Technical ReportPDF Available

Are Consumers Willing to Pay to Let Cars Drive for Them? Analyzing Response to Autonomous Vehicles

Authors:

Abstract and Figures

Autonomous vehicles use sensing and communication technologies to navigate safely and efficiently with little or no input from the driver. These driverless technologies will create an unprecedented revolution in how people move, and policymakers will need appropriate tools to plan for and analyze the large impacts of novel navigation systems. In this paper we derive semiparametric estimates of the willingness to pay for automation. We use data from a nation-wide online panel of 1,260 individuals who answered a vehicle-purchase discrete choice experiment focused on energy efficiency and autonomous features. Several models were estimated with the choice microdata, including a conditional logit with deterministic consumer heterogeneity, a parametric random parameter logit, and a semiparametric random parameter logit. We draw three key results from our analysis. First, we find that the average household is willing to pay a significant amount for automation: about $3,500 for partial automation and $4,900 for full automation. Second, we estimate substantial heterogeneity in preferences for automation, where a significant share of the sample is willing to pay above $\$$10,000 for full automation technology while many are not willing to pay any positive amount for the technology. Third, our semiparametric random parameter logit estimates suggest that the demand for automation is split approximately evenly between high, modest and no demand, highlighting the importance of modeling flexible preferences for emerging vehicle technology.
Content may be subject to copyright.
1616 P St. NW
Washington, DC 20036
202-328-5000 www.rff.org
August 2016 RFF DP 16-35
Are Consumers
Willing to Pay to Let
Cars Drive for Them?
Analyzing Response to
Autonomous Vehicles
Ricardo A. Daziano, M auricio Sarrias, and
Benjamin Leard
DISCUSSION PAPER
Are consumers willing to pay to let cars drive for
them? Analyzing response to autonomous vehicles
Ricardo A. Daziano
, Mauricio Sarrias,and Benjamin Leard
2016
Abstract
Autonomous vehicles use sensing and communication technologies to navigate
safely and efficiently with little or no input from the driver. These driverless
technologies will create an unprecedented revolution in how people move, and
policymakers will need appropriate tools to plan for and analyze the large impacts
of novel navigation systems. In this paper we derive semiparametric estimates of
the willingness to pay for automation. We use data from a nation-wide online
panel of 1,260 individuals who answered a vehicle-purchase discrete choice experiment
focused on energy efficiency and autonomous features. Several models were estimated
with the choice microdata, including a conditional logit with deterministic consumer
heterogeneity, a parametric random parameter logit, and a semiparametric random
parameter logit. We draw three key results from our analysis. First, we find
that the average household is willing to pay a significant amount for automation:
about $3,500 for partial automation and $4,900 for full automation. Second, we
estimate substantial heterogeneity in preferences for automation, where a significant
share of the sample is willing to pay above $10,000 for full automation technology
while many are not willing to pay any positive amount for the technology. Third,
our semiparametric random parameter logit estimates suggest that the demand for
automation is split approximately evenly between high, modest and no demand,
highlighting the importance of modeling flexible preferences for emerging vehicle
technology.
School of Civil and Environmental Engineering, Cornell University, Ithaca, NY 14853; email:
daziano@cornell.edu
Department of Economics, Universidad Catolica del Norte
Resources for the Future, Washington, DC
JEL classification: C25, D12, Q42.
Key words: willingness to pay, autonomous vehicle technology, discrete choice models;
semiparametric heterogeneity;
1 Introduction: Technological change in the auto-
motive market
Personal mobility is about to experience an unprecedented revolution motivated
by technological change in the automotive industry (National Highway Traffic
Safety Administration,2013;Fagnant and Kockelman,2014). The introduction of
automated vehicles–in which at least some (and potentially all) control functions
occur without direct input from the driver–will completely change how people move.
The adoption of automated navigation systems has the potential to dramatically
reduce traffic congestion and accidents–two major externalities in the transportation
market, while creating substantial improvements in the overall trip experience
as well as providing enhanced accessibility opportunities to people with reduced
mobility. Additionally, serious environmental and energy externalities associated
with both carbon-dependent technologies and inefficient driving can experience major
reductions with the advent of automated electric vehicles.
Automated vehicles use sensing and communication technologies to navigate
safely and efficiently with little or no human input. Automated navigation technology
comprises any combination of (1) self-driving navigation systems informed by on-
board sensors (autonomous vehicles) vehicle-to-vehicle (V2V) and (2) vehicle-to-
infrastructure (V2I) communication systems that inform navigation and collision
avoidance applications (connected vehicles). The National Highway Traffic Safety
Administration (NHTSA) has suggested five levels of automated navigation: level
0 (no automation), where the driver is in complete control of safety-critical
functions; level 1 (function-specific automation), where the driver cedes limited
control of certain functions to the vehicle especially in crash-imminent situations
(adaptive cruise control, electronic stability control ESC, automatic braking); level
2 (combined-function automation), which enables hands-off-wheel and foot-off-pedal
operations, but the driver is expected to be available at all times to resume control of
the vehicle (adaptive cruise control and lane centering); level 3 (limited self-driving
or conditional automation), where the vehicle potentially controls all safety functions
under certain traffic and environmental conditions, but some conditions require
transition to driver control; and level 4 (driverless or full self-driving automation),
1
where the vehicle controls all safety functions and monitors conditions for the whole
trip. KPMG (2015) further distinguishes two levels within NHTSA level 4, based on
whether driver control would be required in unusual circumstances or full automation
would be guaranteed in all conditions. A similar categorization into six levels is
proposed by the Society of Automotive Engineers (SAE).
Both automakers and experts in the vehicle industry are predicting that
self-driving cars will be available for commercialization by 2020. Several semi-
autonomous features are already available in the automotive market, mostly in the
form of in-vehicle crash avoidance upgrades with preventive warnings or limited
automated control of safety functions, such as braking when danger is detected.
Self-parking assist systems are another example of a more advanced upgrade that
is currently available in select makes and models. These entry-level automation
packages are possible as a result of vehicles being equipped with radar, cameras, and
other sensors. Even though technology is still evolving, full automation is possible
with the current stage of development. The Google car and its more than 1.5 million
miles of driverless driving is the most publicized effort.
The literature on vehicle-to-vehicle, vehicle-to-infrastructure, and control systems
for safe navigation is extensive. Regulation, insurance, and liability are other areas
where there is strong debate. However, little attention has been devoted to the
analysis of automated vehicles as marketable products. Consumer acceptance is a
critical issue to forecast adoption rates, especially if one considers that there may be
strong barriers to entry (potential high costs, concerns that technology may fail).
Our work contributes to two strands of literature on the demand for new
technology. The first area is the recent development in understanding the demand,
penetration, and policy implications of autonomous vehicle technology. Several
recent studies attempt to understand how consumer preferences for attributes such
as safety, travel time, and performance shape the demand for driverless cars.
Kyriakidis et al. (2015) conducted an international public opinion questionnaire of
5,000 respondents from 109 countries. Responses were diverse: 22 percent of the
respondents did not want to pay any additional price for a fully automated navigation
system, whereas 5 percent indicated they would be willing to pay more than $30,000.
Payre et al. (2014) conducted a similar survey of 421 French drivers with questions
2
eliciting the acceptance of fully automated driving. Among those surveyed, 68.1
percent accepted fully automated driving unconditionally, with higher acceptance
conditional on the type of driving, including usage of highway driving, in the presence
of traffic congestion, and for automated parking. Similar results were obtained in
a survey of Berkeley, California, residents conducted by Howard and Dai (2013).
Individuals in this survey were most attracted to the potential safety, parking, and
multi-tasking benefits. Schoettle and Sivak (2014) conducted a much larger and
more internationally based survey of residents from China, India, Japan, the united
States, the United Kingdom, and Australia. The authors found that respondents
expressed high levels of concern about riding in self-driving vehicles, with the most
pressing issues involving those related to equipment or system failure. While most
expressed a desire to own an autonomous vehicle, many respondents stated that they
were unwilling to pay extra for the technology.
In contrast to these studies, our work presents estimates of the willingness to pay
(WTP) for automation based on discrete choice experiments with realistic choice
settings. In these settings, respondents chose which new vehicle to buy from among
four possible new vehicle models that vary along multiple desirable characteristics,
including operating cost, price, level of automation, and other features. This
allowed us to to estimate stated preferences for automation based on which vehicles
respondents chose in the choice experiments, which contrast with results from
previous work that directly elicited demand.
A paper related to our own is that by Bansal et al. (2016), which estimates
willingness to pay for different levels of automation. They find that for their sample
of 347 residents of Austin, Texas, WTP for full automation is $7,253, which is
substantially higher than our own estimate. The authors also estimate WTP for
partial automation of $3,300, which is closer to our own estimate. Differences in
WTP estimates stem from two differences in methodology. First, Bansal et al.
estimate preferences for an unrepresentative sample of the U.S. population, while
our sample is representative along many observed household demographics. Second,
Bansal et al.’s empirical model is based on a stated preference experiment that does
not include an option for selecting a vehicle with conventional (non-autonomous)
technology. In sharp contrast, in our choice experiments, many of the cases include
3
at least one option that has no automation technology. We believe that this feature
of our experiments is crucial to elicit how respondents trade off other attributes that
differ between alternatives with and without automation.
Our demand estimates also contribute to the assessment of the social costs
and benefits of autonomous vehicles. Fagnant and Kockelman (2015) estimate
the external net benefits from autonomous vehicle penetration. They find that
the social net benefits including crash savings, travel time reduction from less
congestion, fuel efficiency savings, and parking benefits total between $2,000 and
$4,000 per vehicle. These estimates, however, greatly depend on how the presence of
autonomous vehicles will impact both vehicle ownership and utilization. For example
if autonomous vehicles make owning a vehicle more desirable, then the stock and
use of vehicles may increase, reducing the external net benefits. Our estimates of
WTP for privately owned autonomous vehicles take a first step to understanding
the demand for this technology, which is critical for understanding how aggregate
demand for vehicles and vehicle miles traveled will respond to the technology over
time.1
In this paper, our focus is on exploring the willingness to own autonomous
electric vehicles (AEVs). Although intelligent navigation is not necessarily related
to energy storage, the synthesis of automation and electrification provides a personal
transportation alternative that minimizes the total negative environmental impact of
its use. In addition, because of current limitations of battery electric vehicles impose
a particular way of driving to maximize performance of the battery. Eco-driving will
be automatic in autonomous vehicles. Furthermore, because of reductions in the
likelihood of accidents, automation will come with an eventual major reduction in
vehicle weight. In this sense, self-driving features can optimize energy efficiency while
improving lower-performing attributes such as driving range. Finally, it is likely that
1We do not explore demand for autonomous commercial vehicles or for autonomous public
transportation. Initial work in this area includes a study by Greenblatt and Saxena (2015)
which simulates the greenhouse gas impact of autonomous vehicle taxis and finds that they can
dramatically reduce greenhouse gas emissions relative to conventional taxis. A promising area of
future research involves incorporating our survey and econometric methods for eliciting WTP to
determine how households tradeoff cost savings, travel time, safety, and other desirable attributes
with alternative travel modes with and without a human driver.
4
both electrification and automation will be combined to attract early adopters of
disruptive technology.
In sum, for this work, we designed a web-based survey with a discrete choice ex-
periment to determine early-market empirical estimates of the structural parameters
that characterize current preferences for autonomous and semi-autonomous electric
vehicles. The discrete choice experiment contained as experimental attributes three
levels of automation: no automation, some or partial automation (“automated crash
avoidance”), and full automation (“Google car”). Automation was allowed for alter-
native powertrains (hybrid electric, plug-in hybrid and full battery electric).
In addition to the discrete choice experiment of vehicle purchase, the survey
also contained an experiment to elucidate the subjective discount rate of potential
vehicle buyers. Expanding on the work of Newell and Siikam¨aki (2013), we used
the individual-level experimental discount rate to determine the present value of fuel
costs for each alternative.
To derive flexible estimates of the heterogeneity distribution of the willingness to
pay for automation, we implemented the maximum simulated likelihood estimator
of a logit-based model with discrete continuous heterogeneity distributions. The
approach adopted to unobserved preference heterogeneity in this paper takes into
consideration a mixed-mixed logit model (Bujosa et al.,2010;Greene and Hensher,
2013;Keane and Wasi,2013), where the random willingness-to-pay parameters are
distributed according to a Gaussian mixture. The weights of the Gaussian mixture
can include individual-specific covariates that allow us to identify clusters with
differing willingness to pay for automation. The estimator was implemented with
analytical expressions of the score for computation efficiency.
The remainder of the paper is organized as follows. In section 2, we present a
series of discrete choice models that we use to estimate how consumers value personal
vehicle automation. In section 3, we discuss the survey data and provide summary
statistics of the sample. We then present the empirical models and estimation results
in section 4and draw conclusions based on our results in section 5.
5
2 Structural Vehicle Choice Models
The purchase of an automated vehicle can be modeled as the consumer choice to
adopt high technology, durable goods. The use of discrete choice models to analyze
vehicle purchases in general dates back to the earliest econometric applications of
the principle of random utility maximization. Within this literature, great interest in
modeling the adoption of battery electric vehicles has emerged in the last five years
(for literature reviews, see Rezvani et al.,2015;Al-Alawi and Bradley,2013).
Because the transition to energy efficiency in personal transportation is
characterized by the trade-off between higher purchase prices and lower operating
costs, a specific avenue of research has been taking into account time preferences to
represent how consumers discount future savings. Seminal work on the problem of
estimating individual discount rates with discrete choice models includes Hausman
(1979), Lave and Train (1979), and the technical reports cited in Train (1985). In
addition, excellent and relatively recent literature reviews are provided by Frederick
et al. (2002) and Cameron and Gerdes (2005). Expanding on Jaffe and Stavins
(1994), several resource and energy economists have added to the debate about
the energy paradox (Newell and Siikam¨aki,2013;Allcott and Greenstone,2012;
Ansar and Sparks,2009;Van Soest and Bulte,2001;DeCanio,1998;Hassett and
Metcalf,1993). As reviewed in Wang and Daziano (2015), there are two approaches
to introducing discount rates in discrete choice models: endogenous discounting, in
which discount rate estimates are derived from the marginal rate of substitution
between price and operating cost, and exogenous discounting, in which the discount
rate is assumed as known.
Working with exogenous discount rates has been proposed in the energy
economics literature to avoid confounding effects in the determination of discount
rate estimates coming from market failures (Allcott and Wozny,2014;Newell and
Siikam¨aki,2013). Exogenous discounting takes as known the discount rate of
individual i, making it straightforward to calculate the present value of future costs,
PVFCij. Moving future cash flows to the present allows the researcher to use a
static discrete choice specification. If in addition to monetary attributes, vehicle
design attributes xij are considered (such as power, drivetrain, refueling time, and
6
driving range), then the conditional indirect utility can be specified as
Uij =x0
ij ωx,i αipriceij γPVFC,iPVFCij +εij .(1)
Equation (1) represents our benchmark specification and is formulated in preference
space. ωx,i is the change in utility from marginal improvements in the (nonmonetary)
vehicle design attributes that are captured in the vector xij,αiis the marginal utility
of income, and γPVFC,i is the change in utility from a marginal change in the present
value of fuel costs. For a rational consumer γPVFC,i =αi, since both priceij and
PVFCij are monetary attributes at the time of purchase. If γPVFC,i < αi, then there
is evidence for myopic consumption (as consumers weigh more than saving one dollar
in purchase price than the same dollar in discounted future costs), and γPVFC,i > αi
reveals that consumers overvalue fuel costs. In our benchmark specification, we
assume that the idiosyncratic error term εij is i.i.d. distributed Type 1 extreme
value, so that predicted probabilities take on the conditional logit form.
In this paper, in addition to standard assumptions of unobserved heterogeneity
in the parameters (such as normally and lognormally distributed parameters),
we consider a semi-parametric discrete-continous mixture for the heterogeneity
distributions. In fact, following the idea of the mixed-mixed logit model (MM-MNL)
that represents heterogenous preferences as a weighted average of normals (Bujosa
et al.,2010;Greene and Hensher,2013;Keane and Wasi,2013).2
If θ0
i= (αi, γPVFC,i,ω0
x,i) represents the full vector of parameters of interest,
the heterogeneity distribution assumption is the following Gaussian mixture with
Qcomponents: θi N (θq,Σq) with probability wiq for q∈ {1, . . . , Q}or
fΘ(θi) = PQ
q=1 wiqfq(θi), where fΘis the density function of the heterogeneity
distribution of the parameters of interest and fq(θi) is the multivariate normal density
with parameters θqand Σq. The weights of the mixture wiq can be interpreted as
class assignment probabilities, and can be constant or a function of covariates. In
particular, the weights can be specified as a function wiq =wiq(zi|δ), where ziis a
vector of individual-specific characteristics and δis a vector of parameters. As in
latent class discrete choice models, a possibility is to assume a logit-type specification
2Any continuous distribution can be approximated by a discrete mixture of normal distributions
(Train,2008).
7
for the mixture weights:
wiq =exp(z0
iδq)
PQ
q=1 exp(z0
iδq),(2)
where the vector component-specific (or class-specific) parameter vector is normalized
for identification. For example, normalizing δ1=0ensures that the parameters for
the rest of the components are identified.
Assuming that we observe Tchoices made by individual iand that εijt is i.i.d.
type 1 extreme value for t∈ {1, . . . , T}, the MM-MNL probability of the sequence
of choices is given by:
Pi=
Q
X
q=1
wiq(δ)Z
T
Y
t=1
J
Y
j=1
exp x0
ij ωx,i αipriceij γPVFC,iPVFCij
PJ
j=1 exp x0
ij ωx,i αipriceij γPVFC,iPVFCij
yit
fq(θi)dθi.(3)
As in a mixed logit model, the above probability can be approximated using Monte
Carlo integration:
˜
Pi=1
R
Q
X
q=1
wiq(δ)
R
X
r=1
T
Y
t=1
J
Y
j=1
exp x0
ij ω0(r)
x,i,q α(r)
i,q priceij γ(r)
PVFC,i,qPVFCij
PJ
j=1 exp x0
ij ω0(r)
x,i,q α(r)
i,q priceij γ(r)
PVFC,i,qPVFCij
yit
,(4)
where (α(r)
i,q , γ(r)
PVFC,i,q,ω0(r)
x,i,q) represents random draw r∈ {1, . . . , R}from the normal
density fq(θi|θq,Σq).
Finally, using the Monte Carlo approximation of the probability of the sequence
of choices by individual i, it is possible to find the maximum simulated likelihood
estimator by maximizing the following simulated likelihood:
˜
`(θQ,δQ,ΣQ;y|X,Z,price,PVFC) =
N
Y
i=1
˜
Pi,(5)
where θQ= (θ1,...,θQ), δQ= (δ2,...,δQ) (if the first component is normalized),
and ΣQ= (Σ1,...,ΣQ).
3 Vehicle Choice Data
3.1 The survey
To support design of the survey, we first conducted focus groups where new vehicle
preferences and attitudes toward automated cars were discussed by randomly selected
8
potential car buyers. The participants discussed benefits and eventual dangers of
automation. Among the benefits, they mentioned less traffic jams, increased mobility
independence, and easier and quicker parking. Another benefit of automation that
was discussed was the possibility of multitasking and increased productivity. One of
the most relevant features that people look for in a new car is safety. Participants
of the focus groups confirmed that safety is a major concern. However, their
perceptions about driverless cars and safety were divided. Some participants agreed
that automation has great potential to reduce accidents, but a majority also said that
unfortunately machinery fails. Concerns about lighter vehicles being more dangerous
also were raised. The qualitative information that was collected in the focus groups
was used to design an attitudinal module of the survey, which supplements the data
that were collected using the discrete choice experiment.
3.2 The data
We used the Qualtrics online platform to collect the survey data. We surveyed
a sample of individuals who provided valid responses for personal characteristics
questions and all of the vehicle choice experiments. We collected several waves of
responses between September 12, 2014, and October 2, 2014, for a total of 1,260
individuals.3
Table 1reports demographic statistics for respondents in our sample. The sample
is broadly representative of the U.S. population. Mean and median household
incomes are $61,226 and $55,000, respectively, which are close to reported estimates
from the 2013 American Community Survey;4the sample’s fraction of married adults
well represents the estimates of the U.S. marriage rate of around 50 percent; the
unemployment rate of 5.79 percent among our sample respondents is close to the most
recently reported national unemployment rate for September 2014 of 5.9 percent.5
The sample appears to only slightly over-represent white respondents and slightly
3Out of the sample of 1,260, 549 responses were collected between September 12 and September
15, 214 were collected between September 19 and September 23, and the remaining responses were
collected between September 29 and October 2.
4These estimates are available at http://www.census.gov/content/dam/Census/library/
publications/2014/acs/acsbr13-02.pdf.
5See http://data.bls.gov/timeseries/LNS14000000.
9
under-represent minorities; the U.S. Census reports that 77.7 percent of U.S. citizens
are white, while our sample includes 85 percent.6Our sample is slightly more
educated relative to the average for U.S. citizens; 38 percent of respondents state
that they have earned at least a bachelor’s degree, while only about 30 percent
of U.S. citizens have done so. These small differences can be explained by the
screening process of our survey. Two screening questions, whether the respondent
has a driver’s license and whether the respondent has access to a household vehicle,
likely disproportionately discourage minorities and less educated individuals from
taking our survey. Fortunately, however, this effect appears to be quite mild as
suggested by the descriptive statistics of our sample.
Table 2reports statistics for the vehicle holdings data in our sample. These data
represent vehicles that are driven most often among all vehicles held by respondent
households. We merge survey responses on the model year, make, model and trim of
the vehicle with trim-level characteristics data from Ward’s Automotive. Vehicle age
and annual vehicle miles traveled (VMT) are based on two questions in our survey.
The average age among all vehicles is seven years, which is about two years
younger than the average age of all autos held by households in 2008.7This seems
reasonable considering that the reported vehicle holding in our survey is conditional
on being the vehicle that is driven most often and not simply a random vehicle chosen
from the full set of household vehicle holdings.8For the same reason, annual VMT
is slightly over 15,000 miles, which is close to the average reported VMT of new
cars and light trucks.9The selection is also a reason why the average vehicle fuel
economy in our sample is remarkably high.10 Average fuel economy of automobiles
6See http://quickfacts.census.gov/qfd/states/00000.html.
7This is based on the 2009 National Household Transportation Survey, Summary of Travel
Trends: http://nhts.ornl.gov/2009/pub/stt.pdf
8It is well documented that vehicles with more annual miles traveled are generally newer. See
Lu (2006), http://www-nrd.nhtsa.dot.gov/Pubs/809952.pdf, for more details.
9Lu (2006) documents that the average VMT for new cars is 14,231, which falls to 12,325
by age seven; the average VMT for new trucks is 16,085, which falls to 12,356 by year 10. See
http://www-nrd.nhtsa.dot.gov/Pubs/809952.pdf.
10In fact, it is close to the average record high 24.9 miles per gallon fuel economy of new 2013
model year vehicles. See http://www.umich.edu/~umtriswt/EDI_sales-weighted-mpg.html.
10
sold in 2007–the average model year of vehicles in our sample–was around 20 miles
per gallon. Households in our sample, however, likely optimize their fleet utilization
choices by driving their relatively fuel efficient vehicles more than their relatively fuel
inefficient vehicles. Therefore, the vehicles that respondents report are more likely
to have high fuel economy.
Patterns in vehicle characteristics across the different styles are in line with
expectations. Fuel efficiency measured in miles per gallon is higher for smaller cars
including coupes, sedans, and wagons and lower for larger, more powerful autos
including trucks and SUVs. Trucks are older than the average vehicle by about
three years, which is also in line with data from the 2009 National Household
Transportation Survey.11 Trucks are generally driven more per year and over the
entire vehicle lifetime than cars, which is consistent with the reported travel data
from our survey.12
3.3 Design of the choice experiment
The discrete choice experiment that we designed is based on a labeled experiment
with quasi-customized alternative attributes. The alternatives are constructed
according to general new vehicle preferences, including stated price thresholds. The
experimental attributes include purchase price, fuel cost expenses, driving range,
recharging time, and levels of hybridization and automation. Levels are described in
Table 3. Note that purchase price in the experiment was customized to the threshold
stated by the respondent when asked about the willingness to spend in buying a new
vehicle.
For automation we considered an aggregation of the NHTSA levels in three
groups: no automation (base), some automation (“automated crash avoidance”),
and full automation (“Google car”).
The automation level aggregation, examples for each level (e.g., “automated crash
avoidance” for some automation), and the connected icon to graphically represent
automation in the discrete choice experiment were discussed in two focus groups
that were performed before final design of the survey. An example of the image that
11See http://nhts.ornl.gov/2009/pub/stt.pdf..
12For more details, see Lu (2006), http://www-nrd.nhtsa.dot.gov/Pubs/809952.pdf.
11
participants saw during one choice situation appears in Figure 1.
3.4 Elicited subjective discounting
As reviewed in Wang and Daziano (2015), laboratory and field time preferences
experiments have been used in experimental economics to elucidate subjective
discount rates. Expanding on the work of Newell and Siikam¨aki (2013), who
implemented and used the Multiple Price List (MPL) method of Coller and Williams
(1999) to analyze consumers’ response to energy efficiency labels on water heaters,
in our survey we implemented a modified version of the MPL method. MPL is
organized as a series of binary choices between an immediate and a delayed reward,
in which increasing exogenous discount rates are used to determine the values of the
rewards (cf. Kirby et al.,1999). In our survey, only one binary choice was shown to
participants at a time, with scenarios being displayed at an increasing interest rate.
Assuming transitivity in intertemporal preferences, the experiment ended as soon as
the respondent accepted the delayed reward, and the associated discount rate at the
accepted delayed reward was set as the individual’s subjective discount rate. Further
details about the survey implementation of the MPL method (such as avoidance of
immediacy bias) are discussed in Wang and Daziano (2015) with data from a pretest.
The elicited subjective discount rate resulting from the MPL experiment has a
mean of 12.18 percent, standard deviation of 12.86 percent, and a median of 10
percent. Both the median and mean are higher than market interest rates for the
automotive market, but are lower than some subjective discount rates that have been
found using the endogenous discounting approach. Newell and Siikam¨aki (2013) in
their experiment found a mean of 19 percent, standard deviation of 23 percent, and
median of 11 percent.
As in Newell and Siikam¨aki (2013), we combine discrete choice models with the
elicited intertemporal preferences, by calculating the present value of future costs as
PVFCij =
Li
X
l=1
operating costij
(1 + ρi)l,(6)
where Liis the total ownership time stated in the survey by individual i, and ρiis
12
the elicited subjective discount rate.13
4 Model Specification, Estimation, and Inference
4.1 Base models
In Table 4we report estimates for our benchmark conditional logit model with fixed
coefficients defined in (1). We provide three separate versions of the model, with each
version having a different method of defining fuel costs. In the first two versions, we
replace the present value of fuel costs with alternative measures of fuel cost. The first
version allows fuel cost to enter as a monthly cost, which is based on the respondent’s
expected amount of monthly driving and the cost per mile attribute. The second
version is only the cost per mile as a simple attribute. The third version includes the
present value of fuel costs (PVFC) as a function of the respondent’s elicited discount
rate, expected length of ownership, expected amount of driving during ownership
and the cost per mile attribute. We note that to avoid convergence issues in the
search for the maximum likelihood estimate, different tables may scale the attributes
differently. The actual scale for each attribute is discussed in the notes under each
table.
In each model, the coefficients on the vehicle attributes are estimated to have the
expected sign. We report these coefficients in the first panel of Table 4. Respondents
dislike higher purchase prices, higher operating costs, and longer charging times
and like longer ranges and both levels of automation. Purchase price sensitivity
has a point estimate ranging from 0.77 to 0.772 and enters significantly at the
13Our measure of the present value of fuel costs does not consider lifetime fuel costs since we
do not survey whether respondents perceive fuel costs beyond their ownership period. If survey
respondents value these costs beyond their ownership period–for example, if they expect to sell their
vehicle and when they sell, they expect that fuel costs are capitalized in used vehicle prices–then
our measure of fuel costs will be an underestimate of the respondents’ expectations. This will lead
us to overestimate WTP of the present value of fuel costs. We expect this bias to be small since a
large majority of fuel costs are incurred during the initial years of ownership. Furthermore, no prior
papers directly examined whether households value post-ownership fuel costs when purchasing a
new vehicle, although indirect evidence indicates that used vehicle markets do capitalize these costs
(Allcott and Wozny,2014;Busse et al.,2013;Sallee et al.,2016).
13
5 percent confidence level in each model. All three forms of operating costs enter
significantly and with the expected negative sign. Both forms of automation also
enter significantly and have an expected order, where full automation is preferred
over partial automation (“automated crash avoidance”).
To convert the preference parameters into dollar terms, we compute willingness
to pay for an additional unit of each attribute by dividing the marginal utility of
each attribute by the marginal utility of purchase price. Respondents are willing to
pay about $34 in a higher purchase price to reduce the monthly operating cost by
$1. This willingness to pay approximately represents a three-year payback window,
which is consistent with recent survey evidence on the consumer valuation of fuel
costs (Greene et al.,2013).14
Respondents are willing to pay slightly more than $3,500 for partial levels of
automation and about $4,900 for full automation. Are these estimates plausible?
They appear to be surprisingly close to the reported market price for Tesla’s autopilot
package available for $4,250, which was announced a couple of weeks after the survey
data were collected. This autopilot package is closer to our partial automation option
as it involves software that helps avoid collisions from the front or sides or from
leaving the road. The only fully autonomous package that appears close to market
is an add-on package called Cruise RP-1, which is a driving program capable of full
automation on certain highways. The current price tag for this program is $10,000.
4.2 WTP models using parametric and semi-parametric
heterogeneity distributions
The base models were extended to mixed logit specifications in preference space.
Table 5presents the results of a mixed logit model where key parameters are normally
distributed, where we interact key parameters with respondent characteristics,
14The empirical literature on how consumers value fuel cost savings is mixed and varies widely
depending on method, time span, and unit of analysis (Greene,2010). Several recent studies in
the economics literature that leverage variation in gasoline prices, however, suggest that consumers
fully value or only slightly undervalue fuel cost savings in new vehicle markets and only moderately
undervalue these savings in used vehicle markets (Allcott and Wozny,2014;Busse et al.,2013;
Sallee et al.,2016).
14
and where some parameters normally distributed and others are log-normally
distributed. For the model with respondent characteristics interactions, interactions
of sociodemographics with the levels of automation were considered to determine
potential deterministic preference variations. To compute WTP for automation and
other variables, we estimate a fixed parameter for vehicle purchase price then divide
the preference parameters by the purchase price parameter.
In the column labeled MIXL-N, we estimate normally distributed coefficients for
the natural log of range, charging time, and the two levels of automation. The
parameter estimates with (µ) next to them represent estimates of the mean of each
coefficient, while the parameter estimates with (σ) next to them represent estimates
of the standard deviation of each coefficient. Each coefficient has the expected sign,
as respondents dislike higher prices, higher fuel costs and greater charging times
while they like longer ranges and automation. The implied WTP for both levels of
automation are large and significant. Both, however, are substantially smaller than
the estimates from our fixed coefficient logit models in Table 4. Furthermore, the
mean WTP for the first level of automation, $1,453, exceeds the mean WTP for
the second level of automation, $990, which is unexpected and runs contrary to our
benchmark model results.
Moving to the next model, in the column labeled MIXL-N-OH, we present results
of the same model but with respondent interaction terms. We interact the two levels
of automation with several respondent characteristics: whether the respondent has
heard of the Google car, whether the respondent is male, the number of years of
experience driving, and geographic region. The implied WTP estimates for this
model seem more plausible. For some subsets of households, however, the implied
WTP is much higher than the average estimates from the models in Table 4. For
example, we estimate that wealthy female respondents living in the Midwest with
little driving experience that have heard of the Google car are willing to pay in excess
of $20,000 for full automation technology. This seems plausible given the degree of
differentiation among household preferences.
In the next two columns labeled MIXL-LN-I and MIXL-LN-II, we present models
for results where we assume the coefficients for both levels of automation are log-
normally distributed. We report the implied mean of these distributions. Our
15
estimates indicate that respondents are willing to pay about $1,000 for either level
of automation. We summarize the estimates for willingness to pay for automation
from the parametric heterogeneity models in Figure 2. The left and right panels in
Figure 2show the distribution of willingness to pay for the first and second levels
of automation, respectively. We can see from both panels that the heterogeneity in
WTP is large, even for the log normal specifications. These estimates are at odds
with our fixed coefficient model estimates and are likely driven by model fit. This
motivates the use of more flexible methods for estimating heterogeneous preferences
for automation, which we explore next with estimates from semi-parametric discrete-
continuous mixture models.
Table 6presents the results of mixed-mixed multinomial logit specifications with
three classes. For the column labeled Class 1, class assignment is set as base, whereas
for Classes 2 and 3, class assignment is a function of socioeconomic covariates. For
example, the respondent stating that he or she has heard of the Google car increases
the likelihood that the respondent has preferences represented by Class 2 or Class 3,
as inferred by the positive coefficient for the Google car covariate for these classes.
Class 1 includes slightly less than a third of the sample at 29 percent and Class 3
includes slightly more than a third at 38 percent.
As expected, each class dislikes higher prices and fuel costs and likes longer driving
range. The classes, however, have extremely different preferences for automation.
Class 1 respondents have a mean estimate for WTP for automation that is not
statistically different from zero. These respondents vary widely in their WTP for
both types of automation, with each having a standard deviation higher than $10,000.
This class is likely composed of households that are not aware of driverless car
technology or are skeptical of the technology, as these households are less likely
to have heard of the Google car and own fewer vehicles. Hence many households in
this group are not willing to pay a positive amount for the technology.
Class 2 respondents are, on average, willing to pay a substantial amount for
automation. These respondents are willing to pay an average of $2,784 and $6,580 for
partial and full automation, respectively. These values are in the range of the values
from our benchmark estimates appearing in Table 4. This group of respondents
appears to be eager to purchase automation technology once it becomes affordable.
16
Their preferences are driven by knowledge of the Google car, driving long distances,
vehicle ownership, and higher education. It is important to note that the standard
deviation for full automation for Class 2 is statistically significant and is $15,526,
which is more than twice as large as the point estimate for the mean. This implies
that some respondents in this group remain skeptical of the technology and are
not willing to pay anything for it. On the other hand, the large standard deviation
implies that some respondents are willing to pay large sums of money–on the order of
$10,000–for full automation. Households in the United States that share preferences
with these respondents will likely be the first to adopt fully autonomous vehicles
when they become commercially available.
Class 3 respondents appear to have moderate desire for automation and represent
a middle group between Classes 1 and 2. This group, which includes the largest
number of respondents, is willing to pay $1,187 and $1,422 for partial and full
automation, values that are substantially less than mean WTP for Class 2 and are less
than the mean WTP for both groups of automation from our benchmark models.
This group appears to be composed of individuals who have heard of the Google
car and that have driving experience. Class 3 individuals are also more likely to
be married and prefer driving. The price of automation must drop dramatically
before this group completely adopts the technology. Similarly to individuals in
Classes 1 and 2, individuals in Class 3 vary considerably in their preferences for
automation, as the standard deviation estimates for both types are large and
statistically significant. This result solidifies the notion that because automation
is a relatively new technology, preferences for the technology will vary widely until
it becomes more mainstream and consumers gain experience with it.
We plot the implied distributions of WTP for both levels of automation
in Figure 3. Similar to our results from models with parametric hetergeneity,
these distributions illustrate that households vary considerably in their desire
for autonomous features. Furthermore, these distributions appear more intuitive
than those from the parametric heterogeneity estimates. The mean estimates
of each distribution have the following intuitive appeal. The average Class 1
household dislikes automation and especially dislike full automation; the average
Class 2 household is willing to pay a high premium for automation, especially full
17
automation; the average Class 3 household is willing to pay a modest amount for
either type of automation. These results, which are not obvious from the models
with parametric heterogeneity, suggest a fairly even segmentation of the demand for
automation, where about one-third of the population highly desires the technology,
one-third has mild interest, and one-third does not want the technology.
5 Conclusions
The transformative nature of automation in personal mobility will be evidenced by
dramatic improvements in overall efficiency of transportation, from crash avoidance
and congestion reduction to enhanced accessibility for individuals with current
mobility constraints. Automation technology is becoming more mainstream as many
companies are adopting semi-autonomous features in their vehicles. The rate of
penetration of these technologies and more ambitious versions in the form of fully
autonomous vehicles, hinges on consumer demand for these technologies.
We have taken an initial attempt to quantify how households currently perceive
and value these technologies. Our work has combined current discrete choice
experimental methodologies with recent developments in the discrete choice literature
to quantify how much households are willing to pay for multiple levels of automation.
Methodologically, we highlight the importance of allowing for flexible distribu-
tions of preferences for vehicle attributes such as automation by comparing estimates
from a standard mixed logit specification with a more flexible mixed-mixed logit spec-
ification. We find richer heterogeneity estimates with the more flexible specification,
where demand for automation appears evenly split between high, modest and no
demand.
We find that households vary widely in their valuation of the technology. Some
are not willing to pay anything for either type. Others that are more knowledgeable
about current abilities of automation are willing to pay a great deal for full
automation; we estimate that a nontrivial fraction of households are willing to pay
above $10,000 for full automation.
We suggest proceeding in this area of research with caution, given our estimates
and the highly diverse preferences for automation as evidenced by the extremely
large standard deviations of the random parameters. However, we expect to see less
18
extreme heterogeneity as automated technology matures in the market, knowledge
of the technology spreads, and consumers learn about its benefits and costs
Acknowledgments
This research is based upon work supported by the National Science Foundation
Award No. CMMI-1462289.
19
References
Al-Alawi, B. and Bradley, T. (2013). Review of hybrid, plug-in hybrid, and electric
vehicle market modeling studies. Renewable and Sustainable Energy Reviews,
21:190–203.
Allcott, H. and Greenstone, M. (2012). Is there an energy efficiency gap? Journal
of Economic Perspectives, 14(5):3–28.
Allcott, H. and Wozny, N. (2014). Gasoline prices, fuel economy, and the energy
paradox. Review of Economics and Statistics, 96:779–795.
Ansar, J. and Sparks, R. (2009). The experience curve, option value, and the energy
paradox. Energy Policy, 37(3):1012–1020.
Bansal, P., Kockelman, K., and Singh, A. (2016). Assessing public opinions of
an interest in new vehicle technologies: An Austin perspective. Transportation
Research Part C.
Bujosa, A., Riera, A., and Hicks, R. L. (2010). Combining discrete and
continuous representations of preference heterogeneity: A latent class approach.
Environmental and Resource Economics, 47(4):477–493.
Busse, M., Knittel, C., and Zettelmeyer, F. (2013). Are consumers myopic? Evidence
from new and used car purchases. American Economic Review, 103(1):220–256.
Cameron, T. A. and Gerdes, G. R. (2005). Individual subjective discounting: Form,
context, format, and noise. Unpublished manuscript, Department of Economics,
University of Oregon. Eugene, OR.
Coller, M. and Williams, M. B. (1999). Eliciting individual discount rates.
Experimental Economics, 2(2):107–127.
DeCanio, S. J. (1998). The efficiency paradox: Bureaucratic and organizational
barriers to profitable energy-saving investments. Energy Policy, 26(5):441–454.
20
Fagnant, D. and Kockelman, K. (2014). The travel and environmental implications
of shared autonomous vehicles, using agent-based model scenarios. Transportation
Research Part C: Emerging Technologies, 40:1–13.
Fagnant, D. and Kockelman, K. (2015). Preparing a nation for autonomous vehicles:
Opportunities, barriers and policy implications. Transportation Research Part A,
77:167–181.
Frederick, S., Loewenstein, G., and O’Donoghue, T. (2002). Time discounting and
time preference: A critical review. Journal of Economic Literature, pages 351–401.
Greenblatt, J. and Saxena, S. (2015). Autonomous taxis could greatly reduce
greenhouse-gas emissions of US light-duty vehicles. Nature Climate Change, 5:860–
865.
Greene, D. (2010). How Consumers Value Fuel Economy: A Literature Review.
Office of Transportation and Air Quality, U.S. Environmental Protection Agency,
Report EPA-420-R-10-008.
Greene, D., Evans, D., and Hiestand, J. (2013). Survey Evidence on the Willingness
of U.S. Consumers to Pay for Automotive Fuel Economy. Energy Policy, 61:1539–
1550.
Greene, W. H. and Hensher, D. A. (2013). Revealing additional dimensions of
preference heterogeneity in a latent class mixed multinomial logit model. Applied
Economics, 45(14):1897–1902.
Hassett, K. A. and Metcalf, G. E. (1993). Energy conservation investment: Do
consumers discount the future correctly? Energy Policy, 21(6):710–716.
Hausman, J. A. (1979). Individual discount rates and the purchase and utilization
of energy-using durables. The Bell Journal of Economics, pages 33–54.
Howard, D. and Dai, D. (2013). Public perceptions of self-driving cars: The case of
berkeley, california. Annual Meeting of the Transportation Research Board.
21
Jaffe, A. B. and Stavins, R. N. (1994). The energy paradox and the diffusion of
conservation technology. Resource and Energy Economics, 16(2):91–122.
Keane, M. and Wasi, N. (2013). Comparing alternative models of heterogeneity in
consumer choice behavior. Journal of Applied Econometrics, 28(6):1018–1045.
Kirby, K., Petry, N., and Bickel, W. (1999). Heroin addicts have higher discount
rates for delayed rewards than non-drug-using controls. Journal of Experimental
Psychology: General, 128:78–87.
Kyriakidis, M., Happee, R., and de Winter, J. (2015). Public opinion on automated
driving: Results of an international questionnaire among 5000 respondents.
Transportation Research Part F, 32:127–140.
Lave, C. A. and Train, K. (1979). A disaggregate model of auto-type choice.
Transportation Research Part A: general, 13(1):1–9.
Lu, S. (2006). Vehicle Survivability and Travel Mileage Schedules. NHTSA technical
report, Department of Economics, UCB.
National Highway Traffic Safety Administration (2013). Preliminary statement of
policy concerning automated vehicles.
Newell, R. G. and Siikam¨aki, J. V. (2013). Nudging energy efficiency behavior: The
role of information labels. National Bureau of Economic Research(working paper
No. 19224).
Payre, W., Cestac, J., and Delhomme, P. (2014). Intention to use a fully automated
car: Attitudes and a priori acceptability. Transportation Research Part F, 27:252–
263.
Rezvani, Z., Jansson, J., and Bodin, J. (2015). Advances in consumer electric vehicle
adoption research: A review and research agenda. Transportation Research Part
D: Transport and Environment, 34:122–136.
22
Sallee, J., West, S., and Fan, W. (2016). Do consumers recognize the value of fuel
economy? Evidence from used car prices and gasoline price fluctuations. Journal
of Public Economics.
Schoettle, B. and Sivak, M. (2014). Public opinion about self-driving vehicles in
China, India, Japan, the U.S., the U.K., and Australia. University of Michigan
Transportation Research Institute, pages 1–31.
Train, K. (1985). Discount rates in consumers’ energy-related decisions: a review of
the literature. Energy, 10(12):1243–1253.
Train, K. E. (2008). EM algorithms for nonparametric estimation of mixing
distributions. Journal of Choice Modelling, 1(1):40–69.
Van Soest, D. P. and Bulte, E. H. (2001). Does the energy-efficiency paradox exist?
Technological progress and uncertainty. Environmental and Resource Economics,
18(1):101–112.
Wang, C. and Daziano, R. (2015). On the problem of measuring discount rates in
intertemporal transportation choices. Transportation, 42(6):1019–1038.
23
A Tables
Table 1: Sample Demographic Statistics
Variable Mean (S.D.)
Household size 2.717 (1.32)
Age of respondent 47.565 (13.55)
Number of children 1.41 (1.36)
Household income (2014$) 61,226 (42,135)
Years respondent has held license 25.409 (9.98)
Number of household members with license 1.914 (0.74)
Number of vehicles held by household 1.592 (0.79)
Respondent daily one-way commute (miles) 13.903 (12.72)
Respondent characteristics Percentage
Male 50.49
Female 49.51
Married 54.49
Widowed 2.94
Divorced 13.70
Single 21.45
Living with partner 7.42
White 85.24
Black 8.32
Hispanic 7.18
Asian 2.934
High school diploma 98.613
Some college experience 76.84
Bachelor’s degree 38.25
Master’s or professional degree 12.40
Full-time (30 hours per week) job 66.40
Part-time job 8.64
Homemaker 7.83
Student 0.90
Retired 10.44
Unemployed but actively looking for work 5.79
Household income $30,000 22.43
Household income >$30,000 and $60,000 34.01
Household income >$60,000 and $90,000 23.82
Household income >$90,000 19.74
Note: The white, black, Hispanic and Asian percentages sum to more
than 100 percent because some of the respondents have multicultural
backgrounds.
24
Table 2: Vehicle Holdings Statistics
Variable Coupe Convertible Sedan Wagon Hatchback CUV Truck SUV Van Average
Age 5.93 8.14 6.57 6.13 7.71 4.43 10.36 8.32 6.77 7.00
(4.46) (3.28) (4.82) (4.39) (4.95) (3.48) (4.74) (4.41) (3.49) (4.76)
Miles per gallon 26.98 23.95 27.89 26.68 29.82 23.68 19.77 18.76 20.97 24.99
(4.19) (2.69) (5.71) (3.92) (4.64) (2.67) (2.56) (2.40) (1.16) (5.69)
Weight (lb.) 3,044.3 3,505.7 3,184.0 3,080.6 2,759.8 3,644.9 3,900.8 4,257.5 4,223.4 3,456.6
(441.8) (762.6) (458.2) (310.3) (459.6) (523.0) (647.2) (690.9) (309.2) (682.3)
Horsepower 185.74 195.36 171.88 159.80 146.91 196.87 193.33 224.46 219.74 186.56
(72.37) (80.12) (50.40) (32.58) (58.77) (48.67) (52.53) (53.96) (40.98) (57.75)
Torque (lb.-ft.) 182.99 219.14 177.99 163.87 157.02 202.62 230.66 250.53 232.39 197.82
(72.74) (128.39) (56.27) (41.50) (65.27) (63.28) (59.06) (59.66) (32.32) (66.68)
Footprint (sq. in.) 7,455.7 7,181.9 7,644.6 7,117.1 7,004.1 7,814.7 9,041.3 8,365.3 9,064.5 7,906.5
(508.7) (696.5) (615.5) (505.0) (565.8) (675.0) (1356.5) (1005.2) (389.0) (959.6)
Real MSRP (2014$) 31,336 58,654 26,197 24,446 21,396 27,695 23,869 33,715 29,634 27,980
(36,990) (98,073) (11,810) (8,788) (9,656) (7,203) (5,024) (8,026) (4,328) (19,616)
Annual VMT 14,225 13,571 15,018 16,167 14,031 14,846 16,828 16,884 16,908 15,236
(10,994) (15,542) (12,738) (10,516) (14,371) (11,490) (15,144) (13,787) (13,546) (12,869)
Holdings share (%) 13.23 1.17 40.79 1.26 4.10 12.23 11.06 11.39 4.77 100
Notes: We assign trim-level miles per gallon, weight, horsepower, torque, footprint and Real MSRP based on vehicle
characteristics from Ward’s Automotive and inflation rates from the Bureau of Labor Statistics. Vehicle age and annual VMT
are based on two questions in our survey. With-in group standard deviation of each variable are reported in parentheses below
the mean. Real MSRP is the Manufacturer’s suggested retail price of a brand new version of the vehicle, adjusted for inflation.
Since the vehicles held are of different model years, we convert all prices to 2014 dollars using historical inflation rates. Annual
VMT stands for annual vehicle miles traveled. Vehicle footprint is defined as the product of a vehicle’s wheelbase and its width,
which is approximately equal to the rectangular area between a vehicle’s tires. The characteristics weight, torque, and footprint
are measured in pounds, pounds per fo ot, and square inches, respectively.
25
Table 3: Attributes and Attribute Levels for the Vehicle Choice Experiment
Attribute Levels
Cost to drive 100 miles [$] HEV: 7.0, 8.8
PHEV: 5.5, 6.5
BEV: 3.2, 4.0
GAS: 15.2, 15.8
Purchase price [$] HEV: 125%, 170% of referencea
PHEV: 145%, 185% of reference
BEV: 130%, 200% of reference
GAS: 100%, 110% of reference
Electric driving range [miles] PHEV: 15, 40
BEV: 80, 150
Recharging time [hours] PHEV: 2, 4
BEV: 1.5, 8
Autopilot package No automation, some automation, full automation
Notes: The four alternatives available for the respondent to choose include a hybrid electric vehicle (HEV), a plug-in
hybrid electric vehicle (PHEV), a battery electric vehicle (BEV), and a gasoline vehicle (GAS). See Figure 1for an
image of what respondents saw when making the vehicle choice. The reference purchase price is calculated from
purchase price thresholds stated by the respondents.
26
Table 4: Conditional Logit Models
Montly cost Cost per mile PVFC
Est. SE Est. SE Est. SE
Parameter estimates
Price -0.076 0.004 -0.076 0.004 -0.076 0.004
ASC BEV -0.851 0.098 -1.303 0.242 -0.602 0.096
ASC HEV -0.068 0.053 -0.393 0.166 0.106 0.051
ASC PHEV -0.126 0.143 -0.427 0.201 0.022 0.142
Ocost -0.004 0.000
Cost -0.079 0.021
PVFC -0.242 0.049
log(range) 0.073 0.048 0.107 0.052 0.048 0.048
Charging time -0.002 0.009 -0.001 0.009 -0.003 0.009
Automation 1 0.267 0.034 0.267 0.034 0.266 0.034
Automation 2 0.372 0.033 0.372 0.034 0.371 0.033
Implied willingness to pay
Ocost -$48.17 5.209
Cost -$10.34 2.820
PVFC -$0.32 0.067
log(range) $ 9.72 6.549 $14.01 6.993 $6.29 6.392
Charging time -$31.58 116.409 -$7.49 115.664 -$43.95 114.980
Automation 1 $3,538.00 502.423 $3,498.31 494.890 $3,486.71 495.980
Automation 2 $4,916.76 541.311 $4,863.83 533.248 $4,850.39 532.820
LL -12,415 -12,466 -12,461
AIC 24,847 24,950 24,939
BIC 24,912 25,015 25,004
Notes: Price is in thousands of dollars, operating cost (Ocost) is in dollars, cost is in
cents, PVFC is in ten thousands of dollars, log(range) is the natural log of miles of
range, charging time is in hours, and automation 1 and automation 2 are dummies for
partial and full automation. The standard errors of the point estimates for WTPs are
obtained using delta method. WTP of log(Range) corresponds to the marginal WTP
for a baseline driving range of 100 miles. AIC is computed as 2LL + 2 ×K, and BIC
as 2LL + log(NT /J )K, where Kis the number of parameters. Bold estimates are
statistically significant at 5%.
27
Table 5: Models with Parametric Heterogeneity
MIXL-N MIXL-N-OH MIXL-LN-I MIXL-LN-II
Est. SE WTP % >0 Est. SE WTP % >0 Est. SE WTP % >0 Est. SE WTP % >0
Price -0.111 0.005 -0.111 0.005 -0.091 0.005 -0.073 0.004
ASC BEV 0.846 0.133 0.855 0.133 0.435 0.123 0.877 0.100
ASC HEV 0.385 0.065 0.403 0.065 -0.069 0.057 -0.403 0.056
ASC PHEV 1.767 0.207 1.773 0.207 1.302 0.193 2.408 0.110
PVFC -0.243 0.068 -$0.22 -0.230 0.068 -$0.21 -0.292 0.062 -$0.32 -0.346 0.063 -$ 0.47
log(Range) (µ)0.783 0.072 $70.57 77% 0.778 0.072 $69.96 77% 0.760 0.068 $ 83.07 79% 2.907 0.140 $ 34.51 53%
Charging time (µ)-0.210 0.018 -$1,893.52 28% -0.213 0.018 -$1,917.68 28% -0.182 0.016 -$1,986.86 28% -0.117 0.015 -$1,606.24 33%
Automation 1 (µ)0.161 0.047 $1,453.14 55% 0.063 0.144 $6,322.84 0.543 0.037 $956.32 0.537 0.037 $1,022.89
Automation 2 (µ)0.110 0.052 $989.75 52% 0.209 0.148 $10,188.95 0.830 0.034 $ 935.06 0.912 0.039 $1,046.15
log(range) (σ)1.058 0.031 $9,532.37 1.043 0.030 $9,372.10 0.941 0.028 7.053 0.822
Charging time (σ)0.354 0.015 $3,184.50 0.358 0.015 $3,214.68 0.313 0.014 0.264 0.014
Automation 1 (σ)1.360 0.071 $12,251.70 1.339 0.071 $12,039.98 0.681 0.058 0.696 0.060
Automation 2 (σ)1.792 0.069 $16,144.62 1.761 0.069 $ 15,827.21 1.908 0.281 2.305 0.400
Autom1×Google car? 0.245 0.086 $2,206.05
Autom1×Male -0.212 0.079 -$1,903.14
Autom1×log(income) 0.165 0.055 $1,483.92
Autom1×years driving/10 -0.138 0.038 -$1,241.47
Autom1×West 0.429 0.111 $ 3,852.15
Autom1×Midwest 0.481 0.101 $ 4,320.95
Autom1×Northeast 0.108 0.105 $972.40
Autom2×Google Car? 0.464 0.087 $4,167.07
Autom2×Male -0.164 0.081 -$1,476.46
Autom2×Log(Income) 0.225 0.054 $2,023.78
Autom2×Years Driving/10 -0.230 0.040 -$2,069.72
Autom2×West 0.205 0.115 $1,838.86
Autom2×Midwest 0.258 0.102 $2,322.13
Autom2×Northeast -0.184 0.109 -$1,651.46
LL -10,277 -10,242 -10,583 -10,610
AIC 20,579 20,538 21,192 21,245
BIC 20,673 20,733 21,286 21,339
Notes: Price is in thousand of dollars, PVFC in ten thousand of dollars, log(range) is the natural log of miles of range, and charging time is in hours. All models were
estimated holding price, PVFC and ASCs fixed. MIXL-N mo del assumes all the parameters as normally distributed. MIXL-N-OH mo del assumes all the parameters as
normally distributed and Automation 1 and 2 normally distributed with observed heterogeneity. MIXL-LN-I model assumes log(range) and charging time as normally
distributed, whereas Automation 1 and 2 as log-normally distributed. MIXL-LN-II assumes charging time as normally distributed, whereas log(range) Automation 1 and 2
as log-normally distributed. The mean and SD of log normally distributed parameters represent the point estimates of µand σ, where βiLN(µ, σ). The standard errors
of the point estimates for log-normally distributed parameters are obtained using delta method. WTP of log(range) corresp onds to the marginal WTP for a baseline driving
range of 100 miles. WTP of log-normally distributed parameter evaluated at the median. 500 Halton draws for each individual were used to simulate the probability. AIC is
computed as 2LL + 2 ×K, and BIC as 2LL + log(NT /J )K, where Kis the number of parameters. Bold estimates are statistically significant at 5%.
28
Table 6: Models with No Parametric Heterogeneity
Class 1 Class 2 Class 3
Est. SE WTP %>0Est. SE WTP %>0Est. SE WTP %>0
A: Mean and standard deviations
Price -0.119 0.018 -0.043 0.005 -0.227 0.012
ASC BEV -2.328 0.580 1.796 0.157 -0.624 0.260
ASC HEV -2.663 0.204 1.151 0.117 2.788 0.160
ASC PHEV -2.320 0.973 2.256 0.221 1.422 0.392
PVFC -0.747 0.126 -$0.63 -0.344 0.153 -$0.80 -0.168 0.137 -$0.07
log(Range) 0.685 0.339 $57.56 0.039 0.067 $9.13 0.022 0.137 $0.99
Automation 1 (µ) -0.072 0.325 -$607.49 48% 0.119 0.051 $2,784.15 100% 0.269 0.082 $1,186.76 67%
Automation 2 (µ) -0.506 0.347 -$4,248.05 38% 0.281 0.053 $6,580.23 66% 0.322 0.083 $1,422.42 59%
Automation 1 (σ)1.231 0.300 $10,342.25 0.022 1.462 $508.26 0.609 0.144 $2,686.01
Automation 2 (σ)1.612 0.276 $13,541.52 0.663 0.088 $15,526.13 1.469 0.103 $6,484.70
B: Variables for class assignment
Constant -0.193 0.288 -1.559 0.270
Days 80 miles 0.128 0.026 0.080 0.025
Own a car? 0.596 0.156 0.192 0.133
No. of vehicles 0.211 0.065 0.133 0.064
Google car? 0.794 0.067 0.407 0.068
Age/ 10 -0.041 0.045 0.187 0.045
Male 0.106 0.064 -0.293 0.063
Married 0.123 0.065 0.333 0.063
No. of children 0.120 0.024 0.041 0.022
Comp. college 0.212 0.075 -0.034 0.070
High school -0.429 0.084 -0.615 0.080
Single family -0.227 0.118 -0.091 0.126
Apartment -0.783 0.148 0.269 0.146
Own a house -0.040 0.079 0.012 0.079
Years driving / 10 -0.398 0.059 -0.450 0.062
Accident -0.168 0.059 0.215 0.058
Prefer driving 0.299 0.069 0.451 0.066
Fulltime -0.163 0.088 0.124 0.083
Part time -0.232 0.123 0.069 0.117
Homemaker -0.515 0.129 -0.759 0.129
White -0.130 0.083 0.325 0.090
Conservative -0.415 0.068 0.154 0.066
Liberal -0.056 0.078 0.370 0.078
West 0.462 0.088 0.320 0.084
Midwest 0.480 0.080 0.474 0.074
Northeast 0.302 0.080 -0.162 0.079
Urban 0.163 0.062 -0.181 0.060
Log(income) -0.013 0.046 0.196 0.049
Shares for classes 29% 33% 38%
LL -9,075.5
AIC 18,323
BIC 18,941
Notes: Price is in thousands of dollars, PVFC in ten thousands of dollars, log(range) is the natural log of miles of range, and charging
time is in hours. WTP of log(range) corresponds to the marginal WTP for a baseline driving range of 100 miles. 500 Halton draws
for each individual were used to simulate the probability. AIC is computed as 2LL + 2 ×K, and BIC as 2LL + log (NT /J )K,
where Kis the number of parameters. Bold estimates are statistically significant at 5%.
29
B Figures
Figure 1: Sample of a Choice Situation Presented to Respondents
30
Figure 2: Willigness to Pay Distribution for MIXL Models
−40 −20 0 20 40
0.00 0.05 0.10 0.15
Automation 1
MIXL−N
MXIL−N−OH
MIXL−LN−I
MIXL−LN−II
−60 −40 −20 0 20 40 60
0.00 0.05 0.10 0.15
Automation 2
MIXL−N
MIXL−N−OH
MIXL−LN−I
MIXL−LN−II
Notes: The horizontal axis measures WTP in thousands of dollars. Observed
heterogeneity is evaluated at mean of variables.
31
Figure 3: Willigness to Pay Distribution for MM-MNL Model
−20 −10 0 10 20
0.0 0.2 0.4 0.6 0.8
Automation 1
WTP / 1000
Density
Class 1
Class 2
Class 3
−40 −20 0 20 40
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07
Automation 2
WTP / 1000
Density
Class 1
Class 2
Class 3
Notes: The horizontal axis measures WTP in thousands of dollars. Observed
heterogeneity is evaluated at mean of variables.
32
... AV uptake is a complex issue and a highly convenient mobility option comes with negative impacts as well, some of which are hard to anticipate at the moment: greater obesity, loss of competencies and skills (Bazilinskyy et al., 2015), over reliance on machines (Trimble et al., 2008; Howard & Dai, 2015). The range of individual concerns already expressed include: lack of trust in the capabilities of AV and their networking (Fraedrich & Lenz, 2014); specific risks for crashes (Daziano et al., 2016) that may be generated by non-AV traffic participants (Bazilinskyy et al., 2015), hacking of the systems (Fagnant & Kockelman, 2015), data transfer to third parties, deprivation from the joy of driving (Fagnant & Kockelman, 2015; Kyriakidis et al., 2015). Carlson et al. (2014) report that trust increases with the past performance of the system, with research on reliability or validation of the system, and through the reputation of the designer and manufacturer of the system; still, numerous trust issues are still unresolved and under review (Howard & Dai, 2015; Kyriakidis et al., 2015). ...
... Several large sample public opinion studies have highlighted positive attitudes (Payre et al., 2014), as well as " non-negligible level of reluctance " (Kyriakidis et al., 2015: 128). This mixture/ambivalence towards autonomous vehicles is expected considering the uncertainties surrounding the technology (Daziano et al., 2016). However, level of awareness and cultural differences, have led to a wide range of attitudes around the globe. ...
... For example, people in Germany and China were more aware of the AVs compare to those in Japan (Sommer, 2013) and more US respondents have heard about AVs compared to the UK and Australian counterparts (Schoettle & Sivak, 2014a). What is more, using a nation-wide online panel, Daziano et al. (2016) drew attention to the substantial heterogeneity in preferences for automation of the US respondents, with a substantial share of the sample willing to pay more than $10,000 for automation, when an equally large number is unwilling to pay any amount. This suggests that flexible policies may be required for a successful adoption of the technology. ...
Conference Paper
Full-text available
Autonomous vehicle technology and its potential effects on traffic and daily activities is a popular topic in the media and in the research community. It is anticipated that AVs will reduce accidents, improve congestion, increase the utility of time spent travelling and reduce social exclusion. However, knowledge about the way in which AVs will function in a transport system is still modest and a recent international study showed a lower familiarity with AVs in Australia compared to USA and UK. Attitudes towards fully automated driving (or higher levels of autonomy) range from excitement to suspicion. The breadth of feelings may be due to the low level of awareness or reflect polarising attitudinal positions. Whilst experts appear to be more confident about the adoption of AV technology in the near future, public acceptance is key to AVs’ market success. Hence, research that examines local contexts and opinions is needed. This paper reviews existing scholarly work and identifies gaps and directions for future developments, with a focus on the Australian context. The review will address the following broad categories: investigation of AV features and mobility models, implications for road traffic and connectivity to infrastructure (especially in low to medium density urban areas), public attitudes and concerns, potential business models, and policy implications. The aims of the paper are to identify critical issues for the development of a focus group inquiry to understand attitudes of potential users of AVs and to highlight AV development issues for policy makers in Australia.
... AV uptake is a complex issue and a highly convenient mobility option comes with negative impacts as well, some of which are hard to anticipate at the moment, including: greater obesity, loss of competencies and skills (Bazilinskyy et al. 2015), and over reliance on machines (Trimble et al. 2014;Howard & Dai 2014). The range of individual concerns already expressed include: lack of trust in the capabilities of AV and their networking (Fraedrich & Lenz 2014); specific risks for crashes (Daziano, Sarrias & Leard 2016) that may be generated by non-AV traffic participants (Bazilinskyy et al. 2015), hacking of the systems (Fagnant & Kockelman 2015), data transfer to third parties, and deprivation from the joy of driving (Fagnant & Kockelman 2015;Kyriakidis et al. 2015). Carlson et al. (2014) report that trust increases with the past performance of the system, with research on reliability or validation of the system, and through the reputation of the designer and manufacturer of the system; still, numerous trust issues remain unresolved and under review (Howard & Dai 2014;Kyriakidis et al. 2015). ...
... Several large sample public opinion studies have highlighted positive attitudes (Payre, Cestac & Delhomme 2014), as well as 'non-negligible level of reluctance' (Kyriakidis et al. 2015: 128). This mixture/ ambivalence towards AVs is expected considering the uncertainties surrounding the technology (Daziano et al. 2016). However, levels of awareness and cultural differences, have led to a wide range of attitudes around the globe. ...
... Evidently, there may also be considerable heterogeneity in preferences for automation. For instance, using a nationwide online panel, Daziano et al. (2016) found a substantial share of the sample willing to pay more than $10 000 for automation, with an equally large number unwilling to pay any amount. This suggests that flexible policies may be required for a successful adoption of the technology. ...
Article
Full-text available
Autonomous vehicle technology and its potential effects on traffic and daily activities is a popular topic in the media and in the research community. It is anticipated that AVs will reduce accidents, improve congestion, increase the utility of time spent travelling and reduce social exclusion. However, knowledge about the way in which AVs will function in a transport system is still modest and a recent international study showed a lower familiarity with AVs in Australia compared to the USA and UK. Attitudes towards fully automated driving (or higher levels of autonomy) range from ‘excitement’ to ‘suspicion’. The breadth of feelings may be due to the low level of awareness or reflect polarising attitudinal positions. Whilst experts appear to be more confident about the adoption of AV technology in the near future, public acceptance is key to AVs’ market success. Hence, research that examines local contexts and opinions is needed. This paper reviews existing scholarly work and identifies gaps and directions for future developments, with a focus on the Australian context. The review will address the following broad categories: investigation of AV features and mobility models, implications for road traffic and connectivity to infrastructure (especially in low to medium density urban areas), public attitudes and concerns, travel behaviour and demand, potential business models, and policy implications. The aims of the paper are to identify critical issues for the development of a focus group inquiry to understand attitudes of potential users of AVs and to highlight AV development issues for policy makers in Australia.
... After rapid technological developments especially over the past decade, autonomous vehicle (i.e. selfdriving) technology is expected to be ready for wide deployment in the near future with large implications for urban mobility [8,9,10,11]. It is generally accepted that one of the main benefits of self-driving cars could be reduced road congestion, as current roads are expected to have much higher capacity if the majority of traffic is autonomous vehicles [12]. ...
... We note that drawing conclusions for the more wide-spread adoption of shared vehicles is not straighforward, since the car-shared users who were the subjects of these studies might not form a representative sample of the general population. Further studies try to estimate public attitude toward mobiltiy options represented by self-driving vehicles and estimate the potential for adoption based on these [9,10,11]. Several studies then try to estimate the fleet size which could serve a certain population, operational parameters and the associated costs for travelers. Studies based on randomly generated trips find that about 10%-15% of cars could serve mobility demands compared to private vehicles, with significantly reduced costs when compared to either privately owned cars or taxi rides [17,18]. ...
Article
The increasing availability and adoption of shared vehicles as an alternative to personally-owned cars presents ample opportunities for achieving more efficient transportation in cities. With private cars spending on the average over 95\% of the time parked, one of the possible benefits of shared mobility is the reduced need for parking space. While widely discussed, a systematic quantification of these benefits as a function of mobility demand and sharing models is still mostly lacking in the literature. As a first step in this direction, this paper focuses on a type of private mobility which, although specific, is a major contributor to traffic congestion and parking needs, namely, home-work commuting. We develop a data-driven methodology for estimating commuter parking needs in different shared mobility models, including a model where self-driving vehicles are used to partially compensate flow imbalance typical of commuting, and further reduce parking infrastructure at the expense of increased traveled kilometers. We consider the city of Singapore as a case study, and produce very encouraging results showing that the gradual transition to shared mobility models will bring tangible reductions in parking infrastructure. In the future-looking, self-driving vehicle scenario, our analysis suggests that up to 50\% reduction in parking needs can be achieved at the expense of increasing total traveled kilometers of less than 2\%.
... Recent studies on autonomous mobility have focused on either assessing public opinions regarding AVs or studying the potential impacts of introducing such a new service on existing urban mobility systems. In a recent study, Daziano et al., 2016 Therefore, the study of the potential impacts of introducing AVs in urban mobility systems associated with assessing their sharing potential are at the heart of nowadays research regarding autonomous mobility. Although research on studying the impacts of deploying AVs has been growing in popularity in recent years, less amount of research on how to plan and operate their trips is yet available. ...
Thesis
Full-text available
The rise of research into shared mobility systems reflects emerging challenges, such as rising urbanization rates, traffic congestion, oil prices and environmental concerns. The operations research community has turned towards more sharable and sustainable systems of transportation. Although shared mobility comes with many benefits, it has some challenges that are restricting its widespread adoption. More research is thus needed towards developing new shared mobility systems so that a better use of the available transportation assets can be obtained.This thesis aims at developing efficient models and optimization approaches for synchronizing people and freight flows in an urban environment. As such, the following research questions are addressed throughout the thesis:*Q1: What are the variants of shared mobility systems and how to optimize them?*Q2: How can people trips be synchronized and what gains can this synchronization yields?*Q3: How can people and freight flows be combined and what are the intended benefits?*Q4: What impacts uncertainty can have on planning and operating shared mobility systems?First, we review different variants of the shared mobility problem where either (i) travelers share their rides, or (ii) the transportation of passengers and freight is combined. We then classify these variants according to their models, solution approaches and application context and We provide a comprehensive overview of the recently published papers and case studies. Based on this review, we identify two shared mobility problems, which we study further in this thesis.Second, we study a ridesharing problem where individually-owned and on-demand autonomous vehicles (AVs) are used for transporting passengers and a set of meeting points is used for synchronizing their trips. We develop a two-phase method (a pre-processing algorithm and a matching optimization problem) for assessing the sharing potential of different AV ownership models, and we evaluate them on a case study for New York City.Then, we present a model that integrates freight deliveries to a scheduled line for people transportation where passengers demand, and thus the available capacity for transporting freight, is assumed to be stochastic. We model this problem as a two-stage stochastic problem and we provide a MIP formulation and a sample average approximation (SAA) method along with an Adaptive Large Neighborhood Search (ALNS) algorithm to solve it. We then analyze the proposed approach as well as the impacts of stochastic passengers demand on such integrated system on a computational study.Finally, we summarize the key findings, highlight the main challenges facing shared mobility systems, and suggest potential directions for future research.
Article
The growth in technology on autonomous transportation systems is currently motivating a number of research initiatives. This paper first presents a survey of the literature on autonomous marine vessels in general. By identifying the main research interests in this field, we define 10 thematic categories. The collected articles are then classified according to these categories. We show that research on autonomous vessels has increased dramatically in the past decade. However, most of the published articles have focused on navigation control and safety issues. Studies regarding other topics, such as transport and logistics, are very limited. Although our main interest is the literature on autonomous vessels, we contrast its development with respect to the literature on autonomous cars so as to have a better understanding about the future potentials in the research on autonomous vessels. The comparison shows that there are great opportunities for research about transportation and logistics with autonomous vessels. Finally, several potential research areas regarding logistics with autonomous vessels are proposed. As the technology behind remote‐controlled or autonomous ships is maturing rapidly, we believe that it is already time for researchers in the field to start looking into future water‐borne transport and logistics using autonomous vessels.
Article
With the recent progress of techniques in computer vision and processor design, vehicles are able to perform a greater number of functions, and are reaching higher levels of autonomy. As the list of autonomous tasks that the car is supposed to perform grows, two design questions arise: how to group these tasks into modules and which processors and data buses should instantiate these modules and their links in the physical architecture. Both questions are linked, as the processing capacity of the processors influences how centralized the architecture can be, and the modularization influences the overall system latency as well. Furthermore, our interest lies in designing architectures that perform the tasks rapidly, while minimizing cost. This multiobjective problem is intractable without architecture exploration and an analysis tool. This paper presents a linear optimization formulation to capture these tradeoffs, and to systematically find relevant architectures with optimal latency and cost. The results show that enforcing all safety constraints on the architecture leads to a worst case increase of 17% in latency and 18% component cost per vehicle. The increase in latency is significant at the scale of human driver reaction times.
Article
Full-text available
Emerging advancements in automated vehicles may dramatically change how transportation system users engage cars as symbols and tools to exert control over their lives, time use, and activity spaces. Transportation technologies do not only serve users. They can also elicit affective reactions and make users experience feelings, including anxiety, freedom, health, and the sense of being in control of different facets of their lives. By changing the role of humans in actively moving through urban environments, fully-automated vehicles could impact users’ perceived control over their environments and their use of cities. To better understand the use and implications of automated vehicles, this study explores whether affective motivations for automobile ownership are likely to change in a potential future of fully-automated vehicles. This study focuses on AV ownership conditions using a consumer survey of 3201 residents of the Greater Toronto and Hamilton Area to estimate consumer interest in regularly using and purchasing privately-owned AVs using a combination of revealed preference (vehicle ownership) and stated preference (AV interest) data. Structural equation models are estimated to explore whether the affective motivations for vehicle ownership are likely to be different than the affective motivations for using or adopting an automated vehicle. Findings indicate that the role of affective motivations are likely to be different for vehicle ownership than for interest in AV use and adoption. While a car freedom and control affective motivation is a strong predictor of current household vehicle ownership, it is negatively associated with AV interest and adoption. The affective motivation related to being a daring achiever is a strong predictor of AV interest and adoption, but it is unrelated with current vehicle ownership levels. Finally, individual control over activities and travel is negatively associated with current vehicle ownership levels, but it is unrelated with AV interest and adoption. Findings suggest significant differences between individuals’ motivations for vehicle ownership and AV adoption and use. Nevertheless, these findings should be interpreted cautiously, as data related to interest in AV technology remains in the domain of stated preference data, rather than revealed preferences, which are subject to broader household decision-making processes. As such, if transportation planning and forecasting concentrate only on instrumental factors and ignore changes in affective motivations, forecasts may miss key components of behavioral change and may miss the balance of who benefits (or loses) from this new technology.
Article
Full-text available
This study investigated public opinion of persons who are blind or visually impaired related to user acceptance, concerns, and willingness to buy partially and fully automated vehicles; commonly referred to as self-driving vehicles. A 39-question Internet-based survey was distributed in the United States that collected 516 useable responses from persons 18 years and older who self-identified as visually impaired. Respondents generally had an optimistic view of the potential benefits of self-driving vehicles however a majority of respondents expressed concerns regarding the technology. Concerns focused on the possibility of equipment failure, vehicles getting confused by unexpected situations and vehicle interactions with pedestrians and bicycles. A majority of respondents agreed that the needs of individuals who are blind or visually impaired are being considered in the development of self-driving vehicle technology though higher education levels were associated with a decrease in a respondent's belief in this contention.
Article
Full-text available
Technological advances are bringing connected and autonomous vehicles (CAVs) to the ever-evolving transportation system. Anticipating public acceptance and adoption of these technologies is important. A recent internet-based survey polled 347 Austinites to understand their opinions on smart-car technologies and strategies. Results indicate that respondents perceive fewer crashes to be the primary benefit of autonomous vehicles (AVs), with equipment failure being their top concern. Their average willingness to pay (WTP) for adding full (Level 4) automation ($7253) appears to be much higher than that for adding partial (Level 3) automation ($3300) to their current vehicles. Ordered probit and other model specifications estimate the impact of demographics, built-environment variables, and travel characteristics on Austinites’ WTP for adding various automation technologies and connectivity to their current and coming vehicles. It also estimates adoption rates of shared autonomous vehicles (SAVs) under different pricing scenarios ($1, $2, and $3 per mile), choice dependence on friends’ and neighbors’ adoption rates, and home-location decisions after AVs and SAVs become a common mode of transport. Higher-income, technology-savvy males, who live in urban areas, and those who have experienced more crashes have a greater interest in and higher WTP for the new technologies, with less dependence on others’ adoption rates. Such behavioral models are useful to simulate long-term adoption of CAV technologies under different vehicle pricing and demographic scenarios. These results can be used to develop smarter transportation systems for more efficient and sustainable travel.
Article
Full-text available
Users of transportation systems face economic decisions that involve cash flows in different points in time. In making these intertemporal choices, users tradeoff between immediate and future benefits and costs. For example, battery electric vehicles require a higher investment but offer important energy savings that may pay off. Using an appropriate discount rate, future costs are transformed into a present equivalent that can be directly compared with upfront expenses or known prices. In this paper we review different methods for measuring individual discount rates that have been used in the literature to discuss, in particular, the findings of the " energy paradox " or undervaluation of future savings (making more difficult the adoption of energy efficient vehicles). We identify two main methods of treating discount rates in discrete choice models, namely endogenous and exogenous discounting. Exogenous discounting addresses some problems of the endogenous discounting method (attribute correlation, endogeneity, and imposition of rational evaluation of future costs), but lacks correct treatment of unobserved heterogeneity. We also review experimental elicitation of intertem-poral preferences that accounts for heterogeneous discounting. In particular, we tested a modified version of a reward-overtime choice experiment that can be easily incorporated into transportation surveys. In addition, we compared the experimental discount rates with those obtained from a contingent valuation exercise. The continent valuation exercise not only provided a much higher mean for the discount rate (13.93% was the mean experimental discount rate, whereas the contingent valuation mean was 93.9%), but also the level of variability among respondents was unexpectedly high (6.4%-267%).
Article
Full-text available
Autonomous vehicles (AVs) represent a potentially disruptive yet beneficial change to our transportation system. This new technology has the potential to impact vehicle safety, congestion, and travel behavior. All told, major social AV impacts in the form of crash savings, travel time reduction, fuel efficiency and parking benefits are estimated to approach $2000 to per year per AV, and may eventually approach nearly $4000 when comprehensive crash costs are accounted for. Yet barriers to implementation and mass-market penetration remain. Initial costs will likely be unaffordable. Licensing and testing standards in the U.S. are being developed at the state level, rather than nationally, which may lead to inconsistencies across states. Liability details remain undefined, security concerns linger, and without new privacy standards, a default lack of privacy for personal travel may become the norm. The impacts and interactions with other components of the transportation system, as well as implementation details, remain uncertain. To address these concerns, the federal government should expand research in these areas and create a nationally recognized licensing framework for AVs, determining appropriate standards for liability, security, and data privacy.
Article
Debate about the appropriate design of energy policy hinges critically on whether consumers might undervalue energy efficiency, due to myopia or some other manifestation of limited rationality. We contribute to this debate by measuring consumers' willingness to pay for fuel economy using a novel identification strategy and high quality microdata from wholesale used car auctions. We leverage differences in future fuel costs across otherwise identical vehicles that have different current mileage, and therefore different remaining lifetimes. By seeing how price differences across high and low mileage vehicles of different fuel economies change in response to shocks to the price of gasoline, we estimate the relationship between vehicle prices and future fuel costs. Our data suggest that used automobile prices move one for one with changes in present discounted future fuel costs, which implies that consumers fully value fuel economy.
Article
Autonomous vehicles (AVs) are conveyances to move passengers or freight without human intervention. AVs are potentially disruptive both technologically and socially, with claimed benefits including increased safety, road utilization, driver productivity and energy savings. Here we estimate 2014 and 2030 greenhouse-gas (GHG) emissions and costs of autonomous taxis (ATs), a class of fully autonomous shared AVs likely to gain rapid early market share, through three synergistic effects: (1) future decreases in electricity GHG emissions intensity, (2) smaller vehicle sizes resulting from trip-specific AT deployment, and (3) higher annual vehicle-miles travelled (VMT), increasing high-efficiency (especially battery-electric) vehicle cost-effectiveness. Combined, these factors could result in decreased US per-mile GHG emissions in 2030 per AT deployed of 87-94% below current conventionally driven vehicles (CDVs), and 63-82% below projected 2030 hybrid vehicles, without including other energy-saving benefits of AVs. With these substantial GHG savings, ATs could enable GHG reductions even if total VMT, average speed and vehicle size increased substantially. Oil consumption would also be reduced by nearly 100%.