Working Paper Sustainability and Innovation

No. S 4/2014

Patrick Plötz
Till Gnann
Martin Wietschel

Modelling Market Diffusion of Electric

Vehicles with Real World Driving Data
Part I: Model Structure and Validation
 Abstract
The future market diffusion of electric vehicles (EVs) is of great importance for trans-
port related green house gas emissions and energy demand. But most studies on the
market diffusion of EVs focus on average driving patters and neglect the great variations
in daily driving of individuals present in real-world driving data. Yet these variations are
important for EVs since range limitations and the electric driving share of plug-in hy-
brids strongly impact the economic evaluation and consumer acceptance. Additionally,
studies often focus on private cars only and neglect that commercial buyers account
for relevant market shares in vehicle sales. Here, we propose a reliable, user specific
model for the market diffusion of EVs and evaluation of EV market diffusion policies
based on real world driving data. The data and model proposed include both private
and commercial users in Germany and allow the calculation of realistic electric driving
shares for all usage patterns. The proposed model explicitly includes user heterogene-
ity in driving behaviour, different user groups, psychological aspects and the effect of
charge-at-home options. Our results show that the proposed model reproduces group
specific market shares, gives confidence bands of market shares and reliably simulates
individual electric driving shares.
Contents
1 Introduction 1

2 Existing Models of Electric Vehicle Market Diffusion 1

3 Proposed Model for EV Market Diffusion 3

3.1 Motivation and Model Overview . . . . . . . . . . . . . . . . . 3
3.2 Data: Driving Profiles, Willingness-to-pay-more, and Techno-
economical Parameters . . . . . . . . . . . . . . . . . . . . . . 5
3.3 Formal Description of the Model . . . . . . . . . . . . . . . . . 8
3.3.1 EV simulation . . . . . . . . . . . . . . . . . . . . . . . 9
3.3.2 Total cost of Ownership and Utility Calculation . . . . . 10
3.3.3 Aggregation and Market Diffusion . . . . . . . . . . . . 13
3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

4 Results and Validation 15

4.1 Reproduction of Diesel Market Shares . . . . . . . . . . . . . . 15
4.2 Statistical Precision of Market Share Estimates . . . . . . . . . 16
4.3 Simulated Electric Driving Shares . . . . . . . . . . . . . . . . . 19
4.4 Future Availability of EVs in Germany . . . . . . . . . . . . . . 22
4.5 Significance of User Group Assignment . . . . . . . . . . . . . . 24

5 Summary 25
1 Modelling Market Diffusion of EVs with Real World Driving Data

1 Introduction
Global warming and the increase of greenhouse gas (GHG) emissions are some
of the most fundamental challenges of the 21st century [1, 2]. Especially the
transport sector with a growing number of vehicles worldwide has to make
a contribution to reduce GHG emissions radically. Furthermore, the scarcity
of conventional energy resources, in particular crude oil, requires new energy
carriers in the transport sector. Electric vehicles (EVs) such as battery (BEV),
plug-in hybrid (PHEV) or range extended EVs (REEV) are a means to this
problem since they are more energy efficient than conventional cars and produce
less GHG emissions when renewable energy is used [2].
There are several studies and models describing the introduction and mar-
ket diffusion of EVs and studying the effect of policies stimulating EV market
diffusion. Most of these studies focus on average driving patters [3, 4, 5, 6,
7, 8, 9, 10, 11, 12]. This could give incorrect results, since individual driving
patterns show great variations both between different users as well as between
different days for individual users [13, 14]. This is of particular relevance for
electric vehicles for which range limitations and electric driving shares have a
high impact on the economic evaluation and consumer acceptance. Addition-
ally, these studies often focus only on private cars and neglect that commercial
buyers have a relevant market share on selling figures of new vehicles. Their
purchase decision is also different, e.g. the cost-effectiveness is more important
for fleet applications than for private buyers. Furthermore, the average vehicle
kilometres travelled (VKT) of commercial car users are much higher than of
private car users.
The present paper attempts to elaborate and assess a reliable, user-specific
model for market diffusion of EVs and the evaluation of policies influencing
EV market diffusion. User-specific in the sense that all relevant buyer groups
are represented with their specific purchase behaviour and individual driving
patterns. The developed multi agent-based simulation model is applied to and
evaluated for Germany with a time horizon until 2020.
In the following section 2 an overview of existing models for EV market
diffusion is given, motivating the development of a new model. In section 3
the simulation model is described including a critical discussion of the model.
Selected results and different model validations are presented in section 4. In
the final section 5 a summary and conclusions are given. A further application
of the model with different market diffusion scenarios for Germany until 2020
will be developed in [15].

2 Existing Models of Electric Vehicle Market Diffusion

The diffusion of new technologies and EVs in particular has received consid-
erable attention in the literature (see [16] for a recent review of EV market
Modelling Market Diffusion of EVs with Real World Driving Data 2

diffusion models). A general classification of market diffusion models was given

by Geroski [17]. He describes two groups of models for market diffusion of
innovations: population and probit models. We will discuss them briefly and
classify existing market diffusion models accordingly in order to categorise the
model proposed here.
Population models describe users or adopters not as individuals, but as
groups. Population models are epidemic models which assume that the rate of
adoption is proportional to the number of adopters and the remaining population
that has not adopted a technology yet. This leads to the well-known logistic
differential equation and can be interpreted via the spread of information about
a technology [17]. We find population models for EV market diffusion or market
diffusion of other alternative fuel vehicles, in [18, 19, 5, 6, 7, 8, 9, 10, 11], which
range from simple mathematical equations to complex system dynamics models.
Population models offer a simple structure and interpretation. They are
usually applied by calibrating the market diffusion curve to existing market data
or by assuming hypothetical growth rates. This procedure is rather sensitive in
early market phases when little data is available. Furthermore, the heterogeneity
of the individual buying decisions and preferences of users, for example reflected
in the willingness to pay more for new technologies of some users, as well as
the individual economics of the driving behaviour cannot be incorporated into
these models.
The second group of market diffusion models, probit and agent-based mod-
els, studies adopters individually. These models are often applied when the
purchase decision is more complex or the technologies to be adopted are ex-
pensive. For example, a simple probit model for EV adoption would calculate
the average ownership cost difference between conventional and electric vehi-
cles and estimate an EV market share based on this difference. As fuel and
battery prices change over time, these cost differences change and with them
the estimated EV market share. Thus, probit models develop market diffu-
sion bottom-up and acknowledge that individual users can be very different.
Probit models were used to model EV market diffusion in [20, 21, 22, 23, 24]
where the detailed modelling approaches range from determining user shares by
stated preference experiments to agent-based models. Furthermore, there is a
third group of hybrid models which try to combine population and agent-based
modelling approaches [25, 12, 26, 27].
In probit models, actual user behaviour was studied in [21, 22, 23, 24] and
some models are based on actual driving behaviour [21, 22]. This would in
principle allow to analyse user behaviour in more detail. However, the latter
models use driving profiles of only one day which can cause severe inaccuracies
on the individual level as a single day might not show the individual’s typical
driving as is crucial for EVs due to their limited range (see [28] and section 4.3).
Consumer choice models are a third class of models to analyse the market
diffusion of electric vehicles [16]. However, these models face the problem
that consumer statements about their preferences for EVs are often inaccurate.
3 Modelling Market Diffusion of EVs with Real World Driving Data

Given the current market shares of EVs, the vast majority of users has never
experienced an EV and can hardly judge its utility. For example, in trying
to predict the utility of a new mode of transport by discrete choice methods,
reference [29, p. 177] concludes that the use-value of attributes of a new mode
cannot be estimated from the existing modes if the the former exhibits some
important attributes which none of the old ones has. This is certainly the case
for EVs and pinpoints the limitations of consumer choice models for EV market
diffusion when they are based on surveys with users of conventional vehicles.
In summary, agent-based models offer the possibility to include several as-
pects of great relevance for the market diffusion of EVs: individual purchase
preferences, individual driving behaviour (to account for the limited range of EVs
and the vehicle kilometres travelled (VKT) related usage costs), the need for fre-
quent recharging and infrastructure as well as the limited choice of EV brands
and models. In the model proposed here and coined ALADIN (Alternative
Automobiles Diffusion and Infrastructure) we explicitly take these factors into
account in an agent-based model with different user groups and their individual
decision making processes.

3 Proposed Model for EV Market Diffusion

3.1 Motivation and Model Overview
Buying decisions for passenger cars are complex. Many factors play a role, both
in private and commercial purchase decisions. Based on a survey of private
passenger car buyers [30], Figure 1 gives an overview of factors ranked first in
private users’ decision making processes. We identify vehicle size, price and
safety as the most important factors in the purchase decision. The importance
of the different vehicle attributes motivates to model the EV purchase decision
as maximisation of utility among several vehicle alternatives. For the future
market diffusion of EVs we focus on the users’ utility obtained from vehicle size,
price, brand, fuel consumption and fuel type, and to a certain extend engine
power, emissions and acceleration. Since our focus is on the vehicles propulsion
technology, we disregard safety, gear shift and four-wheel drive.
In the model proposed here, the potential utility of each technology is cal-
culated for each user individually. Furthermore, three user groups who differ
in their purchase decisions are distinguished: (1) private car buyers, (2) com-
mercial vehicles being used in commercial vehicle fleets only and (3) company
cars which are used by employees for both commercial and private purposes.
For Germany each group amounts for about one third of the annual passenger
car registrations [31]. We distinguish four vehicle sizes: small, medium, large
and light duty vehicles. To take the importance of vehicle size in the vehicles’
utility into account, we assume that every user will buy a vehicle of the same
size as his current vehicle. Purchase price and fuel consumption of a vehicle
are aggregated to the total cost of ownership (TCO). The fuel consumption
Modelling Market Diffusion of EVs with Real World Driving Data 4

Figure 1: Criteria ranked first in the decision making process for private pas-
senger car purchase differentiated by vehicle size. Figure is based on data from
[30].

costs strongly depend on the annual VKT and the individual driving pattern,
in particular the regularity of driving. For a reliable estimate, each user’s driv-
ing profile is simulated as a vehicle with each of the propulsion systems (BEV,
REEV, PHEV, diesel and gasoline) and the resulting fuel costs are calculated.
Fuel type, emission standards and acceleration are different for conventional
internal combustion engine vehicles and electric vehicles. Furthermore, many
consumers are willing to pay a price premium for a new technology [32] in
general and for EVs in particular [33, 34]. The positive factors of EVs such
as reduced noise, dynamic driving experience, their novelty and innovativeness
are integrated in the model proposed here as willingness-to-pay-more (WTPM)
of some users. Other factors are difficult to model and are assumed to be
comparable between conventional and electric vehicles, such as design, safety
and engine power.
Apart from the positive image of EVs as a new technology, EVs show certain
limitations. One important factor is the need of frequent recharging caused by
the limited electric range of EVs [35, 36]. To address this issue, we integrate
the cost for the primary charging option into the individual buying decision. In
addition to this, the choice of EVs in terms of brands and models as offered by
manufacturers is still limited and likely to remain so for the next years. This will
certainly restrain some users from buying an EV despite their potential benefits.
We include this effect of a limited choice of brands into our EV market diffusion
model by a two-step process: First, we assume users to stick to their current
vehicle brand if possible. Second, if an EV would maximise the user’s individual
utility but is not available from his current manufacturer, then a share of users
(depending on the number of brands offering EVs in that year) is assumed to
choose an EV from another manufacturer and the rest of the users are assumed
5 Modelling Market Diffusion of EVs with Real World Driving Data

to choose their second best vehicle option.

One of the most important aspects of our proposed EV market diffusion
model is the usage of real-world driving profiles. This is a major improvement
over existing models and has to our knowledge not been used comprehensively
in an EV market diffusion model so far. Here and in the following, a driving
profile comprises all trips of an individual vehicle over a fixed observation period
including starting time, duration and purpose of the trip. The distribution and
regularity of trip lengths varies strongly between different users and influences
the TCO and potential use of EVs significantly. We analysed driving profiles
of at least one week and found shorter periods to be too unreliable to draw
conclusions on the potential use of an individual vehicle as EV (cf. figure 5 below
and [37]). Based on the individual driving profile, each vehicle is simulated as
gasoline and diesel vehicle and as BEV, PHEV and REEV. The resulting electric
driving share as PHEV or REEV and the annual VKT are used to calculate the
individual TCO of each driving profile and vehicle option.
Based on the individual TCO and the additional positive and negative factors
integrated in the model as user specific utility, the utility maximising propulsion
technology for each driving profile is chosen. Thus, in each user group, a share
of driving profiles will correspond to EVs. This share is then extrapolated to the
annual registrations of vehicles in this user group. The model outputs are the
individual utility of each vehicle technology and the individual purchase decision
in a given year. The technological and economical parameters vary over time
and the decision process is repeated for each year. The annual registrations are
built up to a stock of EVs via a stock model.
To summarise, the model is structured as in figure 2. There are three main
model steps, (1) the EV simulation, (2) the utility calculation and (3) the stock
model. Within these steps there are certain parts where actual user behaviour
is integrated. We base the EV simulation on driving profiles in an infrastructure
scenario. Furthermore, the cost for infrastructure, the WTPM and the brand
loyalty of each individual user are incorporated into the utility calculation. While
the first two model steps are done individually for every vehicle driving profile,
the stock model aggregates the preceding results to a market diffusion. The
data used for modelling, in particular the driving profiles and the WTPM, will
be discussed in more detail in the following section. The different steps of the
model will be explained in more detail in section 3.3.

3.2 Data: Driving Profiles, Willingness-to-pay-more, and Techno-

economical Parameters
Driving profiles

The whole model is based on driving profiles which are analysed in the EV
simulation. Here and in the following, a driving profile is defined as all trips
of an individual vehicle including the departure and arrival time as well as the
Modelling Market Diffusion of EVs with Real World Driving Data 6

Figure 2: Overview of the proposed model for the market diffusion of electric
vehicles using real-world driving data. Based on individual driving data from pri-
vate, commercial and company cars (left panel) and using techno-economical
parameters (right panel), the market shares of different propulsion technolo-
gies are determined in three steps (central panel): (1) each driving profile is
simulated as EV and conventional vehicle; (2) based on the TCO, the cost for
home charging, the limited choice of EV makes and models and the individual
willingness-to-pay-more the utility maximising vehicle option is chosen for each
driving profile; (3) the vehicle choices are extrapolated to market shares and
aggregated to a vehicle stock.

distance travelled together with information about the purpose of the trip (to
work, to home, leisure, shopping, other) and additional information on the
vehicle (size, brand, age, annual VKT) and its owner. For private car owners and
company cars, we use the German Mobility Panel [38] which has already been
used for EV analyses (see e.g. [28, 39, 40]). For commercial users we use the
REM2030 Driving Profiles collected by the authors with GPS trackers [41]. Both
data sets are described in detail in [28, 42] and are publicly available. We use
these datasets as they cover observations of at least one week, which is crucial
for reliable estimates of a vehicle driving behaviour, in particular for EVs (see
[28, 43, 44] and section 4.3). The additional information for the private driving
profiles (including company cars) contains also socio-economic data about the
car owner (age, education, sex, income, city size of residence, typical over-night
parking spot, household size). Similarly, the commercial driving data contains
additional information about the company (no. of employees, city size of head
quarter location, total number of vehicles in fleet).
7 Modelling Market Diffusion of EVs with Real World Driving Data

EV user? EV interest? purchase intention? group label

yes - - innovators
no yes yes early adopters
no yes no majority
no no no laggards

Table 1: Definition of private adopter groups according to [49]. Participants in

survey answered the indicated questions and were considered members of the
four indicated adopter groups. A small number of respondents answered the
questions as no,no,yes and have been excluded from further analysis.

Willingness-to-pay-more

An important aspect of an EV’s utility are the positive non-monetary effects

of these vehicles. They are perceived as new and innovative, as silent and
environmentally friendly. These positive aspects of EVs are reflected in a WTPM
of some users and the amgnitude of the WTPM depends on the users position
in the adoption process [32, 45]. Of course, a stated willingness-to-pay is not
equal to the actual willingness-to-pay in a buying decision [46, 47]. However, the
stated WTPM gives an indication for the appreciation of a new technology and
an approximation of the actual WTPM. Using a WTPM is a common approach
in market diffusion models for electric vehicles [48, 20].
To assess a private user’s position in the adoption process of EVs and
their individual WTPM, we use large empirical data sets (see [33, 34, 49],
cf. [50, 51, 52]). Here the WTPM has been determined independently for
four adopter groups with a different attraction to electric vehicles: (1) users
of electric vehicles, identified as likely innovators, (2) attracted individuals with
purchase intention in the near future, identified as likely early adopters, (3) at-
tracted individuals without purchase intention, identified as likely early and late
majority, (4) uninterested individuals, identified as likely laggards (cf. table 1).
The four adopter groups were formed by the participants’ answers concerning
their current vehicle usage, the interest in EVs, and intention to buy an EV in
the near future (see [49, 53, 54] for details). Our aim is to combine these survey
results with the driving profiles and to assign each driving profile to one of the
four adopter groups with their WTPM.
Members of the four adopter groups differ significantly in socio-economic
variables like household income, employment status, household size, city size and
the willingness to accept a higher price for an electric vehicle [33, 34, 49, 53]).
As the data set also contains information about age, sex and education of the
user groups, we are able to assign each driving profile to one of the four groups
according to their resemblance with the other group members (see section 3.3.2
for details and 4.5 for a validation of this assignment). The participants stated
an individual WTPM for EVs. We will use the adopter group average WTPM to
include the positive aspects of EVs mentioned earlier. The percentage WTPM
Modelling Market Diffusion of EVs with Real World Driving Data 8

user group share of users willingness-to-pay-more

innovators 0.5% 30%
early adopters 1.5% 15%
majority 48% 10%
laggards 50% 1%

Table 2: Groups of private users and their willingness to pay more (WTPM).
The numerical values for the WTPM are median values of the group members’
answers.

is converted to absolute monetary values by using the conventional reference

vehicle in that vehicle size (Gasoline for small and medium-sized vehicle, Diesel
for large and light duty vehicle (LDV)). For the individual user, the positive
aspects are finally included in the utility calculation by subtracting the absolute
WTPM from the vehicle list price (LPi in eq. (3) below). The specific values
are summarised in table 2.
Although the described data set contains about 1,000 respondents, it is not
representative for the group sizes in Germany [33, 34], i.e. users of EVs and other
EV friendly groups are clearly overrepresented. This is useful for the validity of
the average WTPM in the groups. To correct the non-representative group sizes,
we use a second survey representative for private German car buyers [55, 56].
The groups are defined in the same way, i.e. according to EV ownership, interest
in EVs and purchase intention. Since the latter survey is representative, we use
it to determine the relative size of the adopter groups. The resulting share of
each adopter group is summarised in table 2.

Techno-economical Parameters
The different modelling steps require assumptions for techno-economical param-
eters concerning the vehicles (retail prices, specific fuel consumptions, battery
sizes and depth-of-discharges (DoD)) as well as the car market (annual sales
per segment (small, medium, large, LDV) and user group (private, commercial,
company car) and age-dependent scrapping probabilities) and framework con-
ditions (fuel, electricity and battery prices). Since the purpose of the present
paper is to introduce and discuss our model, the individual techno-economical
parameters are needed. However, for later use in section 4.3, we summarise
the assumed battery sizes, DoDs and energy consumptions of the vehicle types
considering here in table 3. The battery sizes and DoDs have been developed
in cooperation with the major German car manufacturers [57, 31] and the fuel
consumptions are based on [58].

3.3 Formal Description of the Model

Following the informal description of our model and the data used in the previous
sections, we will now discuss the formal modelling steps in more detail. Some of
9 Modelling Market Diffusion of EVs with Real World Driving Data

BEV REEV PHEV

Battery size C in kWh and Depth of Discharge (DoD)
C DoD C DoD C DoD
Small 20 90 % 13 80 % 7 75 %
Medium 24 90 % 16 80 % 10 75 %
Large 28 90 % 19 80 % 13 75 %
LDV 32 90 % 22 80 % 16 75 %
Consumption ce in kWh/100km and cc in litres/100km
ce cc ce cc ce cc
Small 19.1 - 19.1 6.6 17.8 5.6
Medium 23.3 - 23.3 8.2 22.0 7.0
Large 25.1 - 25.1 10.4 23.8 8.9
LDV 37.6 - 37.6 12.8 35.0 11.0

Table 3: Technical parameters for the battery simulation. Battery capacities

C in kWh, electricity consumption ce in kWh/100km and fuel consumption cc
in litres/100km. The sources and assumptions for these parameter values are
described and discussed in [57].

the model steps (the EV simulation and TCO calculation) have been published
already [28, 33, 34, 59, 60] but the other parts and the combination to an EV
market diffusion model are new. The first step of the model is simulate each
vehicle with its individual driving profile as EV. Based on this vehicle simulation,
the utility for each user of their vehicle with each propulsion technology is
calculated individually. Finally, the individual utility-maximizing decisions are
aggregated to vehicle sales in a stock model.

3.3.1 EV simulation
With the driving profiles described above we simulate the state of charge (SOC)
of a battery for a specific point in time t for each user as
{
SOC(t) − d∆t · ce d∆t > 0
SOC(t + ∆t) = for (1)
min{SOC(t) + ∆t · Ploct , C · DoD} d∆t = 0.

where the battery with capacity C and depth of discharge DoD is initially fully
charged SOC(0) = C · DoD. Here SOC(t) denotes the state of charge at time t.
The distance driven between t and t + ∆t is given by d∆t . The consumption of
electricity in kWh/km depends on the car size and is denoted as ce . Furthermore,
Ploct in kW describes the power for charging at the location where the car
was parked at t. If no charging infrastructure is available, Ploct = 0. Note
that different battery capacities are used in the simulation for different vehicle
sizes and EV types (cf. Table 3). If the car is driven (d∆t > 0), the battery
is discharged by the energy needed for driving the distance d∆t . Otherwise
(d∆t = 0), it is charged with the power Ploct for the time ∆t if necessary and
Modelling Market Diffusion of EVs with Real World Driving Data 10

charging infrastructure is available (Ploct > 0). In the simulation, no fixed step
width ∆t is used, but the result of each individual trip with its trip duration ∆t
or the parking duration ∆t are used.
We analyse each driving profile by simulating a battery profile by Eq. (1)
to estimate the technical EV potential. This technical EV potential delineates
whether a BEV would be able to cover the whole driving profile with a fixed
battery size or, in the case of a hybrid vehicle (REEV, PHEV), what electric
driving share si user i would achieve (cf. [28] for details and section 4.3 for
results). Furthermore, a simulated SOC < 0 for a BEV indicates that at least
one trip (or one chain of trips without charging infrastructure at intermediate
parking spots) of the simulated driving profile was longer than the electric driving
range of the BEV. The latter means that the particular driving profile cannot
be driven with a BEV. All technical parameters for the EV battery simulation
which will be used in section 4, are given in Table 3.
As mentioned before, EVs are charged when charging infrastructure is avail-
able at a parking position. The driving profiles contain information on the
purposes of the trips but not their exact location. Hence, we also have to de-
fine scenarios for charging infrastructure (see [28, 42] for details on charging
infrastructure scenarios). In the following, we assume: (1) PHEVs and REEVs
drive electrically until the energy in the battery is used up completely before
the conventional propulsion is used; (2) Private and company car owners charge
their vehicles with a power of 3.7 kW whenever they are at home; (3) Commer-
cial vehicles are only charged overnight with 3.7 kW no matter where they are
parked.
To summarise, the battery simulation is performed for each vehicle as each
EV option for each year. The results are the electric driving shares as PHEV
and REEV and the substitutability by BEV for each individual vehicle i.

3.3.2 Total cost of Ownership and Utility Calculation

In a second step the economic potential and utility is determined for each driving
profile. We calculate each user’s total cost of ownership (TCO) for different
propulsion systems. The annual total cost of ownership TCOa consists of capital
expenditure acapex and operating expenditure aopex

TCOa = acapex + aopex . (2)

We use the discounted cash-flow method with resale values and calculate the
investment annuity for user i as

LPi · (1 + p)T1 − SPi p (1 + p)T2

acapex
i =p + I CI · . (3)
(1 + p)T1 − 1 i
(1 + p)T2 − 1

The list price LPi (for vehicle and battery) is multiplied by the annuity factor
consisting of the interest rate p and the investment horizon T1 . SPi denotes
11 Modelling Market Diffusion of EVs with Real World Driving Data

the sale price of vehicle i for resale after T1 years and depends on the vehicle’s
annual vehicle kilometres travelled (VKT). The resale value is calculated for
each user i with his individual annual VKTi . We use results of [61] with SPi =
exp [α + 12 · β1 T1 + β2 VKTi /12] · LPβi 3 where the parameters α = 0.97948,
β1 = −1.437·10−2 , β2 = −1.17·10−4 and β3 = 0.91569 have been obtained by
regression (see [61] for details) and T1 denotes the vehicle’s age in years at the
time of resale.1 We use different values for T1 for private and commercial users
reflecting the different average vehicle holding times. The second term describes
the investment for the user-specific charging infrastructure ICIi multiplied by
the annuity factor without residual values and its specific investment horizon
T2 = 15 years.
The operating expenditure of user i for one of the propulsion technologies
is calculated as
( )
aopex
i = VKT i · s i c e k e + (1 − s i ) c c k c + k OM + ktax + kCIi . (4)

We multiply the vehicle kilometres travelled per year by user i (VKTi ) with
the cost for driving in electric mode plus the cost for driving in conventional
mode and the cost for operations and maintenance (kOM ). The cost for electric
driving consists of the electric driving share si , the electric consumption ce in
kWh/km and the cost for electricity ke in EUR/kWh. The same holds for the
conventional driving where the share of conventional driving (1−si ) is multiplied
by the conventional consumption cc in litres/km and the cost for conventional
fuel kc in EUR/litre. Finally the annual vehicle taxes ktax in EUR/yr and the
annual operating cost for charging infrastructure kCIi (in EUR/yr) are added.
By adding the infrastructure cost to the TCO calculation, we address the fact
that users must have at least one charging point to charge their vehicle regularly.
For private car users, we distinguish users that have a garage attached to their
homes or leave their car on the street overnight [62].
To assign each driving profile to one of the adopter groups with their WTPM
we used the following algorithm. For each driving profile, we first calculate the
agreement in socio-demographic characteristics with each survey respondent.
Matches were collected from seven variables: sex, age, employment status,
education, house hold size, household income and city size (all variables were
categorical). That is, a driving profile could achieve up to seven matches with
each of the survey respondents from a known adopter group. The number of
matches mijk ≤ 7 of user i with adopter group member j = 1,∑. . . , Lk (out of
the k = 1, . . . , 4 groups) were collected and normalised Mik = j mijk /(7Lk ).
1
Please note that the regression results for the EV resale values imply a higher absolute
resale price SP but lower relative or percentage resale value RV ≡ SP/LP when compared
to internal combustion engine vehicles (ICEs). If we assume the vehicle’s age and annual
VKT fixed at average values, the sales price is given as SP = c · LPβ3 with some constant
c. Thus the relative resale value will be given as RV ≡ SP/LP = c · LPβ3 −1 and accordingly
( )1−β3
RVEV LPICE
RVICE
= LPEV
< 1 since LPICE < LPEV and 0 < β3 < 1.
Modelling Market Diffusion of EVs with Real World Driving Data 12

The driving profile i should then be assigned to group k where the overlap was
the largest Mik > Mil ∀l ̸= k. However, since the relative group size should be
limited (the number of innovators is rather small), we took only the top 0.5%
(cf. table 2), i.e. those 0.5% with the largest overlap with the survey innovators,
as innovators. The other potential innovators were then assigned to their second
best matching adopter group. The same procedure was applied to the following
groups in descending order in the innovation process: innovators, early adopters,
majority and laggards (see [57, p. 182] for computational details). As a result of
this algorithm, each driving profile has been positioned in the adoption process
according to its socio-demographic variables with an associated WTPM. The
validity of this assignment is analysed in section 4.5 below.
To assess the WTPM of commercial vehicle fleets, we used the results from
a survey of approximately 500 German fleet managers [63]. About half the
fleet managers stated a WTPM with an average of 10%. Again, this WTPM
needs to be assigned to individual commercial vehicle driving profiles. We used
company size (measured as number of employees) as a proxy for the position
in the adoption process. Since larger companies seem more likely to engage
early in innovative technologies, commercial vehicles from companies with more
than 250 employees were assigned a WTPM of 10%. About 50% of the driving
profiles are from such a company in agreement with the results from [63]. No
reliable data was available for WTPM of company car buyers. We assume that
company car buyers have zero WTPM and use this in the model.
Since EVs are in an early market phase, the choice of models and brands is
and will remain limited for the next years. This fact slows down the market dif-
fusion of EVs since brand and design are vehicle purchase criteria (cf. Figure 1).
The limited choice of brands and models is included in the EV market diffusion
model proposed here. In a first step the present and near-future choice of EVs
were collected (from press announcements). Announcements for up to two years
in the future were available. Based on this data and the relevant number of
brands within each vehicle segment for normalisation, a logistic regression of the
upcoming brands was performed. The resulting logistic availability function has
been extrapolated into the future. This availability function is integrated into
the purchase decision as follows: If an EV is TCO optimal for a driver of brand
b and this brand has announced a vehicle for the year under consideration (or
earlier) the EV will be bought by that user. If the user’s brand b does not offer
an EV, then some of the users choose an EV from a different brand (according
to the logistic availability function) and the rest chooses the second best TCO
option.
Finally, we combine all factors to the utility of the different vehicle options.
We calculate the utility for each user i for each propulsion technology p and
assume that each user buys the option that maximises his or her individual
utility, i.e.
( )
max − TCOip + WTPMip − limited choiceip (5)
p
13 Modelling Market Diffusion of EVs with Real World Driving Data

where the users individual TCOip includes the cost for the home charging option
or base charging point for commercial vehicles, cf. Eq.(3). Calculating the utility
maximal propulsion system for each user, summing up all drivers for whom this
would be an electric vehicle and dividing it by the total number of driving
profiles, we obtain the shares pl of potential EV users in the sample.

3.3.3 Aggregation and Market Diffusion

The EV simulation and utility calculation above are performed for every driving
profile. We distinguish three different user groups (private, commercial fleet,
company car) and four vehicle sizes (small, medium, large and LDV) where LDV
are almost exclusively purchased by commercial fleets and accordingly neglected
for the other user groups. We thus arrive at 3 · 3 + 1 = 10 vehicle groups l.
The share of driving profiles in year t, pl (t), that are assumed to buy an EV
according to their individual utility is now multiplied with the number of vehicles
in the corresponding user group and vehicle size nl . As parameters change over
the years t we may calculate the EV registrations Nl (t) as Nl (t) = pl (t) · nl .
However, vehicles that were purchased in a given year do not remain in stock
forever. Instead vehicles will be scrapped with an age-dependent probability
P (t). This can also be written with a survival probability L(t) = 1 −
∫ scrapping
t ′ ′
0 scrapping (t )dt for a vehicle to survive until age t. With this distribution at
P
hand, one can write the stock of EVs in vehicle group l and year t, Sl (t), as the
sum of EVs purchased in earlier years Nl (t′ ) that survived until year t:
t
∑
Sl (t) = Nl (t′ )L(t − t′ ). (6)
t′ =t 0

The survival probability has been obtained from the official German statistics
(see [59] for details). A lifetime distribution for the vehicles to remain in stock
is needed for the stock model introduced above. We use data for the complete
German vehicle fleet and the age dependent scrapping probability over ten years.
These probabilities have been calculated considering the age structure of the
German vehicle stock since 2001 by computing the change between adjacent
ages in subsequent years for all years available. The Weibull distribution for the
β
survivor function is given by L(t) = e−(t/τ ) where the parameters τ = 14.7
for scale and β = 3.5 for shape have been obtained from a least square fit.
These imply an average age for scrapping of 13.8 years and an average age of
the vehicles in stock of 7.3 years, both in good agreement with other studies
of the German passenger car stock [57]. This distribution will be used for the
stock model of the German vehicle fleet.

3.4 Discussion
We now turn to a discussion of the EV market diffusion model described in
the previous section. The distinctive features of the present model are the
Modelling Market Diffusion of EVs with Real World Driving Data 14

individual utility maximisation based on a detailed analysis of many individual

driving profiles as well as in the inclusion of commercial vehicles and company
cars. The user specific analysis allows to cover a wide range of usage scenarios
and to study specific user groups such as commercial drivers or potential early
adopters.
The individual EV simulation is probably more abstract or mathematical
than the purchase decision of private users. But it covers the important aspect
of the regularity of an individual users’ driving behaviour. Users are aware of EVs
limited electric range and understand the general economics of low operating
costs for electric driving. Similarly, the TCO calculation of Eq. (2) is rather
complex but the purchase and operation costs of a vehicle are an important
aspect in the purchase decision both for private [30] and commercial buyers [63].
This is indicated by the average annual VKT for diesel vehicles (22,300 km)
and gasoline vehicles (11,800 km) in Germany [64] – reflecting the average
fuel economy under the German conditions of both propulsion technologies.
Accordingly, TCO calculations are a part of many EV market diffusion models
[65, 31, 59, 66, 48, 67]. Along the same direction, recent studies pointed out
that the costs of EVs are a major influence in the purchase decision [68, 33, 34,
52].
Although the TCO are an important factor in the vehicle buying decision,
they alone cannot explain purchase decisions of car users, neither for private nor
commercial car purchases. Furthermore, private buyers of hybrid and conven-
tional vehicles seem to lack knowledge necessary for a TCO-based decision [69].
An analysis of the potential early adopters of EVs in Germany shows that more
criteria than only the vehicle’s TCO are important [33, 34, 54]. Accordingly,
our model covers further important aspects of the purchase decision: (1) The
need of frequent recharging was addressed in the model by adding the cost for
a home charging option to the vehicle’s TCO; (2) the WTPM of some user
groups has been derived from surveys and is added to the driving profiles based
on the vehicle owner’s socio-demographic characteristics; (3) the limited choice
of brands and models is included according to the current share of brands of-
fering EVs. Overall, we attempt to make the most important factors in the
EV buying decision explicit and measurable. They have been included in our
model in an empirical way that allows updates or corrections when more data
on WTPM or choice of models become available in the future.
The model proposed here has some advantages. It is user specific since
individual driving profiles are used and the TCO and utility of EVs are calculated
for each individual driver. This allows us to study a wide range of usage patterns
and economical conditions as opposed to simple models based on average driving
behaviour. Thus, even niche markets can be analysed (as, e.g., in [70, 54]).
Furthermore, the large number of driving profiles allows the modeller to use
statistical methods to assess the statistical quality of the model results (cf.
section 4.2 and 4.3 for examples). Additionally, EV specific purchase decision
factors such as the limited electric range and the need for a base charging spot
15 Modelling Market Diffusion of EVs with Real World Driving Data

are addressed by the model proposed here.

One of the main disadvantages of the model are the data requirements.
For reliable estimates several hundred driving profiles per user group (private,
commercial, company car) are needed. These driving profiles should extend
in time over at least one week (see section 4.3 below) and contain additional
socio-demographic information on the car owner. Quite often, such data is
unavailable and the collection of data and the connection of the different data
sources requires real effort. However, driving data with limited observation
time is available for many industrialised countries and the interest in EVs has
triggered driving data collections over long time spans [43, 44].
The choice of model for EV market penetration depends on the specific
research question asked and the accuracy required. In the present case, the
model proposed here offers a detailed analysis of both private and commercial
user groups, covers the vast heterogeneity of vehicle usage patterns and takes
into account both the most important purchase decision criteria as well as the
peculiarities of EVs. This detailed picture requires noteworthy data input. Thus,
the model proposed here should give reliable and detailed results on the future
market penetration of EVs if sufficiently good data are available and correctly
combined.

4 Results and Validation

The present section is devoted to a validation and first results of the proposed
model using individual driving behaviour. The results will focus on the special
opportunities when using many driving profiles: an analysis of the statistical
significance of the predicted market shares and a statistical analysis of future
electric driving shares of PHEVs. Results of our model on EV market diffusion
and an assessment of of market diffusion policies are given in [15].

4.1 Reproduction of Diesel Market Shares

As a proof of principle, we first show that our TCO based approach for a vehicle
purchase is able to reproduce the current market shares of diesel passenger cars
in German commercial fleets. This supports our general proposal of including
TCO as one important factor in the purchase decision for passenger cars. We
acknowledge the fact that commercial fleets are only one of the three user
groups under consideration here. However, it is responsible for about one third
of the annual registrations of passenger cars in Germany and thus an important
market.
We study a large sample of German commercial passenger cars collected in
2002 [71]. For each vehicle in the database that has been used on the day of
the survey, the lengths of all daily trips are summed up and multiplied by the
average number of working days in Germany (which is 220 days per year) to
Modelling Market Diffusion of EVs with Real World Driving Data 16

obtain an estimate for the vehicles annual VKT. In this case, the latter was not
part of the survey and thus had to be calculated. For each vehicle, the TCO as
gasoline and diesel car was calculated and the vehicle has been assigned the fuel
type with lower TCO. For the validation purpose, we studied only medium-sized
vehicles and assumed a purchase price of 19,560 Euro for gasoline and 21,560
Euro for the diesel vehicle. The average fuel prices in 2002 have been taken
from [59]. The parameters for the 2002 passenger car assumptions in German
commercial fleets are summarised in table 4. The estimated share of diesel
vehicles in the different commercial branches are shown in Figure 3 together
with the actual market share as stated in the corresponding survey.

Parameter Unit Gasoline Diesel

Purchase price Euro 19,560 21,560
Fuel consumption l/100km 7,6 6
Fuel price Euro/l 1.34 1.26
Operation & Maintenance Euro/km 0.025 0.023
Insurance Euro/a 114 242

Table 4: Overview of techno-economical parameters for TCO-based estimate of

Diesel market shares in Germany’s 2002 commercial passenger car fleet. Prices
are without VAT.

Figure 3 shows that the estimated and actual market shares of commercial
diesel passenger cars in Germany are 40–60%. In Figure 3 the commercial
sectors are sorted by sample size which roughly follows the registrations of
passenger cars in these segments. In most cases, the estimated market shares
are very close to the actual market shares with significant deviations in the
sectors HJ (transport and telecommunications), A (agriculture and forestry),
and K (finance). Even if the share of diesel in a sector is well reproduced, one
could still question whether the individual vehicle assignments are correct. In
total, we found 54.2% of the individual assignments to be correct with a lowest
success rate of 38% in branch of industry K (finance) and the highest rate of
66% in branch B (mining). Thus we conclude that a TCO based model is in
principle able to reproduce the market shares of diesel passenger cars of German
commercial vehicles. TCO is thus one important aspect of the purchase decision
and accordingly part of many market diffusion models for EVs.

4.2 Statistical Precision of Market Share Estimates

The use of real world users with their individual driving profile allows estimates
of the statistical accuracy of the market share calculations. The market shares
depend, of course, on the assumptions for the decision making of the individual
as well as the assumptions for fuel and vehicle prices and vehicle configurations.
However, the model proposed here allows for a direct calculation of the uncer-
tainty from a finite sample size. Note that this is not possible when aggregated
17 Modelling Market Diffusion of EVs with Real World Driving Data

Figure 3: Diesel market shares within different branches of industry in Ger-

many. Shown are the actual values from a large-scale survey (solid line) and
the estimate from a simple TCO calculation (dashed line) together with con-
fidence bands (in blue - see section 4.2) from the finite sample sizes (given in
parentheses). See text for details of the calculation.

top-down market diffusion models, such as Bass or Gompertz [17], are used.
The effect of finite sample size can be expressed statistically as a confidence
interval [72]. That is, under the assumption of a representative sample, one
can compute intervals that have a high likelihood to contain the real market
share. This uncertainty stemming from finite sample size can then be passed on
to a result deriving from the market share, e.g. the stock of EVs or their total
energy consumption, by standard error propagation techniques. More precisely,
one estimates the sales share pl of vehicle type l (e.g. propulsion technology
or vehicle size or a combination of such distinctive characteristics) from the
number of driving profiles kl that fulfil the required condition (e.g. that should
be EVs) and the sub sample size nl as p̂l = kl /nl . Here, the hatˆindicates an
estimate for the ”real” market share pl . Given a confidence level 0 < α < 1, the
confidence band ”contains the real value of pl in (1 − α) · 100% of all cases in
which confidence intervals are estimated” [72]. For a given confidence level α
an upper value p+ − − +
l and a lower value pl are calculated, such that p̂l ∈ [pl , pl ]
in (1 − α) · 100% of the cases. In the present situation of market shares,
one has to calculate a confidence interval
( ) for the success probability pl of a
binomial distribution B(kl |pl , nl ) = nkll pkl (1 − p)nl −kl . The calculation of
confidence intervals via a Gaussian distribution is a common approximation for
this case. However, it is not reliable here since market shares of electric vehicles
tend to be rather small, i.e. kl ≪ nl , and the Gaussian approximation tends
to underestimate confidence intervals in that case [73]. Using a conservative
Modelling Market Diffusion of EVs with Real World Driving Data 18

approach, often referred to as ”exact”, the upper and lower confidence interval
boundaries are given by [73]

p−
l = Beta−1 (α/2, kl , nl − kl + 1) (7)
p+
l = Beta −1
(1 − α/2, kl + 1, nl − kl ).

Here, Beta−1 (x; a, b) denotes ∫ x the inverse of the cumulative Beta distribution
Beta(x; a, b) = (B(a, b)) −1 a−1 (1−t)b−1 dt with the Beta function B(x, y) =
∫ 1 x−1 0 t
0 t (1 − t)y−1 dt for normalisation.2 Note that p− l is not defined for kl = 0
and we set p− l = 0 in that case.
An example for the calculated confidence bands is shown in figure 3 for
the estimates of diesel market shares in German commercial sectors. Shown
are the α = 0.1, 1, 5, 10, 30 % confidence bands (from light blue to dark blue),
i.e. the ”true” value should lie within the confidence band in 99.9, 99, 95, 90,
70 % of the cases where confidence bands are estimated. As expected, the
width of the confidence bands increases with decreasing sample size (shown in
parentheses in the abscissa). In most cases the observed market share is within
or close to the range of the confidence bands. Thus, the TCO calculation
seems to capture important aspects of the purchase decision. Furthermore, the
calculated confidence bands help to distinguish purely statistical uncertainty
from possible systematic inaccuracies.
The statistical uncertainty can easily be propagated to derived quantities.
Let us assume the interval to be symmetric around pˆi and denote the half width
of the interval by ∆pl = (p+ −
l − pl )/2. The resulting uncertainty due to finite
sample size of the derived result is then calculated by error propagation. Let y =
f (x1 , . . . , xn ) denote a function of the input parameters xi being uncorrelated
and each having variance 2
∑ (∆xi ) . The variance of the resulting function is
then given by (∆f ) = i (∂f /∂xi )2 (∆xi )2 [74]. For example, let nl denote
2

the total number of vehicle registrations in segment or user ∑ group l. Thus,

2
the variance (∆N ) of the total number of registrations N = l p̂l (t)Nl (t) is
obtained as ∑
(∆N )2 = (∆pl )2 Nl2 . (8)
l
Similar results can be derived for the stock of EVs. Since ∆pl is a function of the
confidence level α we obtain upper N + ∆N and lower boundaries N − ∆N , i.e.
an α-dependent confidence interval, for the registrations of EVs. If N −∆N < 0,
we use zero as lower boundary for the number of registered vehicles.
To summarise, the usage of individual driving profiles and individual mod-
elling of the purchase decision as proposed here, allows a direct calculation of
the statistical uncertainty due to finite sample size. This helps to distinguish
systematic from statistical errors and is not feasible within aggregated market
diffusion models.
2
The inverse of the cumulative Beta distribution Beta−1 (x; a, b) is available in standard
software, e.g. as BETAINV(x, a, b) in MS EXCEL.
19 Modelling Market Diffusion of EVs with Real World Driving Data

4.3 Simulated Electric Driving Shares

The electric driving share of a potential hybrid EV is a key factor for its eco-
nomics. Regular daily driving with almost complete utilisation of the vehicle’s
electric driving range strongly reduces the fuel consumption costs. Thus, the
regularity of driving influences the user’s TCO in case of a plug-in hybrid EV. To
validate the model proposed here, we performed battery simulations as described
in section 3.3.1.

Electric Driving Shares for Different Annual VKT

To understand the variation in daily driving patterns and to estimate the range of
possible yet realistic electric driving shares a large set of driving profiles has been
simulated as PHEVs. The data set comprises privately owned vehicles as well
as commercially owned vehicles both for commercial use only (commercial cars)
and private and commercial use (company cars). Each of these groups makes
up about one third of the annual registrations in Germany [75]. The private and
company car driving profiles have been retrieved from publicly available driving
data from Germany [76] and consist of all trips during one week of more than
6,000 individuals (see below). The driving profiles of commercial vehicles have
been collected by the authors and consist of all trips of 354 vehicles during three
to four weeks and have been described in [41, 42]. All driving profiles have been
simulated as PHEVs. Here, we limit our analysis to medium-sized vehicles and
assumed a battery capacity of 10 kWh with a depth of discharge of 75 % and an
electricity consumption of 0.22 kWh/km, resulting in an electric driving range
of L = 34.1 km.
The simulated electric driving shares si (cf. Eq. (1) in section 3.3.1) are
shown as a function of annual VKT in figure 4 for the three user groups (private,
commercial, and company car). There are 6,339 private and company car driving
profiles in total. 3,727 (58.8%) of all private driving profiles contain the annual
VKT as stated by the owner. The other annual VKT have been extrapolated
from the observed driving behaviour, i.e. the observed VKT of one week have
been multiplied by 52 for private users if the annual VKT was not given. For
commercial fleet vehicles, the annual VKT of each vehicle have been obtained
by extrapolation from their individual observation period (with an average of
18.9 days).3
The electric driving shares in figure 4 vary strongly between different users
and with for different annual VKT. In addition to the simulated individual electric
driving shares a non-parametric kernel regression is shown to highlight typical
electric driving shares for varying annual VKT.√ Shown is a Nadaraya-Watson
kernel regression with Gaussian kernel over N nearest neighbours [72] in which
N is the number of driving profiles in a given user group (medium size vehicles
private N = 3561, commercial fleets N = 117, privately used company car
3
See [77] for a discussion of the accuracy of these estimates.
Modelling Market Diffusion of EVs with Real World Driving Data 20

N = 96). Please note that all driving profiles have been simulated as PHEVs
irrespective of their individual TCO. That is, only very few of the driving profiles
will actually be cost-effective as PHEVs and the first adopters of PHEVs are
unlikely to cover the whole range of annual VKTs [70, 54]. Furthermore, some
electric shares can be exceptionally high (> 365 · L/r where L denotes the
electric driving range and r the annual VKT) even for large annual VKT since
users can recharge up to several times a day if they return home between their
trips.

100% private cars

private car regression
Electric driving share PHEV

company car
80% company car regression
commercial cars
commercial car regression
analytical approximation
60%

40%

20%

0%
0 20.000 40.000 60.000 80.000
Annual VKT [km]

Figure 4: Simulated PHEV electric driving shares as a function of annual VKT

for medium-sized vehicles. Shown are the results of battery simulations of many
individual cars (symbols) and a kernel regression (solid lines): private (blue, N =
3561), company cars (green, N = 117) and commercial cars (red, N = 96). A
simple approximation for the kernel regression given by s(r) = 240·L/r (dashed
black line) is also shown.

The typical electric driving share is decreasing with increasing annual VKT as
expected. For given electric driving range L and annual VKT r, s(r) = 240·L/r
can serve as typical electric driving share s approximating the kernel regression
in figure 4 between 10,000 and 60,000 km annual VKT. That is, an average
user would have an electric driving share as if he was distributing his driving
equally among two out of three days a year (240 ≈ 2/3·365). However, even for
fixed annual VKT the electric driving shares within a user group vary strongly
indicating large differences in the regularity of the users’ driving patterns. This
demonstrates that average electric driving shares are strongly misleading (even
when an average is formulated as a function of annual VKT) and supports our
proposed market diffusion model with individual TCOs based on the individual
driving profiles.
21 Modelling Market Diffusion of EVs with Real World Driving Data

Statistical Precision of Electric Driving Share

Since the electric driving share is an important parameter in the individual TCO
calculation of PHEVs and REEVs, any estimate of it should be as precise as
possible and its precision should be known. In the present section, we estimate
the precision of electric driving shares when obtained from simulating multi-day
driving profiles as EVs. Directly related to this, a short observation period,
e.g. single-day driving profiles, can lead to strongly biased estimates of electric
driving shares and the future market shares of EVs [28, 37].
We consider a user’s simulated electric driving
∑T share as an average over
several days T of observation s¯i (T ) ≡ (1/T ) j=1 sij with the electric driving
share sij of user i on day j = 1, . . . , T .4 We take this as an estimate for the
”real” electric driving shares as obtained from the finite sample of T days. Here,
T denotes the number of observation days and not the number of driving days,
i.e. a vehicle could, e.g., drive on 5 out of T = 7 days. Assuming the individuals
electric driving share distribution to be Gaussian, the width ∆si (T ) around user
i’s average electric driving share s̄i (T )at confidence level α after T days is given
by [72]
σi (T )
∆si (T ) = t(1−α/2,T −1) √ . (9)
T
Here, σi (T ) denotes∑the standard deviation of user i after T days of observation
σi (T ) = [(T −1)−1 Tn=1 (s̄i (T )−sin )2 ]1/2 and t(x, n) is Student’s t-distribution
for n degrees of freedom [72].
As described in section 3.3, we simulated driving profiles of commercial
vehicles as PHEVs. These have an observation period of up to four weeks.
Following the prescription of Eq. (9) we calculated the confidence band width
∆si (T ) as a function of observation days T for each user’s electric driving share
(each user’s average electric driving share along with his annual VKT is shown in
figure 4). Figure 5 shows the empirical cumulative distribution function (CDF)
of the 95% confidence band widths ∆si for the driving profiles being observed
for different number of observation days T .
Figure 5 demonstrates how the distribution of confidence band widths is
changing with increasing duration of the observation time. Note that many
statistical surveys include only a single day of observation which is not able to
capture the day-to-day variation individual driving behaviour usually exhibits.
Furthermore, the Figure shows a clear change in the (distribution of) electric
driving share precision over observation time. The 25%- and 75%-quantile as
well as the median (50%-quantile) of the confidence width distributions for
different observation periods T are summarised in Table 5.
The precision of the estimated median electric driving share increases quickly
with growing number of observational days. However, the effect slows down
4
Above we used si as electric driving share of user i defined as the fraction of all electrically
driven kilometres and the total kilometres driven. This slightly differs this the running average
over time but the difference is negligible for large T .
Modelling Market Diffusion of EVs with Real World Driving Data 22

0.8

CDF(∆si )
0.6

0.4
T=2
T=5
0.2 T =10
T =15
T =20
0
0 0.1 0.2 0.3 0.4
Confidence band width ∆si

Figure 5: Empirical cumulative distribution function CDF(∆si ) of confidence

band widths ∆si (T ) after T = 2 (green), 5 (cyan), 10 (blue), 15 (black) and
20 (red) days.

period T quantile of ∆si

25 % 50 % 75 %
5 days 4.7 % 19.1 % 33.2 %
10 days 5.2 % 12.3 % 18.4 %
15 days 6.4 % 10.2 % 14.6 %
20 days 6.0 % 9.6 % 12.3 %

Table 5: Statistics of 95 %-confidence bands of simulated PHEV electric driving

shares.

after about 10 days and mainly driving profiles that contain trips on a small
number of days only improve noteworthy in the electric driving share precision
after this period of time. But as the observation time increases, the probability
of larger errors (as measured by the 75% quantile or the maximum of the CDF
in Figure 5) decreases. We conclude that driving profiles should contain at least
one week of driving.
The present section’s results highlight the importance of simulating each
individual driving profile instead of relying on assumed average electric driving
shares. Furthermore, driving profiles over several days of observation time allow
to estimate the statistical accuracy of the individual electric driving share and
thus of the accuracy of this very specific parameter in the TCO calculation.

4.4 Future Availability of EVs in Germany

The diffusion of innovations and new technologies typically follows an S-shaped
curve, well described by a logistic function [32, 17, 78, 79, 80]. We assume
that the availability of EVs from different brands can be described by a logistic
function, too. That is the share of brands per segment that offer an EV grows
23 Modelling Market Diffusion of EVs with Real World Driving Data

Segment BEV PHEV/REEV

t0 τ [10−3 ] t0 τ [10−3 ]
Small 2015.2 2.827 2019.2 1.292
Medium 2015.2 2.459 2014.8 1.292
Large 2024.6 3.091 2016.6 1.633
LDV 2018.3 3.153 2018.3 1.292

Table 6: Numerical parameters for the logistic function of the estimated future
EV availability as obtained from regression and partially transferred between the
segments.

logistically over time A(t) = [1 + e−(t−t0 )/τ ]−1 . Here t0 denotes the point in
time when 50% of the brands in a given segment offer an EV and τ is the time
scale of change of EV availability. Technically, we collected EV announcements
from different brands and calculated the cumulative number of brands per year
that already offer or have announced to offer an EV in the given year (see [57,
Ch. 7.4]). This cumulative number of brands has been divided by the number
of brands active in that segment for normalisation. For the case of Germany, we
chose all brands with non-zero new registrations in 2011 as active (26 brands
in the small segment, 32 in medium and 29 in large).
The parameters of the logistic function to estimate future availability of
EVs were obtained by least-squares regression and – if the amount of data is
insufficient for selected segments – assumed to be partly equal between the
groups. Furthermore, PHEVs and REEVs were treated as a single group since
from the perspective of availability these vehicles are rather similar and many
future announcements do not clearly distinguish between the two technologies.
Most announcements were available for medium-sized and large vehicles. The
parameters for the availability of these segment were obtained first. The τ value
of medium sized vehicles has been used for small vehicles and LDVs, too. For
small and large vehicles, t0 was derived from the announced availability in 2015,
i.e. for small vehicles t0 = τ ln[1/(1/26) − 1] + 2015 and t0 = τ ln[1/(1/29) −
1]+2016 for large vehicles. The results of the regression for future availability of
EVs from different brands in Germany are summarised in Table 6. The resulting
availability functions are shown in Figure 6. The data that has been used for the
least square regression is also shown in Figure 6. These are the share of brands
that have announced to offer EVs in the near future for medium BEV (red
squares) and medium PHEV/REEVs (blue diamonds). The agreement between
the regression curve and the actual share of announcements of all active brands
in the segment is good.
We observe from Figure 6 that EVs have been announced mainly for small
and medium-sized vehicles. Furthermore, BEVs have mainly been announced
for small vehicles and PHEV/REEVs in the medium and large vehicle segments.
The obtained availability curves estimate a coverage of about 50% of the brands
already for 2015 for mid-size PHEV/REEV and small BEV, somewhat later for
Modelling Market Diffusion of EVs with Real World Driving Data 24

1
S BEV
0.9 M BEV
L BEV
0.8 LDV BEV
S PHEV/REEV
Availability of EVs
0.7 M PHEV/REEV
L PHEV/REEV
0.6 LDV PHEV/REEV

0.5

0.4

0.3

0.2

0.1

0
2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
Year

Figure 6: Projected future availability of EVs from extarpolated least square

regression of model announcements from manufacturers active in the German
market. PHEV and REEV announcements have been treated as one group.
Also shown for medium BEV (red squares) and medium PHEV/REEVs (blue
diamonds) are the share of brands that have announced to offer EVs in the near
future, i.e. the data that has been used for the regression.

other segments. The expected increase in brand choice is slower only for large
BEVs, about 20% of the brands are expected to offer a BEV in this segment.

4.5 Significance of User Group Assignment

We distinguish four different user groups in different stages of the adoption
process: innovators, early adopters, early and late majority, and laggards [32] in
order to assess each driving profile’s individual WTPM in the proposed EV mar-
ket diffusion model. It is important to understand the validity of this assignment
and the present section aims at estimating the validity of this assignment.
In the original data set the assignment to one of the four adopter groups has
been made by the respondents’ answers to questions concerning their ownership
of an EV, their interest in EVs and their purchase intention for an EV. These
questions have not been asked in the driving profile data and the assignment
was made according to the driving profile’s socio-demographic variables (see
section 3.3). However, in the original data set where the adopter status is known,
the same assignment according to socio-demographic variables can be performed
and cross checked with the actual adopter status as determined from the EV
ownership, interest and purchase intention. Thus we applied the procedure
described in section 3.3 to the original survey data and obtained the respondents
status as determined from the similarity with other adopters. This new and the
original adopter status assignment are compared in table 7.
25 Modelling Market Diffusion of EVs with Real World Driving Data

original new assignment

assignment innov. ea. ad. majority laggards sum
innovators 0 1 47 33 81
early adopters 3 3 118 125 249
majority 2 3 193 162 360
laggards 0 2 97 152 251
sum 5 9 455 472 941

Table 7: Contingency table of original and newly assigened user group for val-
idation of the assignment. Note that the group sizes in the new assignment
have been pre-determined as described in section 3.3 and table 2.

In the new assignment, 0 out of 5 innovators have been identified correctly,

3 out of 9 early adopters, 193 out of 455 from the early (and late) majority
and 152 out of 472 laggards. In total, about 37% of the assignments are
correct. The assignment is accordingly slightly better than an assignment by
pure chance (with about 25% correct). The result is mainly determined by
the two large groups of users (early/late majority and laggards). Taking only
the two dominating groups 193+152 = 345 out of 360+251 = 611 users are
assigned to their correct group, i.e. 57% ± 5.2% (at 99% confidence level,
i.e. α = 0.01) are correct and also better than pure chance (50 %). Focussing
on these two large groups, table 7 can be reduced two the contingency table
(193, 162; 97, 152). Performing a χ2 -test of dependence, we find the old and
new assignment to be significantly dependent (p-value < 0.01) as wished, both
in the case of the reduced as well as the full contingency table.
To summarise, the assignment of a driving profile to a position in the adop-
tion process (and connected to this the WTPM) is significantly better than pure
chance but not very accurate. However, it seems sufficient for the modelling
purposes under consideration here since the driving profile is mainly determining
a user’s TCO and a strong correlation between the driving behaviour (in terms
of annual VKT and regularity of daily driving) and the users’ WTPM is not
very likely. We thus conclude that the assignment of a willingness to pay more
based on a user’s socio-economic characteristic is meaningful and helpful for
modelling the market diffusion of electric vehicles.

5 Summary
We proposed a new approach to model market diffusion of electric vehicles in
a reliable, user specific and empirical way. The model decision for different
propulsion system is based on an individual total cost of ownership calculation
extended by a willingness-to-pay-more for new and environmental friendly vehi-
cles of some vehicle buyers, limited choice of EV cars available on the market and
the cost for charging reflecting the current lack of public charging infrastructure
and the corresponding range anxiety.
Modelling Market Diffusion of EVs with Real World Driving Data 26

One main characteristic of the new model is the integration of differences

in user behaviour, in particular annual VKT and daily driving patterns. This
is important since the detailed analysis of empirical driving patterns has re-
vealed great variations in annual VKT as well as in daily driving distances. To
identify realistic electric driving shares of PHEVs and REEVs on the basis of
empirical driving patterns is of relevance for the TCO calculation. Furthermore,
it is needed to calculate the substitutability of conventional vehicles by BEVs.
Additionally, relevant differences between the use of private, company and com-
mercial cars can be found. In general company cars drive more on a yearly
basis and often show more regular daily driving as compared to private cars.
Accordingly, these user groups are differentiated in the present market diffusion
model. In addition, the limited selection of EV cars is one barrier for the market
penetration of EVs, despite the fact that this situation will improve in the next
years. To estimate the future EV model palette a forecast on S-shaped curves
has been developed.
The model quality has been tested by the forecast of market shares of diesel
vehicles for commercial passenger cars. Compared to statistical data the diesel
market shares calculated with the model show good agreement. The model
evaluation has furthermore shown that a short observation period of one or
two day could lead to strongly biased estimates of electric driving shares and
therefore biased results for EVs. The quality of results increases quickly with
the number of observation days. However, the effect slows down after about
ten days.
In summary, we developed and tested a user specific EV market diffusion
model to project future market shares of EVs and to evaluate policies stimulating
EV market diffusion.

Acknowledgements
The authors would like to thank Elisabeth Dütschke, Dirk Höhmann and André
Kühn. Funding from REM 2030 (regional ecological mobility) is gratefully ac-
knowledged.

References
[1] S. Kahn Ribeiro, S. Kobayashi, M. Beuthe, J. Gasca, D. Greene, D. S.
Lee, S. Muromachi, P. J. Newton, S. Plotkin, D. Sperling, R. Wit, P. J.
Zhou, Transport and its infrastructure, Tech. rep., Working Group III to
the Fourth Assessment Report of the Intergovernmental Panel on Climate
Change [B. Metz, O.R. Davidson, P.R. Bosch, R. Dave, L.A. Meyer (eds)],
Cambridge, United Kingdom and New York, NY, USA (2007).
27 Modelling Market Diffusion of EVs with Real World Driving Data

[2] International Energy Agency (IEA), Energy Technology Perspectives 2010,

Paris, 2010.
[3] J. Dagsvik, T. Wennemo, D. Wetterwald, R. Aaberge, Potential demand for
alternative fuel vehicles, Transportation Research Part B: Methodological
36 (4) (2002) 361–384.
[4] D. Santini, A. Vyas, Suggestions for a New Vehicle Choice Model Sim-
ulating Advanced vehicle Introduction Decisions (AVID): Structure and
Coefficients, 2005.
[5] D. Keles, M. Wietschel, D. Möst, O. Rentz, Market penetra-
tion of fuel cell vehicles analysis based on agent behaviour, In-
ternational Journal of Hydrogen Energy 33 (16) (2008) 4444–4455.
doi:10.1016/j.ijhydene.2008.04.061.
[6] J. Köhler, M. Wietschel, L. Whitmarsh, D. Keles, W. Schade, Infrastructure
investment for a transition to hydrogen automobiles, echnological forecast-
ing and social change 77 (8) (2010) 1237–1248.
[7] P. J. Lamberson, The Diffusion of Hybrid Electric Vehicles, Center for the
Study of Complex Systems, University of Michigan, Michigan, USA, 2008.
[8] P. E. Meyer, J. J. Winebrake, Modeling technology diffu-
sion of complementary goods: The case of hydrogen vehicles
and refueling infrastructure, Technovation 29 (2) (2009) 77–91.
doi:10.1016/j.technovation.2008.05.004.
[9] F. Nemry, F. Brons, Plug-in Hybrid and Battery Electric Vehicles - Market
penetration scenarios of electric drive vehicles, European Commission, Joint
Research Centre, Institute for Prospective Technological Studies, Sevilla
(Spain); Luxemburg, 2010.
[10] S. Shepherd, P. Bonsall, G. Harrison, Factors affecting future demand for
electric vehicles: A model based study, Transport Policy 20 (0) (2012)
62–74. doi:10.1016/j.tranpol.2011.12.006.
[11] J. Wansart, E. Schnieder, Modeling market development of electric vehi-
cles, in: 2010 4th Annual IEEE Systems Conference, 2010, pp. 371 –376.
doi:10.1109/SYSTEMS.2010.5482453.
[12] Y. Orbach, G. E. Fruchter, Forecasting sales and product evolution:
The case of the hybrid/electric car, Technological Forecasting and Social
Change 78 (7) (2011) 1210–1226. doi:10.1016/j.techfore.2011.03.018.
[13] S. Amjad, R. Rudramoorthy, S. Neelakrishnan, K. Sri Raja Varman, T. V.
Arjunan, Impact of real world driving pattern and all-electric range on
battery sizing and cost of plug-in hybrid electric two-wheeler, Journal of
Power Sources 196 (2011) 3371–3377.
Modelling Market Diffusion of EVs with Real World Driving Data 28

[14] R. Smith, S. Shahidinejad, D. Blair, E. D. Bibeau, Characterization of

urban commuter driving profiles to optimize battery size in light-duty plug-
in electric vehicles, Transportation Research Part D 16 (2011) 218–224.
[15] T. Gnann, P. Plötz, M. Wietschel, Modelling market diffusion of electric ve-
hicles with real world driving data – part ii: Results and policy assessment,
in preparation.
[16] B. M. Al-Alawi, T. H. Bradley, Review of hybrid, plug-in hy-
brid, and electric vehicle market modeling studies, Renewable
and Sustainable Energy Reviews 21 (0) (2013) 190 – 203.
doi:http://dx.doi.org/10.1016/j.rser.2012.12.048.
[17] P. Geroski, Models of technology diffusion, Research Policy 29 (45) (2000)
603–625. doi:10.1016/S0048-7333(99)00092-X.
[18] G. Duleep, H. Van Essen, B. Kampan, M. Grünig, Impacts of electric vehi-
cles - deliverable 2: Assessment of electric vehicle and battery technology,
Tech. Rep. 11.4058.04, ICF, CE Delft, Ecologic, Delft, Netherlands (Apr.
2011).
[19] J. Richter, D. Lindenberger, Potenziale der Elektromobilität bis 2050 -
eine szenarienbasierte Analyse der Wirtschaftlichkeit, Umweltauswirkungen
und Systemintegration, Endbericht, Energiewirtschaftliches Institut an der
Universität zu Köln (EWI), Köln (Jun. 2010).
[20] M. J. Eppstein, D. K. Grover, J. S. Marshall, D. M. Rizzo, An agent-
based model to study market penetration of plug-in hybrid electric vehicles,
Energy Policy 39 (6) (2011) 3789–3802. doi:10.1016/j.enpol.2011.04.007.
[21] F. Hacker, R. Harthan, P. Kasten, C. Loreck, W. Zimmer, Marktpoten-
ziale und CO2-Bilanz von Elektromobilität – Arbeitspakete 2 bis 5 des
Forschungsvorhabens OPTUM, Öko-Institut, Freiburg and Berlin, 2011.
[22] P. Mock, D. Hülsebusch, J. Ungethüm, S. Schmid, Electric vehicles - A
model based assessment of future market prospects and environmental
impacts, German Aerospace Centre, Stuttgart, 2009.
[23] B. Propfe, D. Kreyenberg, J. Wind, S. Schmid, Market Penetration Analy-
sis of Electric Vehicles in the German Passenger Car Market Towards 2030,
Stuttgart, Kirchheim/Teck-Nabern, 2012.
[24] T. Zhang, S. Gensler, R. Garcia, A study of the diffusion of alternative fuel
vehicles: An agent-based modeling approach, Journal of Product Innova-
tion Management 28 (2011) 152–168.
[25] H. Hao, H. Wang, R. Yi, Hybrid modeling of chinas vehicle owner-
ship and projection through 2050, Energy 36 (2) (2011) 1351–1361.
doi:10.1016/j.energy.2010.10.055.
29 Modelling Market Diffusion of EVs with Real World Driving Data

[26] L. A. de Santa-Eulalia, D. Neumann, J. Klasen, A simulation-based innova-

tion forecasting approach combining the bass diffusion model, the discrete
choice model and system dynamics - an application in the german market
for electric cars, in: Proceedings of Electric Vehicle Symposium 27 (EVS
27), Barcelona, Spain, 2013.
[27] J. Shin, J. Hong, G. Jeong, J. Lee, Impact of electric vehicles on existing car
usage: A mixed multiple discretecontinuous extreme value model approach,
Transportation Research Part D: Transport and Environment 17 (2) (2012)
138–144. doi:10.1016/j.trd.2011.10.004.
[28] T. Gnann, P. Plötz, F. Kley, Vehicle charging infrastructure demand for
the introduction of plug-in electric vehicles in germany and the us, in:
Proceedings of Electric Vehicle Symposium 26 (EVS 26), Los Angeles,
2012.
[29] K. Train, A validation test of a disagregate mode choice model, Trans-
portation Research 12 (1978) 167–174.
[30] A. Peters, P. de Haan, Der autokäufer - seine charakteristika und
präferenzen. ergebnisbericht im rahmen des projekts ”entscheidungsfak-
toren beim kauf treibstoffeffizienter neuwagen”., Tech. rep., ETH Zurich
(2006).
[31] Nationale Plattform Elektromobilität (NPE), Zweiter Bericht der Na-
tionalen Plattform Elektromobilität, Gemeinsame Geschäftsstelle Elektro-
mobilität der Bundesregierung, Berlin, 2011.
[32] E. M. Rogers, Diffusion of innovations, Free Press of Glencoe, 1962.
[33] A. Peters, E. Dütschke, How do consumers perceive electric vehicles? a
comparison of german consumer groups, Journal of Environmental Policy
& Planningdoi:10.1080/1523908X.2013.879037.
[34] M. Wietschel, E. Dütschke, S. Funke, A. Peters, P. Plötz, U. Schneider,
A. Roser, J. Globisch, Kaufpotenzial für elektrofahrzeuge bei sogenannten
”Early adoptern”, Endbericht, Fraunhofer ISI, IREES GmbH, Karlsruhe
(2012).
[35] E. Tate, M. O. Harpster, P. J. Savagian, The Electrification of the Auto-
mobile: From Conventional Hybrid, to Plug-in Hybrids, to Extended-Range
Electric Vehicles, SAE International, 2008.
[36] F. R. Kalhammer, B. M. Kopf, D. H. Swan, V. P. Roan, M. P. Walsh,
Status and prospects for zero emissions vehicle technology (Apr. 2007).
[37] T. Gnann, P. Plötz, F. Zischler, M. Wietschel, Elektromobilität im Person-
enwirtschaftsverkehr - eine Potenzialanalyse. Working Papers Sustainability
and Innovation No. S 7/2012, Fraunhofer ISI, Karlsruhe, 2012.
Modelling Market Diffusion of EVs with Real World Driving Data 30

[38] MOP, Mobilitätspanel Deutschland 1994-2010, Tech. rep., Projektbear-

beitung durch das Institut für Verkehrswesen der Universität Karlsruhe
(TH). (2010).
URL www.clearingstelle-verkehr.de

[39] A.-G. Paetz, T. Kaschub, M. Kopp, P. Jochem, W. Fichtner, Monetäre

Anreize zur Steuerung der Ladelast von Elektrofahrzeugen – eine mod-
ellgestützte Optimierung, Zeitschrift für Energiewirtschaft 37 (1) (2013)
1–12.

[40] A.-G. Paetz, T. Kaschub, P. Jochem, W. Fichtner, Load-shifting potentials

in households including electric mobility - A comparison of user behaviour
with modelling results, in: 2013 10th International Conference on the Eu-
ropean Energy Market (EEM 2013), pp. 1–7.

[41] Fraunhofer ISI, REM2030 driving profiles database version 2012 (2012).
URL www.rem2030.de/rem2030-de/REM-2030-Fahrprofile.php

[42] T. Gnann, M. Haag, P. Plötz, M. Wietschel, Market potential for electric

vehicles in the German commercial passenger transport sector, in: Electric
Vehicle Symposium 27 (EVS 27), Barcelona, Spain, 2013.

[43] S. Karlsson, L.-H. Kullingsjö, GPS measurement of swedish car movements

for assessment of possible electrification, in: Electric Vehicle Symposium
27 (EVS 27), Barcelona, Spain, 2013.

[44] R. Smith, S. Shahidinejad, D. Blair, E. Bibeau, Characterization of urban

commuter driving profiles to optimize battery size in light-duty plug-in elec-
tric vehicles, Transportation Research Part D: Transport and Environment
16 (3) (2011) 218 – 224. doi:http://dx.doi.org/10.1016/j.trd.2010.09.001.

[45] M. Laroche, J. Bergeron, G. Barbaro-Forleo, Targeting consumers who

are willing to pay more for environmentally friendly products, Journal of
Consumer Marketing 18 (6) (2001) 503–520.

[46] J.-C. Huang, T. C. Haab, J. C. Whitehead, Willingness to pay for quality

improvements: Should revealed and stated preference data be combined?,
Journal of Environmental Economics and Management 34 (3) (1997) 240
– 255. doi:http://dx.doi.org/10.1006/jeem.1997.1013.

[47] M. A. Bradley, A. J. Daly, Estimation of logit choice models using mixed

stated preference and revealed preference information, in: Les methodes
d’analyse des comportements de deplacements pour les annes 1990 - 6e
conference internationale sur les comportements de deplacements, chateau
bonne entente, Quebec, 22-24 May 1991, vol. 1, 1991.
31 Modelling Market Diffusion of EVs with Real World Driving Data

[48] P. Mock, Entwicklung eines Szenariomodells zur Simulation der

zukünftigen Marktanteile und CO2-Emissionen von Kraftfahrzeugen (VEC-
TOR21), Ph.D. thesis, Universität Stuttgart, Stuttgart (Oct. 2010).
[49] A. Peters, A. Agosti, M. Popp, B. Ryf, Electric mobility - a survey of
differnt consumer groups in germany with regard to adoption, Proceedings
to ECEEE summer study, Belambra Presque-ile de Giens, France (Jun.
2011).
[50] Hohe Kaufbereitschaft für Elektroautos. Studie der GfK Panel Services
Deutschland zu Einstellungen und Akzeptanz von Elektrofahrzeugen, GfK,
Landsberg a. Lech, 2012.
URL http://www.gfk.com/imperia/md/content/presse/
pressemeldungen2010/20100921_pm_elektroautos_dfin.pdf
[51] T. Heupel, B. Krol, T. Stender, FOM-Umfrage: Elektromobilität Kauf- und
Mobilitätsverhalten in Bezug auf Elektroautomobile, FOM Hochschule für
Ökonomie und Management, Essen, 2010.
[52] A. Knie, Die Neuerfindung urbaner Automobilität: Elektroautos und ihr
Gebrauch in den USA und Europa, Ed. Sigma, Berlin, 1999.
[53] E. Dütschke, U. Schneider, A. Peters, A.-G. Paetz, P. Jochem, Moving
towards more efficient car use – what can be learnt about consumer ac-
ceptance from analysing the cases of LPG and CNG?, in: Proceedings to
ECEEE Summer Study, Belambra Presqu’ile de Giens, France, 2011.
[54] P. Plötz, U. Schneider, J. Globisch, E. Dütschke, Who will buy electric
vehicles? Identifying the german early adopter, Transportation Research
A,Submitted for publication.
[55] J. Wesche, Nutzerakzeptanz von Stromladetarifen im Bereich der Elektro-
mobilität, Master’s thesis, Kassel University and Fraunhofer ISI (2013).
[56] E. Dütschke, A.-G. Paetz, J. Wesche, Integration Erneuerbarer En-
ergien durch Elektromobilität? Inwieweit sind Konsumenten bereit, einen
Beitrag zu leisten?, uwf UmweltWirtschaftsForum 21 (2013) 233–242.
doi:10.1007/s00550-013-0290-3.
[57] P. Plötz, T. Gnann, M. Wietschel, A. Kühn, Markthochlaufszenarien für
Elektrofahrzeuge – Langfassung, Fraunhofer ISI, Karlsruhe, 2013.
[58] H. Helms, J. Hanusch, Energieverbrauch von Elektrofahrzeugen, ifeu - In-
stitut für Energie- und Umweltforschung Heidelberg GmbH, Heidelberg,
2010.
[59] P. Plötz, T. Gnann, M. Wietschel, Total ownership cost projection for the
german electric vehicle market with implications for its future power and
electricity demand, in: Proccedings of the enerday conference, 2012.
Modelling Market Diffusion of EVs with Real World Driving Data 32

[60] F. Kley, Ladeinfrastrukturen für Elektrofahrzeuge - Analyse und Bewer-

tung einer Aufbaustrategie auf Basis des Fahrverhaltens, Fraunhofer Ver-
lag, Karlsruhe, 2011.

[61] V. Dexheimer, Hedonic methods of price measurement for used cars, Tech.
rep., German Statistical Office (2003).

[62] T. Gnann, P. Plötz, M. Haag, What is the future of public charging infras-
tructure for electric vehicles? - A techno-economic assessment of public
charging points for Germany, in: Proceedings of the 2013 ECEEE summer
study, Hyeres, France, 2013.

[63] Elektrofahrzeuge in deutschen Fuhrparks Zur künftigen Bedeutung von

Elektrofahrzeugen in deutschen Flotten (2011).

[64] R. Follmer, D. Gruschwitz, B. Jesske, Mobilität in Deutschland 2008

Ergebnisbericht, infas Institut für angewandte Sozialwissenschaft, Insti-
tut für Verkehrsforschung des Deutschen Zentrums fr Luft und Raumfahrt
e.V., Berlin, 2010.

[65] M. Holtermann, J. Radeke, J. Weinmann, J. Massiani, G. Hemmert,

A. Gohs, I. Witsch, Marktmodell Elektromoblität - Schlussbericht, Eu-
ropean School of Management and Technology (ESMT), Berlin, 2011.

[66] A. Peters, C. Doll, F. Kley, M. Möckel, P. Plötz, A. Sauer, W. Schade,

A. Thielmann, M. Wietschel, C. Zanker, Konzepte der Elektromobilität
und deren Bedeutung für Wirtschaft, Gesellschaft und Umwelt, Büro
für Technikfolgen-Abschätzung beim Deutschen Bundestag (TAB), Berlin,
2012.

[67] McKinsey & Company, A portfolio of power-trains for europe: a fact-based

analysis - the role of battery electric vehicles, plug-in hybrids and fuel cell
electric vehicles, Tech. rep. (2011).
URL http://www.zeroemissionvehicles.eu/

[68] K. Götz, G. Sunderer, B. Birzle-Harder, J. Deffner, Attraktivität und

Akzeptanz von Elektroautos: Ergebnisse aus dem Projekt OPTUM–
Optimierung der Umweltentlastungspotenziale von Elektrofahrzeugen
(2011).

[69] T. S. Turrentine, K. S. Kurani, Car buyers and fuel

economy?, Energy Policy 35 (2) (2007) 1213 – 1223.
doi:http://dx.doi.org/10.1016/j.enpol.2006.03.005.

[70] P. Plötz, T. Gnann, Who should buy electric vehicles? - the potential
early adopter from an economical perspective, in: Proceedings of the 2013
ECEEE summer study, Hyeres, France, 2013.
33 Modelling Market Diffusion of EVs with Real World Driving Data

[71] IVS, TU Braunschweig, Kraftfahrzeugverkehr in Deutschland 2002

(KiD2002), Tech. rep., IVS Institut für Verkehr und Stadtbauwesen, Tech-
nische Universität Braunschweig, Braunschweig (2002).

[72] L. Fahrmeir, R. Künstler, I. Pegeot, G. Tutz, Statistik – Der Weg zur

Datenanalyse, Springer, 2011.

[73] L. D. Brown, T. T. Cai, A. DasGupta, Interval estimation for a binomial

proportion, Statistical Science 16 (2001) 101–117.

[74] P. R. Bevington, Data Reduction and Error Analysis for the Physical Sci-
ences, New York: McGraw-Hill, 1969.

[75] Nationale Plattform Elektromobilität (NPE), Anhang des zweiten berichts

der nationalen plattform elektromobilität (May 2011).
URL http://www.elektromobilitaet.din.de/sixcms_upload/
media/3310/RZ_0905_NPE_Anhang_Final_Print.pdf

[76] Karlsruher Institut für Technologie - Institut für Verkehrswesen, tns

Infratest, Das Deutsche Mobilitätspanel - MOP-Berichte, Tech. rep.,
Bundesministerium für Verkehr, Bau und Stadtentwicklung.
URL http://mobilitaetspanel.ifv.uni-karlsruhe.de/de/
downloads/mop-berichte/index.html

[77] P. Plötz, How to estimate the probability of rare long-distance

trips, Fraunhofer ISI, 2014, working papers on sustainability and in-
novation No. S1/2014, http://www.isi.fraunhofer.de/isi-de/x/
publikationen/workingpapers_sustainability_innovation.php.

[78] A. van der Vooren, F. Alkemade, The diffusion of infrastructure dependent

technologies. A simple model, Utrecht University, Utrecht, 2010.

[79] J. Massiani, Modelling and Evaluation of the diffusion of electric vehicles:

existing models, results and proposal for a new model for policy in euro-
pean countries, European School of Management and Technology (ESMT),
2010.

[80] N. Meade, T. Islam, Modelling and forecasting the diffusion of innovation

- a 25-year review, International Journal of Forecasting 22 (3) (2006) 519–
545. doi:16/j.ijforecast.2006.01.005.
Authors’ affiliations

Patrick Plötz
Till Gnann
Martin Wietschel
Fraunhofer Institute for Systems and Innovation Research, Karlsruhe, Germany

Contact: Brigitte Kallfaß

Fraunhofer Institute for Systems

and Innovation Research (Fraunhofer ISI)
Breslauer Strasse 48
76139 Karlsruhe
Germany
Phone: +49 / 721 / 6809-150
Fax: +49 / 721 / 6809-272
E-Mail: [email protected]
www.isi.fraunhofer.de

Karlsruhe 2014