Preference elicitation for a dynamic simulation: powertrain
choices in the European Union car market
J. Gémez Vilchez"'*, G. Harrison’, C. Thiel't, H. Lu’, C. Rohr’, L. Kelleher’, A. Smyth®
*Corresponding author: +390332785318. E-mail: jonatan.gomez-vilchez@ ec.europa.eu
‘European Commission, Joint Research Centre (JRC), Ispra, Italy
“Institute for Transport Studies, University of Leeds, 34-40 University Road,
Leeds LS2 9JT, UK
5RAND Europe, 7 Westbrook Centre, Milton road, Cambridge CB4 1YG, UK
4gchool of Architecture, Planning and Environmental Policy, University College Dublin (UCD),
Richview, Dublin D14 E099, Ireland
5Centre for Sustainable Communities, University of Hertfordshire, 17 Hatfield,
Hertfordshire AL10 9AB, UK
‘The views expressed are purely those of the authors and may not in any circumstances be
regarded as stating an official position of the European Commission.
ABSTRACT
A nested logit model was estimated from a survey carried out among European car
owners. For the purpose of improving choice assumptions, this model was embedded
within the Powertrain Technology Transition Market Agent Model. This system
dynamics model focuses on vehicle powertrain uptake in the European Union. This
paper describes the modeling process and shows its application to German car sales
market shares until 2025. In conclusion, the integration of discrete choice frameworks
based on stated preference surveys into system dynamics models remains a useful
approach to explore empirically-grounded factors of technology adoption and feedback
processes.
Keywords: electric vehicles, stated preference survey, discrete choice model, system
dynamics, automotive
37th International Conference of the System Dynamics Society
Albuquerque, New Mexico (US), 2019
1. INTRODUCTION
In 2015, the transport sector emitted 1,048 megatonnes of CO2 equivalent (MtCO2eq) in
the European Union (EU). By 2050, EU transport emissions should not exceed 333
MtCOoeq (EEA, 2019). The EU vehicle market plays a crucial role in achieving that
goal. The uptake of low- and zero-emission vehicle (in particular electric vehicles
(EVs)) technologies is being facilitated mainly by COz2 emissions performance
standards (EU, 2009, 2017b), deployment of alternative fuels infrastructure (EU, 2014)
and financial incentives (ACEA, 2018) (EEA, 2018b).
To simulate the impact of these policy measures on the EU passenger car and light
commercial vehicle markets over time, the Powertrain Technology Transition Market
Agent Model (PTTMAM) was developed?. This is a system dynamics (SD) model
representing feedback structures and capturing the interactions of four agent groups:
users, manufacturers, infrastructure providers and authorities (Harrison, Thiel, & Jones,
2016). At the core of the model lies a key assumption, namely users’ powertrain choice.
Harrison and Thiel (2017) used PTTMAM to construct policy scenarios for the
Netherlands and the United Kingdom (UK). The authors acknowledged that “in future
development the choice model will be further refined to obtain more specific preference
parameters” (p. 37). Hence PTTMAM developers settled for integrating the utility
coefficients of a discrete choice (DC) model into the SD model. The needs of
PTTMAM could, to a certain extent, be accommodated from the outset in the survey
that underpins the DC model. To our knowledge, this is the first attempt to date at
designing and conducting a survey tailored to the requirements of an SD model focusing
on EV market uptake. The objective of this paper is to describe this modeling process
and the corresponding results. The focus of this study is on the car market.
The structure of the paper is as follows: section 2 provides a concise overview of the
literature, the survey and the resulting DC model are briefly described in section 3, in
section 4 the process through which the DC model was integrated into the SD model is
described, section 5 shows the results, and in section 6 conclusions are drawn.
! PTTMAM is available at: https://ec.europa.eu/jrc/en/pttmam. The model used for this paper is a
simplified version with updated data.
2. LITERATURE REVIEW
Consumer choice can be modeled using different methods with the most common ones
being, in the context of electric car deployment, diffusion of innovation theory, agent-
based modeling and discrete choice (DC) analysis. An overview of the former can be
found in Al-Alawi and Bradley (2013). Applied examples of the last two methods are
Gnann (2015) and Hackbarth and Madlener (2013), respectively. A multi-method
approach can also be identified in the literature: for example, whereas Kieckhéfer et al.
(2014) used German data to link agent-based modeling with SD, Jensen et al. (2016)
used Norwegian data to combine the diffusion and DC methods.
Embedding a DC model into an SD model is not entirely new (for the pioneering work
and a more recent example, see respectively Ford (1995) and Shepherd et al. (2012)).
From a review of this body of literature (see Gomez Vilchez and Jochem [under
review]), it can be concluded that in previous studies the development of the logit model
preceded the conceptualisation of the SD model, with the latter sometimes requiring
adaptations to accommodate the set of alternatives and/or attributes included in the
choice set of the former. In some cases, operations to reconcile both models (for an
example related to the units of measurement, see section 5.4.5 in Meyer (2009)) had to
be carried out. In contrast, in this work PTTMAM preceded the DC analysis.
3. STATED PREFERENCE SURVEY AND LOGIT MODEL
The survey sought to answer the following research question: which powertrain
technologies are EU consumers willing to adopt and how do they trade-off between
important attributes of electric and other cars? To this end, a stated preference (SP)
survey was designed and conducted in mid-2017 using an existing online panel by
computer-assisted web interviewing. The sample comprised a total of 1,248 car owners
from six EU countries: France, Germany, Italy, Poland, Spain and the United Kingdom.
The questionnaire and a description of the survey respondents can be found in Gomez
Vilchez et al. (2017). The survey built upon another survey that had been carried out in
2012 (Thiel et al. 2012). In contrast to the latter, the 2017 survey included two choice
experiments, from which a statistical model was estimated after pooling the data
(further details on the design and analysis can be found in Rohr et al. (2019)). The five
powertrain options offered in the second choice experiment (see Figure Al in the
Appendix) were: petrol, diesel, hybrid (with conventional or plug-in hybrid electric
vehicle (PHEV) as a variable), battery electric vehicles (BEVs) and fuel cell electric
vehicles (FCEVs).
Relying on random utility theory (refer to e.g. Ben-Akiva & Lerman (1985)), the model
that was estimated using the survey data was a special case of the Generalized Extreme
Value (GEV) model, namely the Nested Multinomial Logit (NMNL) model. The
formulation of the Multinomial Logit (MNL) model is improved by introducing a
nesting structure, thereby mitigating the undesirable impact of the Independence of
Irrelevant Altematives (IIA) property. Whereas the error terms of the alternatives are
independent and identically distributed (IID) within a nest, they are not across nests.
The nesting structure was empirically tested, finding higher elasticities between hybrids
and zero emission vehicles (ZEVs: BEVs and FCEVs) (‘low emissions’ nest) compared
to intemal combustion engine vehicles (ICEVs: petrol and diesel). In other words,
respondents perceive low emissions (hybrids) and ZEVs to be more “similar” to each
other and thus they are more likely to switch between these alternatives compared to the
petrol and diesel cars. The preferred nesting structure of the model can be seen in Fig. 1.
A statistically significant (at the 90% confidence level) @ value of 0.613 was estimated
for the low emissions nest (see Fig. 1). By lying between zero and one, this value is
consistent with the assumption of utility maximization. A value of one would mean that
the NMNL collapses into the MNL model. As it approaches zero the degree of
independence within a nest reduces, leading to increasing substitution within each nest.
All
Petrol Diesel is
emissions
Hybrid BEV FCEV
Figure 1. Preferred nesting structure of the logit model
Source: own work
The model was specified with nine car attributes (see Table A1 in the Appendix):
purchase price, hire purchase (HP) option, personal contract purchase (PCP) option,
operating cost, retained value (i.e. depreciation), range, re-fuelling/-charging time, level
of emissions and low emission car incentive. The estimated utility coefficients for each
of these attributes can be found in Table 2 in Rohr et al. (2019).
4, MODELING APPROACH
4.1 Reduction of subscript range
PTTMAM is a comprehensive model that disaggregates car technology demand by
country (28 Member States), vehicle category (passenger cars, light commercial
vehicles), user (private, public, fleet), geography (urban, non-urban), size (small,
medium, large) and powertrain (16 technologies). In addition, several vehicle attributes
are taken into account (see Harrison et al. (2016)). This leads to a complex formulation
of powertrain choice, which entered into conflict with the need to reduce the cognitive
burden to survey respondents.
Original NMNL Adapted
PTTMAM model PTTMAM
Biodiesel
ICEV
Bioethanol |
ICEV,
FCEV FCEV
Figure 2. Powertrain options, by model
*This powertrain is further disaggregated into gasoline, diesel, biodiesel and ethanol. Source: own work
Figure 2 shows how the ‘powertrain’ subscript array was simplified. The most
important changes in PTTMAM are the deletion of cars powered by biodiesel and the
substitution of bioethanol cars by flexible fuel vehicles (FFV: ethanol 85).
4,2 Extension of feedback loops related to battery attributes
One of the attributes included in the NMNL model is re-charging time for electric cars
(re-fuelling for the rest). This variable, however, was not explicitly considered in the
original version of PTTMAM. The possibility of simply creating an exogenous variable
was considered less satisfactory than the inclusion of an endogenous variable, as the
potential for representing additional feedback processes would in this way be exploited.
Specifically, three new variables were created (battery capacity [kWh/component],
electric range [km] and recharging time [minute]) and three new feedback loops
represented (see Fig. 3). This approach had been previously implemented in the model
by Gomez Vilchez (2019). For the component cost, the unit of measurement of the
battery component was modified from [euro/component] to [euro/kWh].
+ ————
EU passenger .
combined vty aciual vehicle price
+
recharging time
new regotations An) EU bert +
+
“original
PTTMAM"
Cc
‘Ttumulative component "3rd new loop"
electric range ‘manufacture
component cost
w
“2nd new loop" Le)
" Ist new loop"
battery capacity,
Figure 3. New feedback loops in PTTMAM
Note that this is a highly simplified CLD: various variables are usually present along the causal links
displayed here. Source: own work using Vensim®
4.3 Embedment of the discrete choice model within the system dynamics model
Next, the results of NMNL model were embedded within PTTMAM following
Equations 1-5 (see also Figure A2 in the Appendix).
For the NMNL model, the probability function can be written as two parts (logits):
P; = PiupPp (Eq. 1)
Among them, the conditional probability of choosing alternative i given that an
altemative in nest Bx is chosen is defined as below:
eM
yey
Lyjene J
i
Pig =
(Eq. 2)
The marginal probability of choosing an alternative in nest Bx is determined by:
Ok.Ik
e
P,
B= Teco (Eq. 3)
Then the “logsum” term, which brings information from the lower nest model to the
upper model is:
Ky = InQjew ev) (Eq. 4)
The observable part of the utility function Vv; for each powertrain/fuel type alternative is
written as:
Vspi = Le B_SPixsXix + Base (Eq. 5)
There are two components of the systematic utility coefficient: the coefficients from the
SP models, @_SP;,, that multiply the observed ‘k’ attribute values, i.e. Xj,. It is noted
that some of the coefficients vary across different segments. Specifically, purchase price
and operating cost coefficients vary by vehicle size. Range varies by vehicle type (one
term for ICEVs and diesel cars, one for low emission vehicles). Information on both is
required to run the model. We dropped the coefficient for left-choice bias, which is not
required for implementation of the model (it is included in the estimation of the model
to ensure that the resulting coefficients are not biased by such behaviour).
The concept of willingness-to-consider (WtC) a platform (i.e. powertrain), which was
present in PTTMAM as a model variable, was also dropped. The reason for this being
that the WtC term, which represents “the formation of a driver’s consideration set”
(Struben & Sterman, 2008: 1077), is incorporated implicitly in the DC approach though
the attribute weights and altemnative-specific constants (ASCs). The advantage of this is
that the policy analyst does not need to predict that variable, but rather can focus on car
attributes.
The utility equation for each vehicle type also requires an ASC, which reflects the
additional utility required for the utility for each car type to ensure that the model
incorporates attributes not measured in the choice experiment and replicates observed
market shares. We estimated ASCs from the SC data and these were found to vary by
age, education level and country. However, it is not appropriate to use constants from
SP exercises in forecasting for a number of reasons, including:
- These reflect the choices that were presented in the choice experiments, which may
not reflect real-world conditions (e.g. costs varied substantially in the experiments);
- The SP approach assumes that each respondent has perfect knowledge of all
alternatives and captures stated (not observed) choices;
- Not all alternatives were able to be included in the choice experiments, e.g. FFVs.
It was therefore necessary to calculate these constants from real-world data.
4.4 Calibration to historical data
The calibration of the model presupposes the availability of an up-to-date dataset with
the country-specific historical market shares. Given the aggregate nature of the available
real-world data, we adopted the following sequential approach:
- Step 1: PTTMAM’s database was updated with time series on car sales. For this
purpose, data from EAFO (2018), EEA (2018a), Eurostat (2017) and OICA (2017)
was collected. However, historical car sales market shares disaggregated by country,
size and powertrain were available for all the countries only until 2015. The
categorisation of car size was primarily made based on engine size and, for electric
cars, on segment (e.g. B for small cars or C for medium; see CEC (1999));
- Step 2: The values of the car attributes were simulated in PTTMAM to derive the
‘utility sum of attributes’ by country, size and powertrain (see Fig. A2), which
represents the systematic utility coefficient in Eq. 5;
- Step 3: The term Basc (‘ASC SP” in Fig. A2) was set equal to -50 if a particular powertrain
was not available in the market for a given size, otherwise equal to -zero;
- Step 4: The term Byscep; (INITIAL ASC RP’ in Fig. A2) was calibrated from the
collected market information;
- Step 5: In addition to the ASCs, the lambda scale term (A) would ideally be
calibrated to ensure that the models reflect real-world car type choices. For
simplicity, we assumed that this value is by default 1, i.e. that the scale of choices in
the real world derived from revealed preference (RP) data are equal to the scale of
the SC choices. Although it may also be possible to incorporate other attributes into
the RP utility equations (e.g. number of brands), which could provide an indication
of the supply side of the market and may improve the quality of the choice models,
high-quality market information on this was not available at the time this analysis
was conducted.
- Step 6: For the calculation of the car type probabilities, adjustments in the ASCs as
per Eq. 6 to ensure that the model replicates observed market shares;
eu in(eeserved shame (Eq. 6)
predicted share
- Step 7: The calibrated utility (‘V from RP’ in Fig. A2) is determined following Eq. 7:
Vapi = Le AVspi) + Bascrri (Eq. 7)
- Step 8: Finally, the nesting structure and @ parameter (recall section 3 and Eq. 2-4)
are used to simulate the market shares by country, powertrain and size (see Fig. A2).
To render information exchange between methods feasible, an Excel template was
created thereby reconciling the PTTMAM assumptions for each attribute and the DC
model output. Those assumptions are considered in section 4.6.
Finally, simulation errors were found for these three subscripted elements in the
variable ‘exp V from RP low emission nest’: in 2009 for [France,BEV,Large] and in
2011 for [Bulgaria, FFV,Medium] and [Bulgaria HEV,Medium]. This was caused by
very low registration values and solved by setting them to zero.
4.5 Transferability to the remaining powertrains and countries
As can be suspected from Fig. 1 and 2, the five powertrain options considered in the
discrete choice analysis needed to be re-mapped into the adapted version of PTTMAM.
We assumed that HEVs, PHEVs, BEVs, and FCEVs belong to the low emissions nest.
Conversely, the remaining powertrains were assumed to be outside of this nest (i.e. are
part of the ICEV nest).
Conceming the transferability of results to the remaining EU countries, the generic
operating cost coefficient was used for all the countries, except for France or Italy.
Since we had estimated lower price sensitivity to operating cost for these two countries,
we used their specific coefficients.
4.6 Numerical assumptions of powertrain attributes
Once the choice structure was updated, the future values of the attributes of each
powertrain (and size, as relevant) were required to run PTTMAM. From Fig. 3, electric
range and recharging time are expected to play an important role in BEV choice.
Though not shown in Fig. 3, average recharging time is also affected by the proportion
of normal power and high power (i.e. fast) recharging infrastructure availability. The
assumed dynamic behaviour of these variables is shown in the next three figures. In this
paper, the model time horizon considered extends until 2025.
Fig. 4 shows the simulated (sim) growth in BEV electric range, from ca. 160 km in 2012
to over 600 in 2025. As a reference, data based on the New European Driving Cycle
(NEDC) from three specific BEVs is shown: Renault Zoe (small), Nissan Leaf
(medium) and Tesla S (large). The assumed increase in range is due to higher battery
energy density over time and, especially, to a step increase in battery capacity in 2019.
10
BEV
s
Data (smal!)
. .
8 © Data (medium)
.
Data (large)
3 ‘Sim (small)
Sunt Sim (medium)
100 = sim (large)
Figure 4. Dynamic behaviour of BEV range, by size: data vs. simulation
Source: own work based on data from Wikipedia (2019) own simulations
Fig. 5 shows the evolution of recharging points in the EU, distinguishing between
normal and high power (or fast, with >22 kW following EU (2014)). The 2020 target
corresponds to the value determined in EU (2017a). For simplicity, no growth in
recharging infrastructure is assumed between 2020 and 2025 in this paper. At the
country level, the proportion of normal versus (vs.) rapid recharging infrastructure
varies, which influences country-specific average recharging times. For fast recharging,
avalue of 100 kW is assumed.
Sim (normal) =Total (data) @ Target (total)
150,000
100,000
50,000
publicly accessible re
oc
S$ oS aS 4 oe
ge ae af eh SY
eS
Figure 5. Deployment of normal vs. high power recharging points in the EU
Source: own work based on data from EU (2017a) and EAFO (2018) and own simulations
11
Fig. 6 shows the average simulated recharging time for medium-sized BEVs in five
major European car markets that were covered in the aforementioned survey. By
increasing the proportion of fast recharging, Italy achieves a noticeable reduction in
recharging time between 2013 and 2019. The assumed increase in battery capacity in
2019 adversely impacts average recharging times. As the proportion of normal vs. fast
recharging remains constant post-2020, no changes in recharging times are simulated in
the last five years of the model time horizon.
40
35
30 -—
25
20
2013 2015 2017 2019 2021 2023 2025
estimated average refielling or recharging Medium) : Cument
estimated average refielling or recharging timelGemany, BEV, Medium] : Curent
ethers ling ox meen ne BEV Mion] Cima
estimated average refielling or recharging time(Spain, BEV, Medium) : Curent
estimated average refuelling or recharging time{United Kingdom, BEV, Matfum]: Cumat
Figure 6. Dynamic behaviour of medium BEV recharging time in five countries
Source: own simulations using Vensim®
5. RESULTS
The results of executing the approach described in section 4 are reported for the largest
car market in the EU: Germany. Fig. 7 shows the historical observations vis-a-vis
simulated values of petrol and diesel car sales market shares. These powertrain options
clearly dominated the German market for new cars over the period. As can be seen, the
data could be replicated, via year-by-year adjustments of the ASCs, with the NMNL
framework embedded in PTTMAM.
12
‘Data (petrol) Sim (petrol) Data (diesel) = == Sim (diesel)
2008 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Figure 7. ICEV car sales market share [%] in Germany (2005-2016): data vs. simulation
Source: data from EA FO (2018) own simulations
In this market, alternative powertrain options exhibited very low sales market shares
during the calibration period considered. Because of the potential of electric cars to
replace ICEVs, annual sales of PHEVs and BEVs were calculated. The results for
Germany are shown in Fig. 8. As can be seen, the fit to data worsens, particularly in
2015 (the last year for which disaggregated historical data was considered in the
calibration). Although the simulation matches the data in 2017, it exhibits a more
sluggish behaviour than the 2018 value and current real-world policy developments
suggest.
Daa -== simulation
—
calibrated
choice
>
ww
Figure 8. Electric car annual sales in Germany (2005-2025): data vs. simulation
Source: data from EA FO (2018) own simulations
13
6. CONCLUSIONS AND OUTLOOK
We conclude that the linkage between DC and SD remains useful in this field of
application because the results of the former can be tested in the presence of feedback
loops while the latter benefits from an empirically grounded representation of choice.
The main contribution of this paper is the presentation of how the responses of an SP
survey designed with an SD model in mind may be incorporated into simulated
aggregate market shares in the EU car powertrain system.
A series of limitations related to this work can be pointed out. First, since PTTMAM
does not disaggregate the users market agent by demographic and socio-economic
characteristics, the presence of this information in the DC model could not be exploited
in the simulation part. Second, the modeling assumptions conceming the transferability
of the estimated utility coefficients into other powertrain alternatives and countries can
be challenged as e.g. the attributes of FFVs were not considered in the choice
experiments. This points to a third limitation, namely the need to devote greater
resources to ensure that: (i) a larger sample and more representative by including
respondents from other EU countries can be secured; (ii) the scope of the survey widens
by extending the duration of the survey, so that additional powertrain alternatives can be
inserted in the choice experiments; (iii) more sophisticated DC models such as cross-
nested (Hess et al., 2012), mixed logit and latent-class (Shen, 2009) are estimated and
their relative superiority tested; and (iv) the survey can be replicated in the future, so
that preference stability can be gauged, and be complemented with RP data.
The survey undertaken in 2017 was, by nature, static. Placing the resulting DC model in
a dynamic context raises intriguing questions: how can the aggregation problem be in
practice successfully addressed? Do ASCs become by necessity dynamic when framed
in a time-varying context? These need to be addressed in future research.
Further work along the following lines is required: (i) updating the database to a more
recent year and re-calibrating the model for that period; (ii) assessing the accuracy of
the new formulation by performing e.g. Theil’s inequality tests; (iii) analysing different
policy measures and constructing scenarios at the EU level with an extended model time
horizon; and (iv) scaling these choice assumptions into other users and vehicle types.
14
References
ACEA. (2018). Tax Guide. European Automobile Manufacturers Association (ACEA).
Retrieved from https://www.acea.be/industry-topics/tag/category/tax- guide
Al-Alawi, B. M., & Bradley, T. H. (2013). Review of hybrid, plug-in hybrid, and
electric vehicle market modeling Studies. Renewable and Sustainable Energy
Reviews, 21(Supplement C), 190-203. https://doi.org/10.1016/j.rser.2012.12.048
Ben-Akiva, M. E., & Lerman, S. R. (1985). Discrete Choice Analysis: Theory and
Application to Travel Demand. MIT _ Press. Retrieved from
https://books.google.it/books?id=oLC 6ZY Ps9UoC
CEC. (1999). Case No IV/M.1406 - Hyundai / KIA. Regulation (EEC) No 4064/89
Merger Procedure. Commission of the European Communities (CEC). Retrieved
from http://ec.europa.eu/competition/mergers/cases/decisions/m1406_en.pdf
EAFO. (2018). European Alternative Fuels Observatory (EAFO). Retrieved from
http://www.eafo.eu/
EEA. (2018a). Monitoring of CO2 emissions from passenger cars — Regulation
443/2009. European Environment Agency (EEA). Retrieved from
https://www.eea.europa.eu/data-and-maps/data/co2-cars-emission- 13
EEA. (2018b). Vehicle Emissions and Impacts of Taxes and Incentives in the Evolution
of Past Emissions. Eionet Report - ETC/ACM 2018/1 to European Environment
Agency (EEA). Retrieved from
http://acm.eionet.europa.eu/reports/docs/EIONET_Rep ETCACM_2018 1 Vehicl
e Taxes.pdf
EEA. (2019). Greenhouse gas emissions from transport. European Environment
Agency (EEA). Retrieved from _https://www.eea.europa.eu/data-and-
maps/indicators/transport-emissions- of- greenhouse- gases/transport-emissions-of-
greenhouse- gases- 10
EU. (2009). Regulation (EC) No 443/2009 of the European Parliament and of the
Council of 23 April 2009 setting emission performance standards for new
15
passenger cars as part of the Community’s integrated approach to reduce CO2
emissions from light-duty vehicles. Euro. Retrieved from _http://eur-
lex.europa.eu/legal-content/en/A LL/?uri=C ELEX :32009R0443
EU. (2014). Directive 2014/94/EU of the European Parliament and of the Council of 22
EU.
EU.
October 2014 on the deployment of alternative fuels infrastructure. European
Union Law. Retrieved from http://eur-lex.europa.eu/legal-
content/EN/TX T/?uri=celex%3A 32014L0094
(2017a). COMMISSION STAFF WORKING DOCUMENT Detailed Assessment
of the National Policy Frameworks Accompanying the document
COMMUNICATION FROM THE COMMISSION TO THE EUROPEAN
PARLIAMENT, THE COUNCIL, THE EUROPEAN ECONOMIC AND
SOCIAL COMMITTEE AND THE COMMITTEE OF THE REGIONS Towards
the broadest use of alternative fuels - an Action Plan for Altemative Fuels
Infrastructure under Article 10(6) of Directive 201/94/EU, including the
assessment of national policy frameworks under Article 10(2) of Directive
2014/94/EU. SWD/2017/0365 final. Retrieved from http://eur-lex.europa.eu/legal-
content/EN/TX T/?qid=1512384850915&uri=CELEX :52017SC0365
(2017b). Proposal fora REGULATION OF THE EUROPEAN PARLIAMENT
AND OF THE COUNCIL setting emission performance standards for new
passenger cars and for new light commercial vehicles as part of the Union’s
integrated approach to reduce CO2 emissions from light-dut. Retrieved from
http://eur-lex.europa.eu/legal-content/EN/TX T/?uri=C OM:2017:676:FIN
Eurostat. (2017). Your key to European statistics. European Commission (EC).
Retrieved from http://ec.europa.eu/eurostat/web/main/home
Ford, A. (1995). Simulating the controllability of feebates. System Dynamics Review,
11(1), 3-29. https://doi.org/10.1002/sdr.4260110103
Gnann, T. (2015). Market diffusion of plug-in elecric vehicles and their charging
infrastructure. Ph.D. Thesis. Fraunhofer V erlag.
Gomez Vilchez, J. (2019). The Impacts of Electric Cars on Oil Demand and
16
Greenhouse Gas Emissions in Key Markets. Ph.D. Thesis. Karlsruhe Institute of
Technology (KIT). KIT Scientific Publishing.
http://dx.doi.org/10.5445/K SP/1000091891
Gomez Vilchez, J., Harrison, G., Kelleher, L., Smyth, A., & Thiel, C. (2017).
Quantifying the factors influencing people’s car type choices in Europe: Results of
a stated preference survey. Retrieved from
http://publications.jrc.ec.europa.eu/repository/handle/111111111/50291
Hackbarth, A., & Madlener, R. (2013). Consumer preferences for alternative fuel
vehicles: A discrete choice analysis. Transportation Research Part D: Transport
and Environment, 25, 5-17. https://doi.org/10.1016/j.trd.2013.07.002
Harrison, G., & Thiel, C. (2017). Policy insights and modelling challenges: The case of
passenger car powertrain technology transition in the European Union. European
Transport Research Review, 9(3), 37. https://doi.org/10.1007/s12544-017-0252-x
Harrison, G., Thiel, C., & Jones, L. (2016). Powertrain Technology Transition Market
Agent Model (PTTMAM): An Introduction. Retrieved from
http://publications.jrc.ec.europa.eu/repository/handle/111111111/40434
Hess, S., Fowler, M., Adler, T., & Bahreinian, A. (2012). A joint model for vehicle type
and fuel type choice: evidence from a cross-nested logit study. Transportation,
39(3), 593-625. https://doi.org/10.1007/s11116-011-9366-5
Jensen, A. F., Cherchi, E., Mabit, S. L., & Ortuzar, J. de D. (2016). Predicting the
Potential Market for Electric Vehicles. Transportation Science, 51(2), 427-440.
https://doi.org/10.1287/trsc.2015.0659
Kieckhafer, K., Volling, T., & Spengler, T. S. (2014). A hybrid simulation approach for
estimating the market share evolution of electric vehicles. Transportation Science,
48(4), 651-670. https://doi.org/10.1287/trsc.2014.0526
Meyer, G. (2009). Analyse und technisch-dkonomische Bewertung von Gesetzesfolgen
im Individualverkehr: dargestellt am Beispiel der Automobilindustrie Japans und
Deutschlands. Retrieved from https://books.google.it/books?id=Q6BypwAACAAJ
17
OICA. (2017). Sales statistics: Sales of new passenger cars 2005-2016. Organisation
Internationale des Constructeurs d’Automobiles (OICA). Retrieved from
http://www.oica.net/category/sales-statistics/
Rohr, C., Lu, H., Smyth, A., Kelleher, L., Gomez Vilchez, J., Thiel, C., & Harrison, G.
(2019). Using Stated choice experiments to quantify the impact of vehicle
characteristics that influence European’s propensity to purchase electric vehicles.
In TRB Annual Meeting Online. Transportation Research Board (TRB) 98th
Annual Meeting, January 13-17, Washington D.C. Retrieved from
http://amonline.trb.org/68387-trb- 1.4353651/t0015-1.4368127/1694-
1.4368253/19-01325-1.4364644/19-01325-1.4368274?qr=1
Shen, J. (2009). Latent class model or mixed logit model? A comparison by transport
mode choice data. Applied Economics, 41(22), 2915-2924.
https://doi.org/10.1080/00036840801964633
Shepherd, S., Bonsall, P., & Harrison, G. (2012). Factors affecting future demand for
electric vehicles: A model based study. Transport Policy, 20, 62-74.
https://doi.org/10.1016/J. TRANPOL.2011.12.006
Struben, J., & Sterman, J. D. (2008). Transition Challenges for Altemative Fuel Vehicle
and Transportation Systems. Environment and Planning B: Planning and Design,
35(6), 1070-1097. https://doi.org/10.1068/b33022t
Thiel, C., Alemanno, A., Scarcella, G., Zubaryeva, A., & Pasaoglu, G. (2012). Attitude
of European car drivers towards electric vehicles: a survey. Retrieved from
http://publications.jrc.ec.europa.eu/repository/handle/111111111/26995
Wikipedia. (2019). The Free Encyclopedia. Retrieved from https://www.wikipedia.org/
18
Appendix
Figure A1 shows one of the choice scenarios respondents were presented.
Internal combustion engine Hybrid vehicles Zero emission
Fuel Type Petrol Diesel Plug - in Electric battery Hydrogen fuel cell
Purchase price (outright price) £15,000 £15,000 £40,000 £25,000 £15,000
Personal ContractPlan (monthly price for 36 month}+
£290 per month with a final
payment of £5,000
£290 per month with a final
payment of £5,000
£830 per month with a final
payment of £13,200
£510 per month with a final
payment of £8,250
£290 per month with a final
payment of £5,000
[0 perating cost (pence/ mile)
24p/ mile 22p/ mile
18p/ mile
6p/ mile 12p/ mile
Low Emission Vehicle incentive (daily charge £/ day)
Working or living in an urban | Working or living in an urban
area: £12.00 area: £12.00
Working or living in an urban
Working or living in an urban
area: £2.40
Working or living in an urban
area: £0.00
Other areas: £2.00 Other areas: £2.00
Other areas: £1.50
Other areas: £0.40 Other areas: £0.00
Vehicle value (after 3 years)
lime to recharge the battery to atleast half its capacity)
£3,750 £5,250 £10,000 £6,250 £5,250
Range on a full anki charge (miles) 400 miles 520 miles 400 miles 300 miles 400 miles
pafoal | Racharge Sine at senice maton (or eleckic vehicles, 5 mins 5 mins 5 mins, if Electric: 4 hours 4 hours 5 mins
* with a £1000 deposit
Figure A 1. Scenario in the second choice experiment
Source: Rohr et al. (2019)
Figure A2 shows an excerpt of the module where powertrain choice takes place in the updated version of PTTMAM.
cash or loan.
Proportion uiiity coefficient <average refuelling estimated average
HP proportion refueling time tme> ‘durin ties
sum of purchase <4 oe
opten proportions lly HP optio
uty coefficient _—>" uly refueling estimated average
HP optio
PCP proportion ope tine} melng or recharging
lity PCP opti
ulility coefficient ually PCP option
PEF option uilty coefficient
ay coefficient suis tility range me
lepreciation “Se Tene
<range>
<depreciation> uy Cepecaion nt enisions uilty coefficient
—— po ae emissions
uty coefficient pg
my cost ~ ‘ASC SP year powertrain
ee 2 oping Vikoree becomes available
— ia INITIAL ASC RP
<EU total annual operating cose t wee
fuel costs> ————e __perkm sum of marketshares fromRP sum of market shares
within ICE nest lambda scale tem ——_ within low emission nest
<anmual average yn nd a 4
VKT per vehicle>
pare ee exp V fromRP low market share within
~Sexp V from RP ‘emission nest. ~~ low emission nest
ICE nest
market | market share by gmbbw Scent x ee ee sam. elemates
on ars . sana Y <¢— emission nest> nesting coefficient a mnnest low emission nest
\ ICE nest
“Ss woh ICR wee IV ICE nest ra IV low emission
p ~~ N
nest
exp IV low a“
emission nest’
prob low
market share within exp IV ICE nest on
low emission nest>
Figure A2. NMNL model embedded within PTTMAM
Source: own work using Vensim®
20
Table Al shows a description of the eight attributes and their associated coefficient
terms used in the NMNL model.
Table Al. Attributes and coefficients in the systematic utility equation for
implementation of the model
Atiribute values in the utility equation &) Coefficients in utility @)
Attribute fhe sete descrip Units Coefficient terms Description
Purchase price | Purchase price in euros, itis | 1,000s of euros | purpr_sm, Coelficientis generic (the same)
expected that these will vary purpr_md, across all vehicle-type altematives,
by vehicle type, vehicle size purpr ig but varies by size of vehicle
and across EU countries
HP Proportion of vehicles HP proportions | AP_ot Coelficientis generic across all
purchased by HP multiplied | purchase price vehicle-type altematives, size of
by the price of vehicle (1,000s of vehicle and country
(assumed HP proportion euros)
likely to vary by country,
price varies as above)
PCP Proportion of vehicles PCPx purchase | PCP ct Coelficientis generic across all
purchased by PCP multiplied | price vehicle-type altematives, size of
by the price of vehicle vehicle and country
(assumed PCP proportion
likely to vary by country,
price varies as above).
Operating cost | Operating cost, in euros per | Cents/km oper_ct (all Coelficientis generic across all
Jom, assume that these will vehicles), vehicle-type altematives, but varies
vary by vehicle type, vehicle oper FR (France, | across countries for France and
size and country additive), oper IT | Italy
(Italy, additive)
Relained vehicle | Retained value of vehicle, in| 1,005 of euros | depr ct Coefficient is generic across all
value (i.e euros, assumed that these vehicle-type altematives, size of
depreciation) will vary by vehicle type, vehicle and country
vehicle size(?) and country
Range Range vehicle can travel, in| Im eff range, eff rLo | Separate values for ow emission
Jom, assumed that these will and other vehicles, but the same
vary by vehicle type and across countries and size of vehicle
vehicle size (2)
Refueling/ Time to refuel, these will | Mins refuel Coelficientis generic across all
re-charging time | vary by vehicle type (and vehicle-type altematives, size of
perhaps vehicle size) vehicle and country
Emissions Emission level for vehicle, | Categorical TeroEmiss, Coefficients are generic across all
will vary by vehicle type | variables LowEniss, vehicle-type altematives, size of
(and pethaps vehicle size) MedEmiss, vehicle and country
HighEmiss (set as
reference = 0)
Note: a ninth attribute (low emission car incentive) was included in the second experiment only.
Source: own work
21
