approaches are possible: 1. An orthogonal fraction of a 2J design is used to design presence/absence conditions
and designed attributes profiles are randomly assigned without replacement to make choice in each condition.
2. A fraction of a 2J design is used to design presence/absence conditions, and a second orthogonal fraction of the collective factorial of the attributes of the alternative “present” is used to make the choice sets in each present/absent condition
38
Attribute availability nesting based on fractional design
Set no. A B C
Condition 1 (011): based on the smallest fraction of the 26
1 A 000 000
2 A 001 011
3 A 010 111
4 A 011 100
5 A 100 101
6 A 101 110
7 A 110 010
8 A 111 001
39
Set no. A B C
Condition 2 (101): based on the smallest fraction of the 26
1 000 A 000
2 001 A 011
3 010 A 111
4 011 A 100
5 100 A 101
6 101 A 110
7 110 A 010
8 111 A 001
40
Set no. A B C
Condition 3 (110): based on the smallest fraction of the 26
1 000 000 A
2 001 011 A
3 010 111 A
4 011 100 A
5 100 101 A
6 101 110 A
7 110 010 A
8 111 001 A
41
Overview
• Will present a few examples of stated preference surveys – Maryland Vehicle Preference Survey – Capitol Beltway HOT Lane Study
• Show survey progression from trial to first run for vehicle preference survey with focus on new Fuel Technology Experiment
• Focus on Departure Time Experiment for HOT study
42
Maryland Vehicle Preference Survey
Sources (abbreviated) Cirillo, C. and Maness, M. Estimating Demand for New Technology Vehicles. ETC 2011 Maness, M. and Cirillo, C. Measuring and Modeling Future Vehicle Preferences: A Preliminary Stated Preference Survey in Maryland. forthcoming
43
Objective
• Objectives – Collect data on future household vehicle preferences in Maryland in
relation to vehicle technology, fuel type, and public policy – Determine if respondent could make dynamic vehicle purchase
decisions in a hypothetical short- to medium-term period – Determine if results from this hypothetical survey could be modeled
using discrete choice methods
44
Survey Design
• Respondent and Household Information • Current Vehicle Properties • Stated Preference Survey
– One of the following: • Vehicle Technology Experiment • Fuel Type Experiment • Taxation Policy Experiment
45
Survey Methodology Time Frame Summer – Fall 2010 Target Population Suburban and Urban Maryland Households Sampling Frame Households with internet access in 5 Maryland counties Sample Design Multi-stage cluster design by county and zipcode Use of Interviewer Self-administered Mode of Administration Self-administered via the computer and internet for remaining respondents Computer Assistance Computer-assisted self interview (CASI) and web-based survey Reporting Unit One person age 18 or older per household reports for the entire household Time Dimension Cross-sectional survey with hypothetical longitudinal stated preference
experiments Frequency One two-month phase of collecting responses Levels of Observation Household, vehicle, person
45
46
Experiment Directions • Make realistic decisions. Act as if you were actually buying a vehicle in a
real life purchasing situation. • Take into account the situations presented during the scenarios. If you
would not normally consider buying a vehicle, then do not. But if the situation presented would make you reconsider in real life, then take them into account.
• Assume that you maintain your current living situation with moderate increases in income from year to year.
• Each scenario is independent from one another. Do not take into account the decisions you made in former scenarios. For example, if you purchase a vehicle in 2011, then in the next scenario forget about the new vehicle and just assume you have your current real life vehicle.
46
47
Vehicle Technology Experiment
47
48
Results - Vehicle Technology
48
0
5000
10000
15000
20000
25000
30000
35000
40000
0%
5%
10%
15%
20%
25%
2010 2011 2012 2013 2014 2015
Vehi
cle
Pric
e
Adop
tion
Rate
Vehicle Price vs Adoption Rate
New gasoline New Hybrid New Electric Gasoline Price Hybrid Price Electric Price
49
Results – Vehicle Technology
Coefficient
Included in Utility
Value T-stat Curr
ent
Gas
olin
e
HEV
BEV
ASC – New Gasoline Vehicle -1.320 -3.28 ASC – New Hybrid Vehicle -1.760 -2.93 ASC – New Electric Vehicle -3.450 -5.70 Purchase Price [$10,000] -0.639 -5.42 Fuel Economy Change [MPG] (current veh. MPG known) 0.039 2.68 Fuel Economy Change [MPG] (current veh. MPG unknown) -0.002 -0.21 Recharging Range [100 miles] 0.909 4.37 Current Vehicle Age – Purchased New [yrs] -0.123 -4.34 Current Vehicle Age – Purchased Used [yrs] -0.059 -2.02 Minivan Dummy interacted with Family Households 1.410 2.75 SUV Dummy interacted with Family Households 1.900 4.77 Non-Electric Vehicle Error Component (standard deviation) 2.400 6.00 Non-Hybrid Vehicle Error Component (standard deviation) 2.150 6.71 Vehicle Size (mean) -0.435 -2.42 Vehicle Size (standard deviation) 1.09 6.61
Likelihood with Zero Coefficients -1379.4 "Rho-Squared" 0.406 Likelihood with Constants Only -1088.1 Adjusted "Rho-Squared" 0.395 Final Value of Likelihood -819.6 Number of Observations 995 (83)
49
50
Results – Vehicle Technology
• Gasoline and hybrid vehicles have a similar inherent preference • Families influenced by vehicle size • Fuel economy not significant for respondents who did not know
their own vehicle’s fuel economy • Covariance between Vehicle Types
– current vehicle + new gasoline vehicle (largest cov.) – new gasoline or current vehicle + new hybrid vehicle – new gasoline or current vehicle + new electric vehicle – new hybrid vehicle + new electric vehicle (smallest cov.)
• About 65% of respondents preferred smaller vehicles
50
51
Fuel Type Experiment
51
52
Results – Fuel Type
52
0
1
2
3
4
5
6
7
0%
5%
10%
15%
20%
25%
30%
2010 2011 2012 2013 2014 2015
Pric
e pe
r Gal
lon
(or E
quiv
alen
t)
Adop
tion
Rate
Fuel Price vs Adoption Rate
New Gasoline New Alternative Fuel New Electric New Plug-In Hybrid Gasoline Price Alternative Fuel Price Electricity Price
53
Results – Fuel Type
53
Coefficient
Included in Utility
Value T-stat Curr
ent
Gas
olin
e
AFV
Die
sel
BEV
PHEV
ASC – New Gasoline Vehicle -8.810 -6.81 ASC – New Alternative Fuel Vehicle -9.940 -7.66 ASC – New Diesel Vehicle -10.300 -7.84 ASC – New Battery Electric Vehicle -9.230 -4.07 ASC – New Plug-in Hybrid Electric Vehicle -10.100 -4.79 Fuel Price [$] -1.160 -7.79 Gasoline Price – PHEV [$] -0.358 -2.02 Electricity Price – BEV [$] -0.762 -3.02 Electricity Price – PHEV [$] -0.569 -2.79 Charge Time – BEV [hrs] -0.917 -3.68 Charge Time – PHEV [hrs] -0.164 -0.87 Average Fuel Economy [MPG, MPGe] 0.039 3.91 Current Vehicle Age – Purchased New [yrs] -0.395 -4.21 Current Vehicle Age – Purchased Used [yrs] -0.377 -3.86 Current Vehicle Error Component (standard deviation) 2.290 3.90 Electric Vehicle Error Component (standard deviation) 2.300 3.92 Liquid Fuel Vehicle Error Component (standard deviation) 3.460 4.91
Likelihood with Zero Coefficients -901.3 "Rho-Squared" 0.508 Likelihood with Constants Only -667.7 Adjusted "Rho-Squared" 0.489 Final Value of Likelihood -443.6 Number of Observations 503 (42)
54
Results – Fuel Type
• Respondents less sensitive to electricity price – Maybe lack of familiarity, no rule of thumb?
• Charging time has influence on attractiveness of BEVs but not PHEVs
• Error components shows that groups of respondents may have similar propensity towards electric vehicles (BEV and PHEV) and between liquid fuel vehicles
54
55
Taxation Policy Experiment
55
56
Results – Taxation Policy
56
0
10
20
30
40
50
60
70
80
0%
5%
10%
15%
20%
25%
30%
35%
2010 2011 2012 2013 2014 2015
VMT
Tax
($/1
000
mile
s)
Adop
tion
Rate
VMT Tax vs Adoption Rate
Drive Current Vehicle Less New Gasoline New Hybrid New Electric Current Vehicle VMT Gasoline VMT Hybrid VMT Electric VMT
57
Results – Taxation Policy
57
Coefficient
Included in Utility
Value T-stat Curr
ent
Gas
olin
e
HEV
BEV
ASC – New Gasoline Vehicle -7.170 -6.03 ASC – New Hybrid Vehicle -7.090 -5.94 ASC – New Electric Vehicle -7.590 -6.17 Hybrid Vehicle Deduction [$] divided by HH Income [$1000] 0.093 2.71 Electric Vehicle Deduction [$] divided by HH Income [$1000] 0.245 2.02 VMT Tax interacted with Annual Mileage [$100] -0.186 -5.14 Toll Discount [%] (for HHs near toll facilities) 0.065 2.76 Toll Discount [%] (for HHs not near toll facilities) 0.005 0.75 Current Vehicle Age (new) interacted with Annual Mileage [years x 1000 miles] -0.049 -5.24 Current Vehicle Age (used) interacted with Annual Mileage [years x 1000 miles] -0.026 -2.47 New Vehicle Error Component (standard deviation) 3.760 4.90 Current Vehicle Error Component (fixed to 0) 0.000 Fixed
Likelihood with Zero Coefficients -565.6 "Rho-Squared" 0.455 Likelihood with Constants Only -456.7 Adjusted "Rho-Squared" 0.436 Final Value of Likelihood -308.1 Number of Observations 408 (34)
58
Results – Taxation Policy
• ASCs similar to Vehicle Technology Experiment • Toll discount only significant for residents near toll facilities • Higher VMT tax for gasoline vehicles dissuaded new gasoline
vehicle purchases
58
59
Survey Redesign
• Eliminate the taxation policy experiment – Incorporate VMT tax into fuel type experiment – Incorporate Rebates into vehicle technology experiment
• Added open-ended questions for purchase reason of current vehicles – Able to elicit some opinions about vehicle preferences, attitudes, and
concerns
• All respondents participate in both choice experiments
59
60
Survey Redesign
• Vehicle Technology Experiment – Incorporate MPGe into vehicle technology experiment
• Respondents able to compare mpge and mpg in fuel technology experiment well
– Added fees and rebates for different vehicle types – Added Plug-in Hybrid Vehicle (PHEV) alternative
• Fuel Technology Experiment – Removed diesel vehicle option, added flex-fuel vehicle option – Added VMT tax depending on fuel type
60
61
New Vehicle Technology Experiment
61
62
New Fuel Type Experiment
63
New Fuel Type Experiment
• Purpose – Collect data on future household vehicle preferences in Maryland in
relation to fuel type – Determine if respondent could make dynamic vehicle purchase
decisions in a hypothetical short- to medium-term period
• Respondents given a stated preference survey over a hypothetical five year period with two scenarios per year
64
Prior Data Collection
• Respondent Characteristics – Age, gender, employment, commute
• Household Characteristics – Size, children, workers, location
• Current Vehicle Characteristics – Make and model, fuel economy, purchase reason
65
Alternatives
• Keep Current Vehicle • Buy New Gasoline Vehicle • Buy New Alternative Fuel Vehicle • Buy New Flex-Fuel Vehicle • Buy New Battery Electric Vehicle • Buy New Plug-in Hybrid Vehicle • Sell Current Vehicle
66
Attributes
• Fuel Price – $ per gallon (equivalent) • Miles Traveled Fee – $ per 1000 miles • Average Fuel Economy – miles per gallon (equivalent) • Fueling Station Availability – distance from home in miles • Battery Charging Time – hours per charge
67
Attribute Levels
• 6134 design • Fuel Price – 6 levels • Miles Traveled Fee – 3 levels • Average Fuel Economy – 3 levels • Fueling Station Availability – 3 levels • Battery Charging Time – 3 levels
68
Attribute Levels 2011 2012 2013
Fuel Cost VMT MPG Avail / Charge Fuel Cost VMT MPG
Avail / Charge Fuel Cost VMT MPG
Avail / Charge
Gasoline Fuel
2.50 20 5 2.75 22 5 3.03 1.80 24 5 2.75 25 5 3.06 28 5 3.41 3.00 31 5 3.00 30 5 3.35 34 5 3.73 4.50 38 5 3.50 3.91 4.37 4.00 4.48 5.02 4.50 5.05 5.66
Alternative Fuel (E85)
2.25 16 50 2.48 18 50 2.72 1.00 20 50 2.48 21 25 2.75 24 25 3.07 1.80 27 25 2.70 26 15 3.01 30 15 3.36 2.50 34 15 3.15 3.52 3.93 3.60 4.03 4.52 4.05 4.54 5.10
Electricity
3.70 60 4 3.81 65 4 3.93 0.50 70 3 4.40 80 5 4.58 85 5 4.76 1.00 90 4 4.90 100 6 5.15 105 6 5.40 1.80 110 5 5.30 5.62 5.96 5.70 6.10 6.53 6.05 6.53 7.06
Attribute levels for first three years of the experiment
69
Experimental Design Attribute
Design # Price VMT Fee MPG Availability Charge Time 1 0 0 0 0 0 2 1 2 2 0 1 3 2 1 2 1 0 4 3 1 0 2 2 5 4 0 1 2 1 6 5 2 1 1 2 7 0 1 1 1 1 8 1 0 0 1 2 9 2 2 0 2 1
10 3 2 1 0 0 11 4 1 2 0 2 12 5 0 2 2 0 13 0 2 2 2 2 14 1 1 1 2 0 15 2 0 1 0 2 16 3 0 2 1 1 17 4 2 0 1 0 18 5 1 0 0 1
70
Preliminary Model (New Data)
71
Preliminary Results
72
Capitol Beltway HOT Lane Study
Estimating Drivers’ Willingness to Pay for HOT Lanes on I-495 in Maryland
73
Overview
• Purpose – Determine preferences for use of high-occupancy toll (HOT) lanes on
I-495 in Maryland – Determine cost and time preferences as well as high-occupancy
vehicle preference
• Respondents given two experiments, both deal with lane choice and the second has a departure time component
74
Prior Data Collection
• Recent Trip (via I-495) Information – Passengers, Route Choice, Trip Purpose – Preferred Departure Time, Arrival Time – Actual Travel Time – Trip Distance on Beltway (D) – Actual Departure Time (DT), Arrival Time – Shortest Travel Time on Beltway (TTmin) – Longest Travel Time on Beltway(TTmax) – Fuel Cost (FC)
75
Departure Time Experiment
76
Alternatives
• Normal Lanes • HOT Lane without passenger (paid) • HOT Lane with passenger (free) • Use alternative route
77
Attributes
• 5431 design • Some attribute levels change depending on time of trip
• Departure Time • Travel Time
– Minimum Travel Time – Travel Time Range
• Fuel Cost • Toll Cost
78
Attribute Levels
≥≥
Variable Normal Lane HOT Lane HOV Lane (passengers 2)
Departure time
DT-40min DT-40min DT-40min
DT-20min DT-20min DT-20min
DT DT DT
DT+20min DT+20min DT+20min
DT+40min DT+40min DT+40min
Minimum Travel Time (minutes)
TTmin TTmin TTmin
TTmin + 5 TTmin + 5 TTmin + 5
TTmin + 10 TTmin + 10 TTmin + 10
TTmin + 15 TTmin + 15 TTmin + 15
TTmin + 20 TTmin + 20 TTmin + 20
79
Attribute Levels
Variable Normal Lane HOT Lane HOV Lane (passengers 2)
Travel Time Range (minutes) [during rush hour]
30 10 10
35 15 15
40 20 20
45 25 25
50 30 30
Travel Time Range (minutes) [not rush hour]
5 5 5
15 10 10
25 15 15
35 20 20
45 25 25
80
Attribute Levels
Variable Normal Lane HOT Lane HOV Lane (passengers 2)
Toll Cost ($) [during rush hour]
0 0.30 * D 0
0 0.35 * D 0
0 0.40 * D 0
0 0.45 * D 0
0 0.50 * D 0
Toll Cost ($) [not rush hour]
0 0.10 * D 0
0 0.15 * D 0
0 0.20 * D 0
0 0.25 * D 0
0 0.30 * D 0
81
Attribute Levels
Variable Normal Lane HOT Lane HOV Lane (passengers 2)
Fuel Cost [during rush hour]
FC * 110% FC FC
FC * 120% FC * 110% FC * 110%
FC * 130% FC * 120% FC * 120%
Fuel Cost [not rush hour]
FC * 110% FC FC
FC * 115% FC * 115% FC * 115%
FC * 120% FC * 120% FC * 120%
82
Experimental Design Scenario # Depart Time Min TT TT Range Fuel Cost Toll Cost
1 0 0 0 0 0 2 0 1 2 1 4 3 0 2 3 2 1 4 0 3 4 1 2 5 0 4 1 2 3 6 1 0 1 1 1 7 1 1 4 0 3 8 1 2 0 1 2 9 1 3 3 2 4
10 1 4 2 2 0 11 2 0 2 2 2 12 2 1 3 1 0 13 2 2 1 0 4 14 2 3 0 2 3 15 2 4 4 1 1 16 3 0 3 1 3 17 3 1 1 2 2 18 3 2 4 2 0 19 3 3 2 0 1 20 3 4 0 1 4 21 4 0 4 2 4 22 4 1 0 2 1 23 4 2 2 1 3 24 4 3 1 1 0 25 4 4 3 0 2