Cuadernos de Economía (2023) Volume 46, Issue 132, 59-68

Cuadernos de economía
www.cude.es

ARTÍCULO

Predicting the Diffusion of the Electric Vehicle: A Dynamic

Approach to Model the Impact of Imitation and Experience
Javier Bas1*, Elisabetta Cherchi2, Cinzia Cirillo3

1 Department of Economics, Universidad de Alcalá, 0.7 Facultad de Ciencias Económicas, Empresariales y Turismo
Alcalá de Henares, 28801 (Spain).
Email: [email protected]
Tel: +34 91 885 40 00
2 School of Engineering, ewcastle University, Cassie Building, NE1 7RU Newcastle upon Tyne (UK).
Email: [email protected]
Tel. +45 4525 6549
3 Department of Civil and Environmental Engineering, University of Maryland, 3250 Kim Bldg., College Park, MD 20742.
Email: [email protected]
Tel: +1 301-405-6864 Fax: +1 301-405-2585

*Corresponding Author Email: [email protected]

Abstract: Driven by environmental awareness and new regulations for fuel efficiency,
Jel Codes: O33, R49, electric vehicles (EVs) have significantly evolved in the last decade, though their
D91, M20. market share is still much lower than expected. Besides understanding the reasons
for this slow market penetration, it is crucial to have appropriate models to predict
the right diffusion of these innovative automobiles. Recent studies predicting the
Keywords:
evolution of the market for EVs combine substitution with diffusion models. In these
Electric Vehicle,
Diffusion, Forecast, models advanced discrete choice models are used to measure the substitution effect
Discrete Choice Models, among alternative vehicle’s engines, while Bass-type methods are used to account for
Survey Data. the diffusion effect of innovation. However, the most recent substitution/diffusion
models are not explicitly dynamic, nor measure the fact that innovation is
communicated through certain channels over time among members of a social system.
In this paper, we extend these substitution/diffusion models by including explicitly
the dynamic effect. This dimension makes the EV demand in a given period dependent
on the EV sold in a previous period. In this modeling scheme, we also account for and
measure, for the first time, some of the effects of social conformity on individuals’
choices. The model also includes the impact of policy incentives, in particular in the
availability of parking spaces and parking cost strategies. We illustrate our model for
the Danish EV market using data for the period 2013-2017. Results show an initial slow
penetration of the EV in the market, that progressively increases in the 2050 horizon.

Author Correspondence: [email protected]

https://doi.org/10.32826/cude.v46i132.1206
0210-0266/© 2023 asociación Cuadernos de economía. Todos los derechos reservados
Predicting the Diffusion of the Electric Vehicle: A Dynamic Approach to Model the Impact of Imitation and Experience
60

1. Introduction acquaintances influence our behavior. Even people who we

do not know personally may indirectly influence our
decisions. SC can occur because one wants to be accepted
The number of cities in Europe and the U.S. that have by the members of a certain group (Normative Conformity),
implemented traffic restrictions due to pollution is or because they want to act like it is supposedly right (then
considerable. In many cases, these policies are they consult members of their group to obtain information,
accompanied by other pro electric vehicle (EV) measures, which is called Informational Conformity) (Asch, 1961;
such as allowing drivers to park free of charge in regulated Crutchfield, 1955; Sherif, 1935). Therefore, conformity is a
areas, drive in High Occupancy Lanes, or access locations type of social influence involving a change in attitudes,
that other more polluting vehicles are not allowed to. beliefs, and behaviors to fit in a group, matching the group’s
These benefits, along with the prospect of savings —given norms and beliefs (Cialdini & Goldstein, 2004).
current fuel costs, present an economic incentive for users In summary, although these three aspects (model
to adopt the EV technology. The European market specification, dynamism, and social conformity) are
experienced a substantial surge with more than one million considered in some works, they have not always been
PHEV reached in 2018, experiencing 42% Year over Year properly addressed and, certainly, not considered jointly, as
growth (Pontes, 2019), with some countries nearly the literature review in the next section shows. Our study
doubling sales in the first half of 2018 (Denmark, Finland, aims to do so, by articulating in a single methodology these
Portugal, Netherlands, and Spain). Likewise, 360,800 plug- elements. This will fill the gap in the existing literature
in hybrid electric vehicles (PHEV) were sold in the U.S. in regarding new methodologies for predicting the diffusion of
2018 (Irle et al., 2019), representing 81% market growth. new technologies. Especially, the consideration of social
Backed by this growing attention on the part of users, the elements is an area not too much explored, at a general
interest of the automobile industry in promoting the EV has level, and particularly in the case of electric vehicle
increased, too. Producers are making significant diffusion. For this purpose, we use a combined substitution-
investments both to expand their catalogs and to improve diffusion model that explicitly includes SC. Our objective is
the models that they already offer. Finally, public to provide an accurate prediction that can serve as the basis
administrations, from local to national, have been on which stakeholders (industry, public agencies,
adjusting their regulations on EVs in recent years to consumers, and practitioners) can make informed decisions.
promote their acquisition and use. Most of these This paper is organized as follows: Section 2 describes the
modifications have to do with fees, public utilities, and efforts made in the field to combine DCM and diffusion
infrastructure; and their particularities are, overall, models, as well as Diffusion and social influence. Section 3
similar among the European countries and American states fully describes the data collection and the methodology;
(Lutsey, 2017; Tietge et al., 2016). while results and conclusions are presented in Sections 4 and
Altogether, these factors place the EV market in a unique 5, respectively.
and interesting situation, albeit fraught with uncertainty.
Thus, the need for reliable information on the future of EVs
is greater than ever, as solid predictions are the basis for 2. Literature Review
both industry strategic decisions and proper policymaking.
Hence, great efforts have been made in the last decade to 2.1. Discrete Choice and Diffusion Models
predict EV market penetration and the evolution of its sales.
Unfortunately, the forecasts published so far —in academia Several studies have been aimed to predict the evolution of
or other spheres— have differed substantially from the the EV market in the last decade, mainly through three
actual trends taking place thereafter, falling short in some approaches; i) Discrete Choice models (Ben-Akiva et al., 1999;
cases, and being too optimistic in others. Some of the Train, 2009), to provide a quantification of the willingness to
reasons we identify for this are methodological, while others purchase an EV; ii) Diffusion-type models (Bass, 1969), to
have to do with how the topic has been approached. For predict the evolution of sales over time and; iii) a combination
instance, Discrete Choice Models (DCM) have been a popular of these two to take advantage of the benefits of both types.
tool in some fields such as marketing and transportation for Regarding the DCM approach, two of the best attempts were
predicting demand. However, DCM rely on the responses those of Eppstein et al. (2011) and Kieckhäfer, Volling, and
provided in hypothetical scenarios, i.e. Stated Preferences Spengler (2014). The first used an agent-based choice model
(SP) data. Thus, when these models are used for prediction, where the purchasing decision of customers was affected by
they need to be calibrated to reflect the unobserved factors media coverage and social network activity, while the latter
present in real markets, which can be quite particular in the estimated a disaggregate demand model that was integrated
case of innovative products. A second main reason that may into a dynamic simulation to analyze the effect of the
explain the divergence between predictions and actual evolution of EV characteristics. However, neither of these two
market evolution is the neglect of the dynamism of the papers accounted for the diffusion effect. Thus, authors, in
demand. Although classic diffusion models such as Bass general, have been leaning towards other specifications in
(Bass, 1969) or Gompertz (Gompertz, 1862) have been which time and the evolution of sales play a major role.
applied to model the particular adoption behavior of new Ayyadi and Maaroufi (2018) Ayyadi and Maaroufi (2018) used a
technology —slow introduction until critical mass is reached, Generalized Bass diffusion model to evaluate the interaction
they have had limited success. Finally, diffusion is a process between accumulated sales of EVs and the battery price.
that occurs through social channels. In the words of Rogers Among its main results, they showed that the Moroccan
(Rogers, Someya, & Huang, 2010), “the diffusion process is market reached the maximum sales of EV after 14 years and
defined as that by which an innovation is communicated that the cost of the battery had a significant effect in
through certain channels over time among members of a accelerating the diffusion. Redondo and Cagigas (2015)
social system”. The behavioral assumption is that an performed a forecast of EV sales in Spain until the year 2040,
innovation is first adopted by a small segment of innovators, using a version of the Bass model that allowed to consider
and then later embraced by an increasing number of different product generations, as well as the jump
customers, the imitators, who are influenced by the number phenomenon (switch to a different generation). Similarly,
of adoptions that have already occurred. This conception Becker, Sidhu, and Tenderich (2009) adopted a Bass model to
corresponds to what in the psychological literature is forecast sales using as inputs: market size of the new
defined as Social Conformity (SC). Friends, family, and technology, a parameter that captured the percentage of
61
Javier Bas, Elisabetta Cherchi, Cinzia Cirillo
buyers whose purchase decision was not influenced by the more effective crowdfunding campaigns. Consumer behavior
purchasing behavior of others, and a metric that captured the is also to some extent related to so-called herd behavior, a
likelihood that additional consumers would adopt the phenomenon by which individuals act as part of a group,
technology in response to the buying experience of others. making decisions that they would not make as an individual.
Jiang and Jain (2012) proposed a Generalized Bass Model in This conduct has been reviewed by Biel et al. (2010) in the
which marketing mix variables were also incorporated, which
context of stock markets, where quick decisions are made,
offered better overall performance than other specifications,
sometimes led by what the majority does.
both in terms of fit and forecast performance. In a recent
study, Gnann et al. (2017) conducted a thorough analysis of Regarding the field of transportation, Pike and Lubell
40 research articles that developed diffusion models as well (2018), as well as Sherwin, Chatterjee, and Jain (2014),
as other approaches to the adoption of EV. They aimed to conducted research to test the hypothesis that certain
study the similarities among the models to offer factors condition social influence in the adoption of
recommendations for their implementation. Also, Massiani bicycles. They concluded that social influence played a
and Gohs (2015) studied the potential of Bass models to vital role in promoting cycling, although its degree of
evaluate the policies to promote the diffusion of the EV leverage depended on specific aspects of the trip such as
market in Germany. They questioned the varying values of the distance. Kormos, Gifford, and Brown (2015) evaluated the
parameters of these models found in the literature, impact of the opinion of others in a person’s choice of
highlighting the uncertainty faced by researchers when private vehicle use. Goetzke et al. (2015) studied the
developing further research upon them. Finally, Klasen and relationship between the choice of an EV made by an
Neumann (2011), also concluded that the Bass model is not individual and the choice made by others. Axsen and Kurani
easily parameterized when there is no available market data. (2012) used the personal network and the experiences of
However, the most recent and advanced studies combine individuals with a hybrid vehicle to study the effect of
substitution and diffusion models. For instance, Shepherd, interpersonal influence in the adoption of this technology.
Bonsall, and Harrison (2012) (based on Struben and TyreeHageman, Kurani, and Caperello (2014) described EV
Sterman (2008)) proposed a simulation system that buyers and their relationships with EV communities,
integrated disaggregate demand and system dynamic finding that early drivers used forums to gather
models, including the diffusion effect. However, the information about the EV characteristics. However, the
number of works that directly measure the effect of social
parameters were exogenously defined rather than
influence as an attribute in the stated preference
estimated. de Santa-Eulalia et al. (2011) in their research,
experiment is reduced, but two of them deserve special
presented a Bass model and a DCM with a dynamic attention. The first is Kuwano, Chikaraishi, and FUJIWARA
perspective to assess how consumer preferences and social (2014), who included the market share of EV in the choice
forces influence the introduction of EV on the market. The tasks. The second, Rasouli and Timmermans (2013),
model estimated both time and market share and was extended this methodology, dividing the market of EV into
flexible regarding the number of products and attributes, four reference groups. They also added an attribute
without the need for any market data. From their part, measuring a positive/negative overview of the EV.
Higgins et al. (2012) developed a diffusion specification Concerning how social influence in general, and SC, in
that incorporated multicriteria analysis and choice models, particular, affect Diffusion, there are remarkable
focusing on the geographical uptake of EV and the effect contributions. Pyo et al. (2023) focused on the role of the
of policy incentives. More recently, Jensen et al. (2017) social network in the diffusion process, paying particular
suggested a method that combined diffusion, as typically attention to the imitation component. They proposed an
estimated in the marketing literature, with advanced DCM. extension of the Bass model to make the social network a
All the parameters of the joint substitution/diffusion function of the number of customers who adopted the
model were estimated jointly, with the disaggregated product. Smaldino (2022) looked at the role of social
model estimated with SP being was adjusted to the real identity and how the diffusion of a product in society is
market endogenously in the diffusion process. However, affected by identity signaling. They also extended a Bass
this extension only included innovation through one single model, assuming that the probability of adoption is
term, whose effect on the probability of choosing EV varies influenced by three factors: a background ratio of
over time linearly. Moreover, imitation was left aside since spontaneous adoption, social influence from one’s group
it was dependent on the number of individuals that had members, and social influence from members of the
adopted the product already. outgroup. Morvinski, Amir, and Muller (2017) found that
2.2. Diffusion and Social Influence information about a large number of previous adopters
positively influences adoption only if those previous
Although SC, as described in the previous sections, has been adopters were described as similar to the potential ones.
extensively researched in psychology, (Ash & Woodward, Finally, Cherchi (2017) measured the effect of both
1989; Cialdini, 2007; Cialdini & Goldstein, 2004; Goldstein & Informational and Normative conformity in the preference
Cialdini, 2007; Schultz, Khazian, & Zaleski, 2008), for just a for EV. She found the social conformity effects highly
few examples), it has also been the object of study in other significant, and also that their impact on the overall utility
disciplines, including the two on which this work focuses: could be large enough to compensate for major differences
Economics/Business and Transportation. Regarding the in the characteristics of EV and Internal Combustion
former, efforts have focused mainly on unveiling the Vehicles (ICV).
consumer decision-making process. In this line, Baddeley
(2010) advocated for considering the incidence of social
influence and not exclusively the rational cost-benefit 3. Data collection and Methodology
approach when evaluating this process, since it is rooted in
psychological and social motivations. Cecere, Le Guel, and This paper builds on the work by Jensen et al. (2017), improving
Rochelandet (2017) studied what people consider when it by considering intrinsic innovation and imitation in a
making small investments in prosocial projects, to create diffusion-substitution model. The specification is also dynamic
Predicting the Diffusion of the Electric Vehicle: A Dynamic Approach to Model the Impact of Imitation and Experience
62

since the demand of EV in time t is dependent on the number decision about the type of car.
of EVs sold in time t-1. In doing so, we avoid the conceptual and 3. A Stated Choice Experiment, pivoted around the
values collected in the first section. It included
methodological pitfalls mentioned above. This procedure can
attributes related to the car characteristics and to the
be described in two steps: parking options, plus attributes that allow measuring
1. Estimation of a disaggregated DCM to account for the the effect of conformity.
substitution effect that occurs when individuals 4. The third section was dedicated to gathering
substitute another means of transportation, especially socioeconomic and residential information.
a car with a different type of engine, by an EV. This 5. Individuals’ attitude and perception towards several
formulation includes social influence and social aspects related to EV, injunctive social norms,
conformity so that the effect of the influence of affections, and values in life. Injunctive norms define
others’ behavior on that of the individual making the when the individual’s behavior is affected by what
choice is also considered. Information of previous other people think of them doing something. In this
periods is also included to gather the dynamic effect. case, the norm is measured by asking respondents
2. Projection of the actual data into the future and about the level of agreement to the following
estimation of a diffusion-substitution model in which statements:
the coefficients of the DCM are integrated, to a. People who are important to me (friends, family)
forecast EV sales. would approve of me using an electric vehicle instead
3.1. Data of my conventional car.
b. People who are important to me (friends, family)
The data used in this research comes from different think that using an EV instead of my conventional car
sources. The vehicle characteristics, as well as the Danish is appropriate.
EV monthly sales, have been computed by the Danish c. People who are important to me (friends, family)
Energy Agency (Statistics, 2016) until the year 2018. The think that more people should use an EV instead of
coefficients interacting with these characteristics in the their conventional car.
diffusion model proceed from Jensen et al. (2016). As for 6. Finally, information about personal and family income
the social conformity parameters, they are derived from a was asked.
survey performed in Denmark in the period between The sample was gathered from a list of individuals who had
December 2014 and January 2015 (Cherchi, 2017) and are signed up to participate in a real-life experiment in 2010
indicated in Table 1. in which they could use an EV for three months. 39% of the
Table 1: Social conformity parameters from (Cherchi, participants had already heard and been informed about
2017). EV. 73% of the users were males and the average age was
Coefficient Value 47. Regarding vehicle ownership, 52% of the users lived in
a household with one car, while 46% lived in a household
Number EV sold t-1 0.067
with more than one car. Additionally, 76% of the
Negative information on the need to change
-0.339 interviewees stated that the next purchase of a vehicle
activities
would replace an old one. Interestingly, the average daily
Negative information on range and activities -0.437
distance traveled was about 55km. A table illustrating
1. The survey was built specifically to study the effect of other characteristics of the complete sample is presented
parking policies on the choice of EV versus ICVs, as in Appendix A1.
well as the role played by social conformity on this Regarding the forecast exercise, it was necessary to make
choice. It consisted of five sections: assumptions about the evolution of the EV attributes. In
2. Detailed information about the last parking activity, this case, we designed a scenario in which the EV
vehicle ownership and use, most likely future vehicle characteristics experience gradual improvements thanks to
purchase, and whether a new EV car would be an technological progress. Table 2 provides these values,
additional vehicle in the household or if it would which we consider realistic since, as stated above, they are
replace an existing one. Users were also asked to based on reliable sources (a description of the variables is
indicate the degree of influence that they had in the provided in Appendix B).
Table 2: Forecasting Scenario 2018 - 2050.

Finally, the potential market (M in Equations (5) and (6) 3.2. Model Specification
below), is defined as 877,000, half of the car-owning
families in Denmark. More complex assumptions could have The model of reference follows Jensen et al. (2) which, in
been made, but they would have been difficult to validate, turn, is based on the diffusion model that accounts for
while this one seemed to us simple yet good enough. substitution effect developed by Jun and Park (1999).

1
It is worth mentioning that although the sample presents an not included in any model and, therefore, had no influence on
imbalance in some of the socioeconomic variables, those were the estimates.
63
Javier Bas, Elisabetta Cherchi, Cinzia Cirillo
Concretely, Jun & Park (1999) included the diffusion effect to change activities becomes positive and, therefore, Info
into the utility 𝑉𝑡𝑖,𝑘 of technical new technology k at time t. takes the value 0 onwards. The reason for considering this
𝑉𝑡𝑖,𝑘 = 𝑞 𝑖,𝑘 (𝑡 − 𝜏 𝑘 + 1) + 𝛽 𝑖,𝑘 𝑥 𝑖,𝑘 (1) variable in negative terms is due to the negativity bias
effect. This refers to the understanding that “negative
where 𝑥 (𝑖,𝑘) is a vector of the technology attributes, 𝛽 (𝑖,𝑘) information tends to influence evaluation stronger than
its corresponding coefficients, 𝑞 (𝑖,𝑘) the time-dependent comparably extreme positive information” (Ito et al.,
diff effect, and 𝜏 𝑘 is the period of the introduction of this 1998). Cherchi (2017) showed in her study that only
technology in the market. The superindex (𝑖, 𝑘) refers to the negative feedback is significant.
case of an individual owning a technology 𝑖 who switches to On the other hand, the preliminary data analysis showed a
𝑘. The probability of purchasing a product of generation k is peak in sales in December 2015. This was caused by the
𝑒𝑥𝑝(𝑉𝑡𝑖,𝑘 )
Danish government announcement that the registration tax
𝑃𝑡𝑖,𝑘 = 𝑖,𝑗 𝑖, 𝑗 ≤ 𝑘(2) for EV would be increased. Instead of considering this
𝑒𝑥𝑝(𝑐)+∑𝑗 𝑒𝑥𝑝(𝑉𝑡 )
information as an outlier, a dummy variable was defined to
Considering 𝑀𝑡 the potential market at time 𝑡, and 𝑌𝑡−1 the model the anticipation to this policy.
total number of units of a product at time 𝑡 − 1, the number Considering all these aspects, the utility function that is the
of sales in each period is: base of our model specification is:
𝑆𝑡𝑘 = (𝑀𝑡 − 𝑌𝑡−1 ) · 𝑃𝑡𝑘 𝑉𝑡𝐸𝑉 = 𝐴𝑆𝐶 𝐸𝑉 + 𝑞 𝐸𝑉 (𝑡 − 𝜏 𝐸𝑉 + 1) + 𝜆(𝛽̂ · ln 𝑁𝑡−1
𝐸𝑉𝑠𝑜𝑙𝑑
+ 𝛽̂ ·
= (𝑀𝑡 − 𝑌𝑡−1 ) ·
𝑒𝑥𝑝(𝑉𝑡𝑖,𝑘 )
(3) 𝐼𝑛𝑓𝑜𝑡−1 + 𝛽̂ 𝑋𝑡𝐸𝑉 ) + 𝛼 · 𝑇𝑎𝑥𝑡 (5)
𝑖,𝑗
𝑒𝑥𝑝(𝑐)+∑𝑗 𝑒𝑥𝑝(𝑉𝑡 )
The elements common to Jensen et al. (2017) maintain
From here, Jensen et al. (2016) defined their model as their meaning, while Info and Tax stand for the concepts
discussed above about information received and
𝑆𝑡𝐸𝑉 = (𝑀𝐸𝑉 − 𝑌𝑡𝐸𝑉 ) · 𝑃(𝐸𝑉𝑡 )
= (𝑀𝐸𝑉 − 𝑌𝑡−1 𝐸𝑉 )
·
anticipation to tax policy, respectively. λ is the
𝑒𝑥𝑝 (𝐴𝑆𝐶 𝐸𝑉 + 𝑞 𝐸𝑉 (𝑡 − 𝜏 𝐸𝑉 + 1) + 𝜆(𝛽̂𝐸𝑉 𝑥𝑇𝐸𝑉 ))
substitution parameter, which reflects the overall effect
· of vehicle attributes, information received, and number of
𝑒𝑥𝑝(𝜆(𝐴𝑆𝐶𝐼𝐶𝑉 + 𝛽̂ 𝐼𝐶𝑉 𝑥𝑇𝐼𝐶𝑉 )) + 𝑒𝑥𝑝 (𝐴𝑆𝐶𝐸𝑉 + 𝑞 𝐸𝑉 (𝑡 − 𝜏 𝐸𝑉 + 1) + 𝜆(𝛽̂ 𝐸𝑉 𝑥𝑇𝐸𝑉 )) EV sold. It is worth mentioning that this last element is
(4) considered in logarithms since the relation of this variable
with its lags is clearly not linear. Equation 5 leads to a
Where 𝛽̂ 𝐸𝑉 and 𝛽̂ 𝐼𝐶𝑉 were estimated using SP data and fixed
number of sales in each period equal to:
in the diffusion process. The three parameters estimated are
the EV Alternative Specific Constant, 𝑆𝑡𝐸𝑉 = (𝑀𝐸𝑉 − 𝑌𝑡𝐸𝑉 ) · 𝑃(𝐸𝑉𝑡 )
𝐴𝑆𝐶 𝐸𝑉 (𝑠𝑖𝑛𝑐𝑒 𝐴𝑆𝐶 𝐼𝐶𝑉 𝑖𝑠 𝑓𝑖𝑥𝑒𝑑), 𝑞, and 𝜆, which represented
𝑒𝑥𝑝(𝑉 𝐸𝑉 )
the alternative specific constant, the diffusion parameter, = (𝑀𝐸𝑉 − 𝑌𝑡𝐸𝑉 ) · 𝑒𝑥𝑝(𝑉 𝐼𝐶𝑉 )+𝑒𝑥𝑝(𝑉
𝑡
𝐸𝑉
)(6)
and a scaling coefficient, respectively. 𝑡 𝑡

That said, the research presented here brings several

improvements to the methodology. In order to consider the 4. Results
effect of social influence on the individual choices, two
elements are included; the number of EV sold in the This section provides the results of the estimation of the
previous period 𝑡 − 1, which makes the model dynamic; model described above as well as those obtained with the
and the type of information that the potential customer specification of Jensen et al. (2017), for comparison
receives about specific characteristics of the EV. This purposes. shows these results. Model 1 is the original
information that other users report on is related to parking specification from Jensen et al. (2016), estimated with the
spaces reserved to EV, EV range, and the need to change new data we count with. Model 2 is the model proposed in
activities due to low battery life. We compiled all three (3.6) and (3.7), but in which the variable Info (feedback
provided by others) has not been included. Finally, Model
into one dummy variable named Info, which equals 1 when
3 is the full model proposed in (3.6) and (3.7). In brief,
the feedback is negative. Moreover, we consider it equal
Model 1 does not include any social element, while Model
to 1 for all periods until the penetration of the charging 2 includes SC (EV sales), and Model 3 includes Social
infrastructure reaches 33% and the EV range also reaches Conformity and Social Influence (Info). This distinction is
33% of the ICV range. At that point, it is assumed that the made in order to quantify separately the impact of these
negative feedback about parking spaces, range, and need elements in the prediction.
Table 3: Estimation Results.
Model 1 Jensen et al. (2016) Model 2 Social Conformity Model 3 Social conformityand Social Influence
Value p-value Value p-value Value p-value
ASC EV −13.26 0.0001 −12.78 0.0001 −11.38 0.0001
q 0.94 0.03** 0.83 0.06* 0.58 0.05**
λ 0.05 0.81 0.14 0.52 0.12 0.59
Tax 2.64 0.0001*** 2.61 0.0001*** 2.59 0.0001***
R2 0.707 0.707 0.711
*significant at 10%, ** significant at 5%, ***significant at 1%
forecast of the EV spread. On the other hand, λ is not
For the full model, the values of both ASC and q are in line significant in any of the models, reflecting that
with the finds of Jensen et al. (2016), although in the substitution plays a minor role in choice. The variable that
lowest bound of the confidence interval in the case of the gathers the tax effect of December 2015 is highly
later. They obtained a range of (-17.99, -6.98) for the ASC, significant, revealing the consumer's rush to take
and (0.83, 2.57) for q. The diffusion parameter, q, is advantage of lower final prices. It is worth mentioning that
significant at the 95% level in models 1 and 3, and Model 1, being simpler than the others, yields the highest
significant at the 90% in model 2; which evidences that value of the parameter q; that is, a more intense effect on
considering the effect of diffusion allows a more realistic
Predicting the Diffusion of the Electric Vehicle: A Dynamic Approach to Model the Impact of Imitation and Experience
64
the diffusion due solely to the passage of time. A possible interesting is the forecast obtained for the period 2018 –
reason for that may be that the elimination of both the 2050, shown in Figure 1. It presents, for the three models,
social conformity and social influence variables causes the forecasted EV monthly sales, and the cumulative number
entire explanatory power to fall on this effect. More of EV sold expressed as a percentage of the total.

Figure 1: Actual, fitted, and forecast sales. Cumulative number and percentage of EV sold.
The classic S-shape that characterizes the introduction of shares from the initial phase to the sharp increase in sales that
a new technology can be appreciated, with low market occurs after reaching critical mass. This evolution is not due to
penetration in the early stages, and later progressive an improvement of the vehicle characteristics, but exclusively
increase once the product is more established in the to the inclusion of these social elements, as shown by the
market. In Model 2 and Model 3 the share of EV evolves comparison of the predictions of Model 1 with those of Models
from just 2% at the beginning of 2018 to around 40% by 2 and 3. Relying on the abundant literature on conformity
2050. On the contrary, Model 1 (which does not include (Baddeley, 2010; Cialdini, 2007; Cialdini & Goldstein, 2004;
social elements) reaches a more conservative final market Kormos, Gifford, & Brown, 2015) the reason is that individuals
share. These results conform the main finding of this work: change their behaviors because they want to fit in with a group,
explicitly considering elements of social conformity in the because they need help in making a decision, or because they
substitution model represents both a qualitative and a want to do what is supposed to be right. Therefore, when we
quantitative leap in predicting the development of the EV succeed in capturing these effects, we can achieve a better
market since models 2 and 3 prediction of EV sold by 2050 understanding of the process through which a technology
is about 66% higher than model 1 predictions (1,250/750 * diffuses in society. Concretely, we can now understand that the
100). It especially helps to predict the shift in market negative information about the EV that individuals may have
shares from the initial phase to the sharp increase in sales received until recently may have been, in part, behind the
that occurs after reaching critical mass. This evolution is slowdown experienced in the market. In our models, the
not due to an improvement of the vehicle characteristics, variable capturing the effect of informational conformity
but exclusively to the inclusion of these social elements, gathered relevant aspects regarding the flexibility in one´s
as our methodology reveals. As soon as the feedback about activity plans or schedule when using an EV. Although these
the use of the EV turns positive, a significant inclination aspects are clearly related to EV attributes (range, charging
towards this type of vehicle occurs. Relying on the time, etc.), they are in fact difficult to evaluate intuitively
abundant literature on social conformity (Baddeley, 2010; when someone is considering the purchase of such a vehicle and
Cialdini, 2007; Cialdini & Goldstein, 2004; Kormos, Gifford, is presented with a range figure. However, it is much easier to
& Brown, 2015), the rationale behind this phenomenon is value the opinion of a close relative or friend, who gives trusty
that individuals change their attitudes, beliefs, and feedback on these aspects.
behaviors because they want to fit in with a group, because Therefore, an effective policy to promote the use or
they need help in making a decision, or because they purchase of vehicles would involve allowing individuals to
simply want to do what is supposed to be right. participate in trial tests. Although it would be difficult for
a significant number of individuals to have access to these
5. Conclusions and Discussion tests, our results show that the participation of some of
them would be reflected in the feedback they give to
others, boosting the expansion of the EV if it were positive.
This paper proposes a methodology that aims to overcome On the other hand, in light of these results, other policies
the problems that are commonly incurred in forecasting EV
to promote the EV should consist of expanding and
sales. These shortcomings have to do with the use of
improving the charging infrastructure. A denser network
methodologies that are not the most convenient for that
task (DCM or Bass-type models, only), the lack of dynamism, that provides shorter charging times would be crucial in
and the neglection of social elements. We suggest a method the diffusion of EVs, since part of the weight of the spread
that combines a substitution model estimated with real of this technology still falls on the characteristics of the
disaggregated data, with a diffusion model based on a vehicle itself. Finally, although less related to the
realistic projection of the variables involved. We pay background of this work, public and private investment
particular attention to the role of Social Conformity, which dedicated to the improvement of vehicle characteristics
is considered in psychology (and increasingly in other fields) (such as the investment in research for new generations of
capital in the decision-making process of individuals. Thus, batteries to increase autonomy, or public subsidies to
our approach considers richer information than previous reduce the final purchasing price), could provide the final
work, and integrates the concepts of substitution and boost, in conjunction with the other aforementioned
diffusion into a single methodological paradigm. aspects, to the acquisition of this technology. In any case,
Our results lead to several conclusions. Explicitly considering it is necessary to emphasize the importance of other
elements of social conformity in the substitution model aspects of social conformity not addressed in this study
represents a qualitative leap in predicting the development of such as social-signaling —for which we refer the reader to
the EV market. It especially helps to predict the shift in market Cherchi (2017), which any policy, to be truly effective,
65
Javier Bas, Elisabetta Cherchi, Cinzia Cirillo
must combine. In this line, it is worth mentioning that, if Bass, F. M. (1969). A new product growth for model consumer
we take for good the hypothesis of technology diffusion durables. Management science, 15(5), 215-227. doi:
through social channels, such diffusion will of course be https://doi.org/10.1287/mnsc.15.5.215
dependent on the scenarios designed to make the Becker, T. A., Sidhu, I., & Tenderich, B. (2009). Electric
predictions. This is the reason why we relied on reliable vehicles in the United States: a new model with
sources to elaborate a design that would reflect a realistic forecasts to 2030. Center for Entrepreneurship and
evolution of all elements involved in the model estimation Technology, University of California, Berkeley, 24,
and forecast exercise. 1-32. Retrieved from https://www.electric-
In any case, we are confident that our methodology can, vehicles.info/library/rapport/rapport051.pdf
on the one hand, improve the diffusion models used in the Ben-Akiva, M., McFadden, D., Gärling, T., Gopinath, D.,
industry. On the other hand, we also hope that they can be Walker, J., Bolduc, D., et al. (1999). Extended
used as an evaluation tool for public agents, who may make framework for modeling choice behavior. Marketing
use of a tool that provide more reliable predictions on EV letters, 10, 187-203. doi:
technology deployment. A tool on which they can leverage https://doi.org/10.1023/A:1008046730291
to carry out more accurate and successful policies. Biel, A., Andersson, M., Hedesström, M., Jansson, M., Sundblad,
Finally, there are some improvements that could enhance E.-L., & Gärling, T. (2010). Social Influence in
this study, although the one that could have the greatest Stockmarkets: A Conceptual Analysis of Social Influence
impact would be a better collection of the social Processes in Stock Markets. Sustainable Investment
information. That is, to know in greater detail the Research Platform. Retrieved from
composition of the social network of each individual in the https://www.researchgate.net/publication/46466196
sample, as well as to improve the Stated Choice Cecere, G., Le Guel, F., & Rochelandet, F. (2017).
Experiment so that it could capture even better the Crowdfunding and social influence: an empirical
elements of Normative and Informational Conformity. This, investigation. Applied economics, 49(57), 5802-5813.
in fact, is a work that is already being developed by the doi: https://doi.org/10.1080/00036846.2017.1343450
authors of this article (Bas, Cirillo, & Cherchi, 2021). Cherchi, E. (2017). A stated choice experiment to measure the
effect of informational and normative conformity in the
Acknowledgments preference for electric vehicles. Transportation
Research Part A: Policy and Practice, 100, 88-104. doi:
https://doi.org/10.1016/j.tra.2017.04.009
The authors would like to thank Anders F. Jensen for his Cialdini, R. B. (2007). Descriptive social norms as
contribution in updating the data on vehicle characteristics underappreciated sources of social control.
used in this paper. Psychometrika, 72(2), 263-268. doi:
Authors Contributions https://doi.org/10.1007/s11336-006-1560-6
Cialdini, R. B., & Goldstein, N. J. (2004). Social influence:
Javier Bas: Conceptualization, Data curation, Formal Compliance and conformity. Annual Review of
analysis, Investigation, Methodology, Writing. Elisabetta Psychology, 55, 591-621. doi:
Cherchi: Conceptualization, Investigation, Methodology, https://doi.org/10.1146/annurev.psych.55.090902.
Review & editing, Supervision. Cinzia Cirillo: Funding 142015
acquisition, Review & editing, Supervision. Crutchfield, R. S. (1955). Conformity and character.
American psychologist, 10(5), 191-198. doi:
References https://doi.org/10.1037/h0040237
de Santa-Eulalia, L. A., D’Amours, S., Frayret, J.-M.,
Asch, S. E. (1961). Issues in the Study of Social Influences Menegusso, C. C., & Azevedo, R. C. (2011).
on Judgment. Conformity and deviation, 143–158. Advanced Supply Chain Planning Systems (APS)
doi: https://doi.org/10.1037/11122-005 Today and Tomorrow. In O. Dilek (Ed.), Supply Chain
Ash, M. G., & Woodward, W. R. (1989). Psychology in twentieth- Management (pp. 171-200). IntechOpen. doi:
century thought and society. CUP Archive. Retrieved from
https://doi.org/10.5772/19098
https://api.semanticscholar.org/CorpusID:140830678
Eppstein, M. J., Grover, D. K., Marshall, J. S., & Rizzo, D.
Axsen, J., & Kurani, K. S. (2012). Social influence, consumer
M. (2011). An agent-based model to study market
behavior, and low-carbon energy transitions. Annual
Review of Environment and Resources, 37, 311-340. doi: penetration of plug-in hybrid electric vehicles.
https://doi.org/10.1146/annurev-environ-062111- Energy Policy, 39(6), 3789-3802. doi:
145049 https://doi.org/10.1016/j.enpol.2011.04.007
Ayyadi, S., & Maaroufi, M. (2018). Diffusion models for Gnann, T., Plötz, P., Kühn, A., & Wietschel, M. (2017). How
predicting electric vehicles market in Morocco. 2018 to decarbonise heavy road transport. Accessed
International Conference and Exposition on November, 16, 901-909. Retrieved from
Electrical And Power Engineering, 0046-0051. doi: https://www.isi.fraunhofer.de/content/dam/isi/d
https://doi.org/10.1109/ICEPE.2018.8559858 okumente/cce/2017/4-346-17_Gnann.pdf
Baddeley, M. (2010). Herding, social influence and economic Goetzke, F., Gerike, R., Páez, A., & Dugundji, E. (2015). Social
decision-making: socio-psychological and interactions in transportation: Analyzing groups and
neuroscientific analyses. Philosophical Transactions of spatial networks. Transportation, 42, 723-731. doi:
the Royal Society B: Biological Sciences, 365(1538), 281- https://doi.org/10.1007/s11116-015-9643-9
290. doi: https://doi.org/10.1098/rstb.2009.0169 Goldstein, N. J., & Cialdini, R. B. (2007). Using social norms as a
Bas, J., Cirillo, C., & Cherchi, E. (2021). Classification of lever of social influence. In The science of social influence
potential electric vehicle purchasers: A machine (pp. 167-191). Psychology Press. Retrieved from
learning approach. Technological Forecasting and https://www.taylorfrancis.com/chapters/edit/10.4324/
Social Change, 168, 120759. doi: 9780203818565-7/using-social-norms-lever-social-
https://doi.org/10.1016/j.techfore.2021.120759 influence-noah-goldstein-robert-cialdini
Predicting the Diffusion of the Electric Vehicle: A Dynamic Approach to Model the Impact of Imitation and Experience
66
Gompertz, B. (1862). A supplement to two papers published in Trial. Marketing Science, 36(2), 290-300. doi:
the Transactions of the Royal Society," On the science https://doi.org/10.1287/mksc.2016.1011
connected with human mortality;" the one published in Pike, S., & Lubell, M. (2018). The conditional effects of
1820, and the other in 1825. Philosophical Transactions social influence in transportation mode choice.
of the Royal Society of London, 152, 511-559. doi: Research in transportation economics, 68, 2-10. doi:
https://doi.org/10.1098/rstl.1862.0026 https://doi.org/10.1016/j.retrec.2018.05.010
Higgins, A., Paevere, P., Gardner, J., & Quezada, G. (2012). Pontes, J. (2019). Electric Vehicle Sales Jump 67% In Europe—
Combining choice modelling and multi-criteria analysis CleanTechnica EV Sales Report. Retrieved from
for technology diffusion: An application to the uptake of https://cleantechnica.com/2019/03/04/electric-
electric vehicles. Technological Forecasting and Social vehicle-sales-jump-67-in-europe-cleantechnicas-europe-
Change, 79(8), 1399-1412. doi: ev-sales-report/
https://doi.org/10.1016/j.techfore.2012.04.008 Pyo, T.-H., Tamrakar, C., Lee, J. Y., & Choi, Y. S. (2023). Is social
Irle, M., Privat, F., Couret, L., Belloncle, C., Déroubaix, capital always “Capital”?: Measuring and leveraging social
G., Bonnin, E., et al. (2019). Advanced recycling of capital in online user communities for in-group diffusion.
post-consumer solid wood and MDF. Wood Material Journal of Business Research, 158, 113690. doi:
https://doi.org/10.1016/j.jbusres.2023.113690
Science & Engineering, 14(1), 19-23. doi:
Rasouli, S., & Timmermans, H. (2013). Incorporating
https://doi.org/10.1080/17480272.2018.1427144 mechanisms of social adoption in design and analysis
Ito, T. A., Larsen, J. T., Smith, N. K., & Cacioppo, J. T. (1998). of stated-choice experiments: illustration and
Negative information weighs more heavily on the brain: application to choice of electric cars.
the negativity bias in evaluative categorizations. Journal Transportation research record, 2344(1), 10-19.
of personality and social psychology, 75(4), 887. doi: doi: https://doi.org/10.3141/2344-02
https://doi.org/10.1037//0022-3514.75.4.887 Redondo, A. N., & Cagigas, A. P. (2015). Sales forecast of
Jensen, A. F., Cherchi, E., Mabit, S. L., & Ortúzar, J. d. D. electric vehicles. J. Eng. Arch, 3(1), 79-88. doi:
(2017). Predicting the potential market for electric https://doi.org/10.15640/jea.v3n1a8
Rogers, J. A., Someya, T., & Huang, Y. (2010). Materials and
vehicles. Transportation Science, 51(2), 427-440.
mechanics for stretchable electronics. science,
doi: https://doi.org/10.1287/trsc.2015.0659
327(5973), 1603-1607. doi:
Jiang, Z., & Jain, D. C. (2012). A generalized Norton–Bass
https://doi.org/10.1126/science.1182383
model for multigeneration diffusion. Management Schultz, W. P., Khazian, A. M., & Zaleski, A. C. (2008). Using
Science, 58(10), 1887-1897. doi: normative social influence to promote conservation
https://doi.org/10.1287/mnsc.1120.1529 among hotel guests. Social influence, 3(1), 4-23. doi:
Jun, D. B., & Park, Y. S. (1999). A choice-based diffusion https://doi.org/10.1080/15534510701755614
model for multiple generations of products. Shepherd, S., Bonsall, P., & Harrison, G. (2012). Factors
Technological Forecasting and Social Change, 61(1), affecting future demand for electric vehicles: A
45-58. doi: https://doi.org/10.1016/S0040- model based study. Transport Policy, 20, 62-74. doi:
1625(98)00049-3 https://doi.org/10.1016/j.tranpol.2011.12.006
Kieckhäfer, K., Volling, T., & Spengler, T. S. (2014). A Sherif, M. (1935). A study of some social factors in
hybrid simulation approach for estimating the perception. Archives of Psychology (Columbia
market share evolution of electric vehicles. University), 27(187), 1-60. Retrieved from
https://psycnet.apa.org/record/1936-01332-001
Transportation Science, 48(4), 651-670. doi:
Sherwin, H., Chatterjee, K., & Jain, J. (2014). An exploration of
https://doi.org/10.1287/trsc.2014.0526
the importance of social influence in the decision to start
Klasen, J. r., & Neumann, D. (2011). An agent-based method bicycling in England. Transportation Research Part A:
for planning innovations. International Journal of Policy and Practice, 68, 32-45. doi:
Innovation and Sustainable Development, 5(2-3), 159- https://doi.org/10.1016/j.tra.2014.05.001
184. doi: https://doi.org/10.1504/IJISD.2011.043073 Smaldino, P. E. (2022). Models of identity signaling. Current
Kormos, C., Gifford, R., & Brown, E. (2015). The influence of Directions in Psychological Science, 31(3), 231-237. doi:
descriptive social norm information on sustainable https://doi.org/10.1177/09637214221075609
transportation behavior: A field experiment. Statistics, d., key figures and energy maps. (2016).
Environment and Behavior, 47(5), 479-501. doi: Energistyrelsen. Retrieved from https://ens.dk/en/our-
services/statistics-data-key-figures-and-energy-maps
https://doi.org/10.1177/0013916513520416
Struben, J., & Sterman, J. D. (2008). Transition challenges for
Kuwano, M., Chikaraishi, M., & FUJIWARA, A. (2014).
alternative fuel vehicle and transportation systems.
Factors that promote personal mobility in relation
Environment and Planning B: Planning and Design, 35(6),
to the social network in old newtown. Asian
1070-1097. doi: https://doi.org/10.1068/b33022t
Transport Studies, 3(1), 108-124. doi:
Tietge, U., Mock, P., Lutsey, N., & Campestrini, A. (2016).
https://doi.org/10.11175/eastsats.3.108
Comparison of leading electric vehicle policy and
Lutsey, N. (2017). Integrating electric vehicles within US
deployment in Europe. International Council Clean
and European efficiency regulations. International
Transportation, 49, 847129-847102. Retrieved from
Council on Clean Transportation, 6, 1-16. doi:
https://www.researchgate.net/publication/303805519
https://doi.org/10.13140/RG.2.2.33048.80640 Train, K. E. (2009). Discrete choice methods with
Massiani, J., & Gohs, A. (2015). The choice of Bass model simulation (2nd ed.). Cambridge university press.
coefficients to forecast diffusion for innovative doi: https://doi.org/10.1017/CBO9780511805271
products: An empirical investigation for new TyreeHageman, J., Kurani, K. S., & Caperello, N. (2014).
automotive technologies. Research in What does community and social media use look like
transportation economics, 50, 17-28. doi: among early PEV drivers? Exploring how drivers build
https://doi.org/10.1016/j.retrec.2015.06.003 an online resource through community relations and
Morvinski, C., Amir, O., & Muller, E. (2017). “Ten Million social media tools. Transportation Research Part D:
Readers Can’t Be Wrong!,” or Can They? On the Role Transport and Environment, 33, 125-134. doi:
of Information About Adoption Stock in New Product https://doi.org/10.1016/j.trd.2014.09.005
67
Javier Bas, Elisabetta Cherchi, Cinzia Cirillo

APPENDIX A – Sample sociodemographics. Source (Cherchi, 2017)

Predicting the Diffusion of the Electric Vehicle: A Dynamic Approach to Model the Impact of Imitation and Experience
68

Appendix B - Table 2 variables’ description.

CTY_SL: Percentage of penetration of slow-charging stations in city centers.

SHO_SL: Percentage of penetration of slow-charging stations in shopping malls.
CTY_FA: Percentage of penetration of fast-charging stations in city centers.
SHO_FA: Percentage of penetration of fast-charging stations in shopping malls.
PP_EV: Purchasing price of the electric vehicle.
PP_GAS: Purchasing price of the gasoline vehicle.
FU_EV: Fuel cost of the electric vehicle.
FU_GAS: Fuel cost of the gasoline vehicle.
RA_EV: Driving range of the electric vehicle.
RA_GAS: Driving range of the gasoline vehicle.
CO2_EV: CO2 emissions of the electric vehicle.
CO2_GAS: CO2 emissions of the gasoline vehicle.