ArticlePDF Available

The diffusion of internet voting. Usage patterns of internet voting in Estonia between 2005 and 2015


Abstract and Figures

E-voting has the potential to lower participation thresholds and increase turnout, but its technical complexity may produce other barriers to participation. Using Rogers' theory of the diffusion of innovations, we examined how the use of e-voting has changed over time. Data from eight e-enabled elections between 2005 and 2015 in Estonia, were used to investigate changes to the profile of e-voters and contrast them to those voting by conventional means. Owing to the aggregate share of e-voters increasing with each election, with one third of voters now casting their vote remotely over the internet, there was a lack of conclusive evidence regarding whether the new voting technology had diffused homogenously among the voting population, or remained a channel for the resourceful and privileged. Our findings show that diffusion has taken place, but not until after the first three e-enabled elections. Thus, internet voting has the potential to be used by a wide range of voter types, bridge societal divisions, and emerge as an inclusive innovative voting technology.
Content may be subject to copyright.
The diffusion of internet voting. Usage patterns of internet voting in
Estonia between 2005 and 2015
Kristjan Vassil
, Mihkel Solvak
, Priit Vinkel
, Alexander H. Trechsel
, R. Michael Alvarez
University of Tartu, Estonia
Tallinn Technical University, Estonia
European University Institute, Italy
California Institute of Technology, United States
abstractarticle info
Article history:
Received 28 December 2015
Received in revised form 21 June 2016
Accepted 22 June 2016
Available online xxxx
E-voting has the potential to lower participation thresholds and increase turnout, but its technical complexity
may produce other barriers to participation. Using Rogers' theory of the diffusion of innovations, we examined
how the use of e-voting has changed over time. Data from eight e-enabled elections between 2005 and 2015
in Estonia, were used to investigate changes to the prole of e-voters and contrast them to those voting by con-
ventionalmeans. Owing to the aggregate shareof e-voters increasing with each election, with one third of voters
now castingtheir vote remotely over the internet, there wasa lack of conclusive evidence regarding whether the
new voting technology had diffused homogenously among the voting population, or remained a channel for the
resourceful and privileged. Our ndings show that diffusion has taken place, but not until after the rst three e-
enabled elections. Thus, internet voting has the potential to be used by a wide range of voter types,bridge societal
divisions, and emerge as an inclusive innovative voting technology.
© 2016 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license
Internet voting
Voting behavior
1. Introduction
Remote internet voting
has long been discussed as a means of in-
creasing voter turnout in developed democracies, especially among
younger people (Alvarez & Hall, 2004; Alvarez, Hall, & Llewellyn,
2008; Norris, 2001, 2003). However, such technology can only have a
signicant impact on political participation when its usage becomes
widely diffused. Voting technologies can empower people who have
faced participation hurdles (Vassil & Weber, 2011). Socially excluded
groups or people with reduced mobility should especially benetfrom
modes that make it easier to cast a vote (Alvarez & Hall, 2004; Gibson,
2001). Such increased empowerment might also increase voter con-
dence and their willingness to participate in elections (Alvarez & Hall,
2006; Alvarez et al., 2008). As participation is required for effective rep-
resentation, easily usable voting modes should, in theory, ensure a bet-
ter overlap between the elected representatives and society. However,
technology can also present additional barriers to thealready disadvan-
taged, in effect nullifying its theoretical promise (Berinsky, 2005; Norris,
2003). It also needs to be acknowledged that e-voting will not address
underlining reasons for abstention, such as political disillusionment or
a lack of political interest. This does not mean that internet voting is a
technological xto an issue that cannot be xed using technology.
E-voting can impact turnout among those who have accessibility prob-
lems, such as the disabled and elderly. Moreover, it can also mobilize
those who do not have clear mobility problems, but who simply do
not vote due to inconveniences related to conventional voting. Thus,
e-voting is rst and foremost a convenient voting method and therefore
should appeal to those parts of theelectorate who have abstained due to
paper voting being too cumbersome.
The actual practice of remote e-voting has been implemented in a
limited number of countries. Exactly how remote e-voting inuences
voting behavior and parties' strategies is unknown. Studies on technol-
ogy usage show that the most likely users and beneciaries are young,
technology savvy, well-resourced, and connected people (Schlozman,
Verba, & Brady, 2010; van Dijk, 2000, 2005). There is clear evidence
that the same applies to the early adopters of e-voting (Alvarez, Hall,
& Trechsel, 2009; Trechsel & Vassil, 2011). However, what we do not
know is whether e-voting has the potential to diffuse beyond this sub-
population to a broader and less homogenous group of voters, or
whether it remains a tool for those with skills and resources. As diffu-
sion is the prerequisite of e-voting having a large impact upon turnout,
discussions about how and why new modes of voting might improve
participation or representation, require empirical evidence of the
Government Information Quarterly xxx (2016) xxxxxx
This research was supported by Estonian Research Council grant nr. PUT523.
Corresponding author at: Institute of Government and Politics, University of Tartu,
Lossi 36, 51003 Tartu, Estonia.
E-mail address: (K. Vassil).
We use the terms e-voting, remote internet voting, and internet voting interchange-
ably throughout this paper to describe online voting using a remote computer and digital
identication, i.e. voting without visiting a polling station.
GOVINF-01181; No. of pages: 7; 4C:
0740-624X/© 2016 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (
Contents lists available at ScienceDirect
Government Information Quarterly
journal homepage:
Please cite this article as: Vassil, K., et al., The diffusion of internet voting. Usage patterns of internet voting in Estonia between 2005 and 2015,
Government Information Quarterly (2016),
conditions and patterns by which new technologies are adopted over
time. If the rate of adoption of a new voting technology is slow and its
diffusion limited to specic subpopulations of the electorate, it is unlike-
ly that e-voting will have a positive impact upon voter turnoutand qual-
ity of representation.
This paper addresses precisely the question: Who are the e-voters
and has their prole changed over time? We used unique cross-
sectional survey data from all eight of the legally binding e-enabled
electionsin Estonia between 2005 and 2015. Our goal was to determine
whether the technology has diffused among the voter population, or
whether it remains a convenient technical solution for a group of people
already engaged in politics and who face limited barriers to participa-
tion in the rst place.
1.1. E-voting in Estonia
Since 2005, Estonia has had a total of eight e-enabled elections
where eligible voters could cast binding ballots over the internet. Inter-
net voting has been used for local, national and European elections. The
number of e-voters in the rst e-enabled election was only 9317 (Fig. 1).
However, the number increased in each succeeding election, reaching
176,491 in the 2015 national elections. In relative terms, the share of in-
ternet votes of total votes grew from a mere 2% in 2005 to N30% in 2014
and 2015.
A prerequisite for casting an electronic vote is a credit card sized
electronic ID-card,
which are compulsory for all Estonian residents.
Using digital identication, voters can use their personal computers
when connected to the internet and equipped with a smart card reader,
to cast an electronic vote (Alvarez et al., 2009). E-voting is available dur-
ing the advanced voting period via a website hosted by the EstonianNa-
tional Electoral Committee (20052011). E-voting itself involves three
steps; rst, the user opens the website and with their ID-card and rst
PIN-code to identify themselves, enters the system; second, after the
system has veried the identity of the voter, it displays the list of candi-
dates by party in the voter's respective district; third, by clicking on a
candidate's name and then entering their second PIN-code, the voter
casts their vote.
The rst ve elections were reasonably similar for the user-end, with
the only marked difference being the length of period during which e-
voting was available: three days in 2005 and 2007; and 7 days in
2009, 2011 and 2013. From 2009, e-voters needed to download a voting
program instead of voting via the web-embedded application. In 2013, a
vote verication feature was introduced to the e-voting system that
allowed voters to verifyusing a smartphone or tabletwhether their
electronic vote was received as cast. Other than these differences, the
eight e-enabled elections were reasonably similar, providing a valid
point of comparison of the related dynamics in user behavior.
On the technical side, e-voting requires internet access and a mini-
mum level of computer literacy, both of which are not universal in
Estonia. However, the act of e-voting is no more difcult than other on-
line activities, such as banking or shopping.
2. Measuring diffusion
Theories on the diffusion of technological innovations provide a
foundation for measuring and explaining the potential spread of e-
voting in a society. The classical accounts of the diffusion of innovations
provided by Ryan and Gross (1943) and Rogers (2003[1962]) have
stood the test of time, being used over the years to explain a wide vari-
ety of phenomena, rangingfrom the spread of agricultural practices (e.g.
Fliegel, 1993) to political reforms and policies (e.g. Starr, 1991; Jahn,
2006), medical practices (e.g. Greenhalgh, Robert, Macfarlane, Bate, &
Kyriakidou, 2004), management (e.g. Abrahamson, 1991), and most
crucially, technological applications in very different elds (e.g.
MacVaugh & Schiavone, 2010). Rogers' (2003) account sees the diffu-
sion of technology as a sequence of steps in an innovation decision pro-
cess. This process includes gaining knowledge of the technology, being
convinced of its usefulness, and ultimately, deciding to implement it.
Adoption occurs if expectations are positively conrmed by experience.
Once a distinct subgroup has reached the adoption stage and built up a
critical mass of users, subsequent diffusion is reminiscent of a bank-run,
where the number of people adoptingit is partly a function of the num-
ber of prior adopters (Rogers, 2003: 206). This sequence has been dem-
onstrated to apply to both collective and individual actors (see Wejnert,
The crucial aspect of using Rogers' account to explain e-voting
regards the changing prole of adopters of technology at different
stages of the process. The rst adopters tend to be a small number of
well-informed, innovative risk-takers (Rogers, 2003: 263). The second-
ary and tertiary adopters should more closely resemble thegeneral pop-
ulation,and the unique characteristics associated with therst adopters
should continually become less prominent. Eventually, even technolog-
ical laggards might be motivated to adopt the technology, as the relative
gains outweigh the costs of adopting (Rogers, 2003:263265).
As with every new internet technology, adoption requires a certain
level of digital literacy, which is not always evenly distributed across so-
cial groups. This suggests that internet voting is most likely to appealto
Fig. 1. Dynamics of e-voting in Estonia, 20052015.
Since 2011voters can alsouse a smartphone-based mobileID (using a special SIM card
and PIN-codes) to authenticate themselves to the e-voting system. The ID card, however,
is the more widely used identication method.
For further details on the process of e-voting, see:
estonia/engindex/; Estonian National Electoral Committee (2005);OSCE/ODIHR (2007,
2011);Vassil and Weber (2011);Trechsel and Vassil (2011).
2K. Vassil et al. / Government Information Quarterly xxx (2016) xxxxxx
Please cite this article as: Vassil, K., et al., The diffusion of internet voting. Usage patterns of internet voting in Estonia between 2005 and 2015,
Government Information Quarterly (2016),
those with a good command of modern technologies. It is precisely this
mechanism that has fueled accounts claiming that e-democracy in gen-
eral reects underlining societal divisions and even augments these by
further marginalizing the marginalized and connecting the connected
(Alvarez & Nagler, 2000; Van Dijk, 2000, 2005; Margolis and Resnick,
2000; Putnam, 2001; Wilhelm, 2000). New voting technologies might
therefore also only diffuse amonga distinct non-random subpopulation
of voters distinguished by higher socio-economic status, but not
Eventually it remains an empirical question that can be tested using
Roger's theoretical framework. The literature on diffusion suggested
two different expectations for our subsequent analyses. First, we ex-
pected a gradual dispersion of e-voting usage, driven by early adopters,
who would be distinguishable by their socio-demographic prole:
younger, better-educated, and comparatively economically well-off in-
dividuals. Presumably, they should also be more technology savvy and
trust the technology more. As the use of technology widens with each
additional e-enabled election, a gradual increase in the number of e-
voters should be observable. Over time, these new e-voters should re-
sult in the total group of e-voters becoming less distinct from paper bal-
lot voters, both in terms of socio-demographics, as well as behavioral
and attitudinal characteristics. If diffusion of a technology is actually tak-
ing place, the population of e-voters will become more heterogeneous
over time. Thus, we expected that the characteristics associated with
the likelihood of e-voting during the rst e-enabled elections would
subsequently become less pronounced and lose their explanatory
power. Based on this, we formulated a rst hypothesis to identify the
presence of diffusion:
H1. Characteristics that explained e-voting during the rst e-enabled
elections will lose their predictive power over time, suggesting that
usage of internet voting has diffused among the electorate.
A contrary expectation, however, follows from the fact that a new
voting technology can present a barrier to potential users. Thus, al-
though a technology mightspread according to the initial sequence sug-
gested by Rogers and user numbers initially rise, a barrier may prevent
larger segments of potential users from adopting the new technology.
We posited that the likely barriers would be an insufcient level of dig-
ital literacy, a lack of trust in thee-voting system, andage related factors.
If such barriers do indeed exist e-voters will remain a distinct subgroup
of total voters, because the growth of e-voters will plateau owing to the
technology failing to bridge the gap between more and lesstechnology-
savvy voters. It would also mean that rst-time e-voters should remain
clearly distinguishable frompaper ballot voters, irrespective of time or a
growth in e-voters. Thus, our competing hypothesis was:
H2. Characteristics that explained e-voting during the rst e-enabled
elections would retain their predictive power over time, suggesting
that usage of internet voting has not diffused among the electorate.
3. Data, variables and model specication
In order to investigate whether diffusion of e-voting occurred, we
used a unique series of individual-level surveys, whereby data were col-
lected after each of the eight e-enabled elections in Estonia. We chose to
work with different types of elections in one temporal sequence. Thus,
we deliberately ignored the possibility that different election types
may mobilize different voter types, vary in saliency, and inuence over-
all turnout. However, as we were interested in measuring diffusion as a
function of recurring experiences with e-voting, we compared elections
over time. We argue that diffusion should be observable irrespective of
whether elections are treated as one temporal sequence or grouped ac-
cording to type or cycle, as time is a proxy for cycle and vice versa. We
return to the empirical implications of this analytical choice in the dis-
cussion section.
The rst ve surveys consisted of quota sampling (with the sample
containing an almost equal share of internet voters, ballot-paper voters,
and non-voters), to ensure a sufcient number of e-voters for analysis;
all surveys had a sample size of 1000 respondents and used the CATI
method. The three subsequent surveys consisted of stratied random
sampling, because the number of e-voters in the overall voting popula-
tion had become sufciently large for analysis (Fig. 1). They also had a
sample size of 1000 respondents and used the CAPI method. The sur-
veys had response rates of 62.3%, 61.7% and 60.0% in 2013, 2014 and
2015 respectively.
All were post-election surveys conducted during
the three week period post-election day and are representative of the
voting eligible population. Table 1 shows the sample composition ac-
3.1. Variable selection
Our dependent variable consisted of a dichotomy of factors that dis-
tinguished e-voters from ballot-paper voters. However, we must be ex-
plicit about our choice of response category of interest, because
comparing e-voters to regular voters over time across elections inevita-
bly introduces noise. Such noise was owing to the fact that in later elec-
tions, the population of e-voters contained rst-time e-voters (early
adopters in the rst elections) and those who had voted online in mul-
tiple elections. As the motivations and characteristics of these groups
may differ considerably (Rogers, 2003), we preferredin-line with our
theoretical argumentto decompose the response category to distin-
guish between rst time and recurring e-voters in each election. We
chose to compare ballot-paper voters (coded as 0) only to rst time e-
voters (coded 1) in each e-enabled election. Effectively, our
operationalization yielded a response category that captured rst time
e-voters that were: early adopters in the rst elections; the early major-
ity and majority voters in subsequent elections; and late majority and
laggards in the most recent elections. Any change in the prole of rst
time e-voters would thus reveal whether diffusion of the new voting
technology occurred.
Prior studies in diverse settings on diffusion patterns have demon-
strated effects of socio-demographic and economic factors, including
ethnicity, in predicting diffusion among actors (e.g. Berry & Berry,
1990; Hedström, 1994; Tolnay & Glynn, 1994). More importantly,
prior studies on e-voting in Estonia have shown that age, education,
trust in the e-voting system, and rst-language, should be particularly
strong predictors of whether someone e-votes (Trechsel & Vassil,
2011). The latter is true due to the fact that Estonia is a multilingual so-
ciety with about one third of the population Russian-speaking. As the
system of e-voting is only offered in the Estonian language, it could
limit its use among the Russian-speaking minority (Trechsel & Vassil,
2010). In addition, economic well-beingand literacy with new technol-
ogies, have been shown to be systematically correlated with the likeli-
hood of internet voting (Alvarez et al., 2009; Trechsel & Vassil, 2011;
Vassil & Weber, 2011). Finally, people's political self-positioning (e.g.
left or right) may provide a useful control for our models, because ag-
gregate election results consistently show that liberal parties gain
more e-votes than those on the ideological left.
Followingthese empirical accounts, we used the following indepen-
dent variables for our analysis: both age in years and age squared (to
allow for non-linear effects); education (using a dummy variable for
higher education, with secondary and elementary education as the ref-
erence); gender (male = 1; female = 0); ethnicity (Estonian as home
language = 1; Russian as home language = 0); income (measured by
decile); computer literacy (using a dummy for good skills level, with av-
erage and basic levels as reference); trust toward the e-voting system
Comparable responserates for earlier surveys cannotbe computed because they used
quota-sampling and CATI methods.
Further information is available on the National Electoral Committee website: www.
3K. Vassil et al. / Government Information Quarterly xxx (2016) xxxxxx
Please cite this article as: Vassil, K., et al., The diffusion of internet voting. Usage patterns of internet voting in Estonia between 2005 and 2015,
Government Information Quarterly (2016),
(trust = 1; no trust= 0)
; and political left-right self-positioning (using
3.2. Model specication
In order to evaluate whether a diffusion mechanism was at play, we
estimated a separate logistic regression model for each of the eight elec-
tions, and compared the coefcients for each independent variable and
model t over time. If the diffusion process happened as stated in H1,
we should observe a gradual reduction in model tand disappearing co-
variate effects in terms of magnitude and statistical signicance over
time, which would conrm that the sample of rst time e-voters had in-
deed become more heterogeneous. By contrast, if the effects retained
their power and signicance, it would suggest that the sample of rst
time e-voters in later elections were as distinct in terms of their charac-
teristics as during the rst few elections, thus conrming H2. The model
took the following generic form (Eq. (1)):
ln Pr evote ¼1ðÞ
1Pr evote ¼1ðÞ
where X
is the vector of the listed independent variables for individual
i; the model was estimated separately for each of the eight elections.
Our particular interest was in how β
changes over time. We were not
interested in any particular covariate effects per se, but whether and
how the effects of independent variables changed over time.
For improved interpretation, we converted the logistic regression
coefcients into average marginal effects, which show the average of
the variation induced in the probability of interest by a marginal change
in an independent variable for each individual in the sample (Baum,
2006). An average marginal effect is interpreted as an effect of one-
unit change of the independent variable on the change of probability
of interest. The appeal of average marginal effects in our analytical set-
ting, was that they are less affected by unobserved heterogeneity that
is unrelated to the independent variables in the model, and can thus
be compared across models, groups, samples, or years (Mood, 2010:
78). Missing values were multiply imputed for all datasets.
4. Findings
The ndings from all eight regression models, with relevant tdiag-
nostics, are presented in Table 2. First and foremost, we saw that associ-
ations between e-voting and age, ethnicity, computer literacy, and trust
toward e-voting, weakened substantially with time. More specically,
Fig. 2 displays the non-linear impact of age-squared on the likelihood
of internet voting. The ndings clearly show that the effect of age
attens gradually over time. Only for the rst three elections did age
have the expected inverted U-shape effect on the probability of voting
online. The likelihood of internet voting was initially highest among
40 to 50 year olds, and lowest for the younger and older. However,
this once strong relationship started to gradually disappear after the
third e-enabled election in 2009, attening and losing its predictive
power entirely by the fourth election. We assumed from this that over
time, the likelihood of e-voting becomes almost equally probable for
all age groups.
Similarly, and in accordance with previous studies, we found that
the Estonian language was an important predictor of e-voting during
the rst e-enabled elections. Between the 2005 and 2009 elections, eth-
nic Estonians were approximately 2638 percentage points more likely
to vote online compared to non-Estonians. However, this difference was
lower by more than half by 2011, and had completely vanished by the
2014 elections. We infer from this that ethnicity has lost its explanatory
power over internet voting, and aswith the effect of age, most of thisex-
planatory power seemed to disappeared after the third election, render-
ing the once signicant disparity between Estonian and Russian-
speakers, with respect to internet voting, negligible (Fig. 3).
Computer literacy followed the same trail: it shows strong associa-
tion with the likelihood of internet voting, with those with high PC-
skills approximately 17 percentage points more likely to vote online
than those with average and poor skills. This is not surprising, as the
general setup of the e-voting system requires several interactions with
a computer, relevantperipherals, andthe ID-card. The effect was consis-
tent over the rst two elections, after which it became insignicant, pro-
viding evidence that voters may have become more familiar with the
system and learned to useit (Fig. 3). However, we did nd it surprising-
ly that the effect of PC-literacy reappeared during the last election of
2015, with a small effect at the 0.01 level. We believe this was partly
due to differences in the electorates between different election types.
Trust toward the system of e-voting has been shown to be one of the
strongest predictors of e-voting (Trechsel & Vassil, 2011). We observed
the same, but only for the rst e-enabled elections. In particular, we saw
that thosewho trusted the system of e-voting were about 49 percentage
points more likely to vote online than those who found it less trustwor-
thy. The effect hovered between 35 and 70 percentage points in the rst
four e-enabled elections and then signicantly lost its explanatory
power. Unlike previous variables, trust decreased substantially in effect
size, but retained its statistical signicance (Fig. 3).
Regarding education, gender, income, and left-right self-position, we
found no substantially strong relationships (Table 2). A higher educa-
tion appeared to be weakly but positively associated with internet vot-
ing, though its effect was not consistent. The same was the case for
gender and income. As for left-right self-positioning, we found it partic-
ularly reassuring that nowhere in the data did we nd a statistically sig-
nicant and sizable effect to provide evidence that internet voting is
unequally likely for those on the left compared to the right of the polit-
ical spectrum.
Taken together, we found that multiple socio-demographic and atti-
tudinal variables were strongly associated with the likelihood of
Table 1
Voter types in the sample.
Voter type 2005 local 2007 national 2009 EP 2009 local 2011 national 2013 local 2014 EP 2015 national Total
Normal voter 318 450 448 403 480 560 477 613 3749
(%) (33.9) (45.8) (44.93) (40.3) (47.7) (53.8) (48.0) (61.4) (47.1)
1st time e-voter 315 309 264 108 139 65 48 62 1310
(%) (33.6) (31.5) (26.5) (10.8) (13.8) (6.2) (4.8) (6.2) (16.5)
Recurring e-voter 0 60 85 142 72 106 90 134 689
(%) (0.0) (6.1) (8.5) (14.2) (7.2) (10.2) (9.1) (13.4) (8.7)
Non-voter 306 163 200 347 316 311 379 190 2212
(%) (32.6) (16.6) (20.1) (34.7) (31.4) (29.9) (38.1) (19.0) (27.8)
Total 939 982 993 1000 1007 1042 994 999 7956
(%) (100.0) (100.0) (100.0) (100.0) (100.0) (100.0) (100.0) (100.0) (100.0)
Column percentages might not sum to 100% due to rounding.
The dummy was created from a four item Likert scale (20052011) or a 10 category
ordinal item scale (201 32015). Both variables were split at the middle, with people
who trusted or tended to trust coded as 1, and those with no trust or a tendency not to
trust coded as 0.
We used STATA's margins package to compute averagemarginal effects, and their mi-
impute package to multiply impute missing values.
4K. Vassil et al. / Government Information Quarterly xxx (2016) xxxxxx
Please cite this article as: Vassil, K., et al., The diffusion of internet voting. Usage patterns of internet voting in Estonia between 2005 and 2015,
Government Information Quarterly (2016),
internet voting in the rst three e-enabled elections, but they started to
gradually lose their explanatory power over time, becoming less and
less relevant. Thus, we took this as therst evidence of diffusion having
taken place and that the data supports H1.
However, as the diffusion process should render the rst-time e-
voter population more heterogeneous overtime, not only the individual
effects of covariates should weaken, but also the t of the model. Con-
versely, ifno diffusion has taken place, model t should remain relative-
ly immune to change over time. Fig. 4 presents various model t
parameters over time. First, there was a signicant drop in model per-
formance, as measured by the drop in pseudo-R
healthy 0.52 in 2005 to 0.18 in 2015. This clearly points to increasing
heterogeneity among rst time e-voters, providing evidence that diffu-
sion was indeed taking place. More importantly, there was a radical
drop in the model's sensitivity, i.e. its ability to correctly classify rst
time e-voters (the true positives). Fig. 4 shows that it substantially
dropped after the third election. Moreover, from the 2013 election on-
ward the model failed to classify e-voters. In sum, the diminishing effect
of covariates in explaining rst time e-voters, as well as a lower overall
model performance, with time, allowed us to refute H2 and support the
thesis of ongoing diffusion based upon the increasing heterogeneity of
e-voters. Notice that the general classication accuracy of the model
remained high throughout the years; this was because it initially cor-
rectly predicted e-voters and ballot-paper voters, and continued to
Table 2
Predicting rst time e-voting (base: only paper ballot voters).
2005 local 2007 national 2009 EP 2009 local 2011 national 2013 local 2014 EP 2015 national
Age 1.72⁎⁎ 1.77⁎⁎ 1.97⁎⁎⁎ 1.28 0.54 0.24 0.23 0.57
(0.58) (0.57) (0.58) (0.69) (0.52) (0.42) (0.37) (0.39)
0.01 0.00 0.00 0.01
(0.01) (0.01) (0.01) (0.01) (0.01) (0.00) (0.00) (0.00)
Estonian language 26.05⁎⁎⁎ 38.40⁎⁎⁎ 29.86⁎⁎⁎ 11.8614.66⁎⁎ 2.311.68 5.92
(5.23) (7.08) (7.50) (4.92) (5.57) (3.43) (2.61) (3.60)
PC literacy: good 17.35⁎⁎⁎ 16.60⁎⁎⁎ 3.75 0.86 5.40 5.10 3.57 9.88⁎⁎
(3.56) (3.41) (3.68) (3.75) (3.34) (2.82) (2.72) (3.06)
Trust e-voting 49.25⁎⁎⁎ 36.23⁎⁎⁎ 69.56⁎⁎⁎ 35.30⁎⁎⁎ 25.25⁎⁎⁎ 14.80⁎⁎⁎ 9.25⁎⁎ 13.50⁎⁎⁎
(4.00) (5.35) (11.51) (7.67) (3.73) (3.63) (3.28) (3.38)
Left-right self-position 0.17 0.13 0.04 0.26 0.54 0.42 0.32 0.19
(0.72) (0.78) (0.66) (0.90) (0.70) (0.53) (0.48) (0.63)
Education: higher 5.24 8.81⁎⁎ 9.01⁎⁎ 5.06 8.98⁎⁎ 5.501.50 4.34
(3.33) (3.29) (3.26) (3.52) (3.03) (2.48) (2.28) (2.63)
Male 1.04 2.05 0.47 0.09 6.381.81 1.56 5.11
(3.13) (3.12) (3.18) (3.35) (2.91) (2.36) (2.10) (2.41)
Income decile 0.31 1.79⁎⁎ 0.85 0.97 0.97
0.10 0.23 0.58
(0.58) (0.59) (0.60) (0.63) (0.48) (0.46) (0.42) (0.53)
Constant 7.49⁎⁎⁎
(1.12) (0.96) (1.25) (1.32) (1.11) (1.38) (1.50) (1.39)
Observations 633 759 712 511 619 625 646 617
Nagelkerke Pseudo R
0.52 0.36 0.41 0.34 0.37 0.24 0.16 0.18
Correctly classied 0.78 0.74 0.72 0.81 0.81 0.90 0.93 0.90
Sensitivity 0.90 0.67 0.68 0.24 0.40 0 0 0
Specicity 0.66 0.79 0.75 0.96 0.76 1 1 1
Average marginal effects as percentages. Standard errors in parentheses.
⁎⁎⁎ pb0.001.
⁎⁎ pb0.01.
Fig. 2. Effect of age on the likelihood of e-voting.
5K. Vassil et al. / Government Information Quarterly xxx (2016) xxxxxx
Please cite this article as: Vassil, K., et al., The diffusion of internet voting. Usage patterns of internet voting in Estonia between 2005 and 2015,
Government Information Quarterly (2016),
accurately predict (only) on-paper voters, this further supporting the
evidence on diffusion.
5. Discussion
E-voting has become a widely used voting mode in Estonia. Howev-
er, the aggregate number of e-voters might disguise a situation where
the technology has not diffused across societal boundaries, but instead
is only being increasingly used by a distinct subpopulation of well-
resourced, technologically savvy voters. Real diffusion over time
would mean that voters from a broad cross-section of the population,
regardless of their social status or level of resources, use e-voting.
Followingthe expectations derived from Rogers' diffusion of innova-
tion, we examined the prole of rst time e-voters over the course of
eight e-enable elections (a period of ten years, 20052015), to deter-
mine to what degree usage of this new voting technology has been
adopted by the wider voter population. The aggregate number of e-
voters increased sizably over time, with every third vote being cast on-
line during the past two elections. At the level of the individual, we
found that the characteristics of rst time e-voters became more similar
to the characteristics of traditional paper ballot voters over time. E-
voters used to be Estonian-speakers from a distinct age group, who
have good computer literacy and trust in the system of internet voting.
However, this was only the case for the rst three elections in which e-
voting was used. From the fourth election onward, we consistently saw
that these characteristics were only weakly, if at all, associated with the
choice to vote online. As a result, our model's ability to predict and cor-
rectly classify rst-time e-voters based only on socio-demographic and
attitudinal data, becomes increasingly limited. Our results show that e-
voting has diffused among the overall voter population, and not just
remained an activity of the privileged few. Importantly, we found that
the process of diffusion did not occur immediately, but was shown via
a plateau effect, by which diffusion became visible only after the rst
three elections.
We focused on elections in one temporal sequence, irrespective of
their type. Alternatively, one could focus on electoral cycles by separat-
ing them by type. We chose not to do so, because we modelled the dif-
fusion of e-voting in one country, where eligible voters substantially
overlap from one election to another. However, if one did focus on elec-
toral cycles, the evidence provided in this paper should also point to-
ward the diffusion. Namely, some of the effects already started to
disappear after the second, and others after the third, electoral cycle.
As a result of our ndings, we draw two main conclusions. First,
technology has the potential to bridge societal divisions and ease polit-
ical participation, not only for the already connected and resourceful,
but also for the less privileged, who have fewer resources and remain
at the periphery of using modern technologies. Similarly, internet vot-
ing may appeal to those nding conventional voting to cumbersome.
As a more convenient mode of participation e-voting may also have
the potential to ease participation for those who are connected and en-
gaged but may still abstain due to the inconveniences related to on-
Fig. 3. Impact of computer literacy, ethnicity, trust and ideological auto-position on the
likelihood of internet voting. Whiskers represent 95% condence intervals. (L local, N
national, EP European Parliament election.).
Fig. 4. Model tdiagnostics(Llocal, N national, EP European Parliament election).
6K. Vassil et al. / Government Information Quarterly xxx (2016) xxxxxx
Please cite this article as: Vassil, K., et al., The diffusion of internet voting. Usage patterns of internet voting in Estonia between 2005 and 2015,
Government Information Quarterly (2016),
paper voting. The experience of e-voting usage in Estonia shows that
technology should not be considered as a hurdle, but as an enabler for
political participation. The caveat is that technology only provides an ef-
cient mode for participation; structural hurdles that inhibit participa-
tion in general, regardless of the mode voting, will most likely stay
unaffected. However, what we have demonstrated in this paper is that
technology itself does not seem to exclude anybody, as the skeptics
have suggested.
The second conclusion is that the potential enabling effects did not
surface immediately in the electoral realm after the introduction of
the new voting technology, but required a period of at least three elec-
tions to appear. Adoption amonga select subgroup can happen immedi-
ately, but this is limited to people who already have the resources and
skills to use new technologies. A wider public benet can only be real-
ized once usage has diffused and this does take time. Policymakers are
well advised not to expect immediate results following the introduction
of new voting technologies, but should recognize that different sub-
groups of the electorate adopt and use new technologies at different
rates. From a positive perspective, our evidence showed the process to
be fairly quick; characteristics that used to predict internet voting
started to lose their predictive power after only three separate elections
within four years. What seemed to matter most was not time as such,
but the frequency of being exposed to the possibility of casting their
vote over the internet.
Regarding the generalizability of the three-election argument to
other contexts, we point out that Estonia was an early adopter of inter-
net voting. Ten years ago internet penetration, broadband communica-
tions, and the use of social media, was markedly lower than today. In
countries where such factors are higher, therate of diffusion of internet
voting may be substantially accelerated following its introduced.
Taken together, we found evidence that e-voting has diffused among
a wide and heterogeneous group of Estonian voters, and has not just be-
come an exclusive form of participation for a privileged few. That
Estonian e-voters are a widespread and heterogeneous group was con-
vincingly shown by the model t, which went from excellent to ex-
tremely poor in just over the course of eight elections. Therefore, we
are condent that new voting technologies are not necessarily exclu-
sive, as early studies on e-voting have suggested, but are inclusive for
a wide range of voter types.
Abrahamson, E. (1991).Managerial fads and fashions:The diffusion and rejection of inno-
vations. Academy of Management Review,16(3), 586612.
Alvarez, R. M., & Hall, T. E. (2004). Point, click, and cote: The futur e of internet voting.
Brookings Institution Press.
Alvarez, R. M., & Hall, T. E. (2006). Controlling democracy: The principalagent problems
in election administration. Policy Studies Journal,34(4), 491510.
Alvarez, R. M., & Nagler, J. (2000). The likely consequences of internet voting for political
representation. Loyola of Los Angeles Law Review,34,11151154.
Alvarez, R. M., Hall, T. E., & Llewellyn, M. H. (2008). Are Americans condenttheir ballots
are counted? The Journal of Politics,70(3), 754766.
Alvarez, R. M., Hall, T. E.,& Trechsel, A. H. (2009). Internet voting in comparative perspec-
tive: The case of Estonia. Political Science & Politics,42(3), 497505.
Baum, C. F. (2006). An Introduction to Modern Econometrics Using Stata. Stata Press.
Berinsky,A. J. (2005). The perverse consequences of electoral reform in the United States.
American Politics Research,33(4), 471491.
Berry, Frances S]>F. S., & Berry, W . D. (1990). State lotte ry adoptions as policy i nnova-
tions: An event history an alysis. The American Political Science Review,84(2),
van Dijk, J. (2005). The deepening divide: Inequality inthe information society. Sage Publica-
tions, Inc.
van Dijk, J. (2000). Widening information gapsand policies of prevention. In K. L. Hacker,
& J. van Dijk (Eds.), Digital democracy: Issues of theory and practice (pp. 166183).
London: Sage.
Estonian National Electoral Committee (2005). Estonian I-voting system: General de-
scription. (February 12, 2014)
Fliegel, F. C. (1993). Diffusion research in rural sociology: The record for the future. (West-
port, CT).
Gibson, R. (2001). Electionsonline: Assessing internet voting in light of the Arizona Dem-
ocratic Primary.Political Science Quarterly,116(4), 561583.
Greenhalgh, T., Robert, G., Macfarlane, F., Bate, P., & Kyriakidou,O. (2004). Diffusion of in-
novations in ser vice organizations: Systematic review an d recommendations.
Milbank Quarterly,82(4), 581629.
Hedström, P. (1994). Contagious collectivities: On the spatial diffusion of Swedish trade
unions, 18901940. American Journal of Sociology,99(5), 11571179.
Jahn, D. (2006). Globalization as 'Galton's Problem':The missing link in the analysis of dif-
fusion patterns in welfare state devel opment. International Organization,60(2),
MacVaugh,J., & Schiavone, Francesco]>F. (2010). Limitsto the diffusion of innovation: A
literature review and integrative model. European Journal of Innovation Management,
13(2), 197221.
Margolis, M., & Resnick, D. (2000). Politics as usual: The cyberspace revolution. Inc: Sage
Mood, C. (2010). Logistic regression: Why we cannot do what we think we can do, and
what we can do about it. European Sociological Review,26(1), 6782.
Norris, P. (2001). Digital divide: Civic engagement, information poverty, and the internet
worldwide. Cambridge University Press.
Norris, P. (2003). Will new technology boost turnout? Evaluating experiments in e-voting
v. All-postal voting facilities in UK local elections. Paper for the British Study Group
Seminar Friday 31st October 2-4 pm, Minda de Gunzberg Center for European studies.
Harvard University.
OSCE/ODIHR (2007). Election assessment mission report in the 2007 parliamentary elec-
tions in Estonia . (December 20,
OSCE/ODIHR (2011). Election assessment mission report in the 2011 parliamentary elec-
tions in Estonia. (December 20, 2013)
Putnam,R.(2001).Bowling alone: The collapse and revival of American community. Simon
and Schuster.
Rogers, M. E. (2003). Diffusion of innovations (5th ed.). New York: Free Press (1962).
Ryan, B., & Gross, N. C. (1943). The diffusion of hybrid seed corn in two Iowa communi-
ties. Rural Sociology,8,1524.
Schlozman, K. L., Verba, S., & Brady, H. E. (2010). Weapon of the strong? Participatory in-
equality and the internet. Perspectives on Politics,8(2), 487509.
Starr, H. (1991). Democratic dominoes. Diffusion approaches to the spread of democracy
in the international system. Journal of Conict Resolution,35(2), 356381.
Tolnay, S. E., & Glynn, P. J. (1994). The persistence of high fertility in the American south
on the Eve of the baby boom. Demography,31(4), 615631.
Trechsel, A. H., & Vassil, K. (2010). Internet voting in Estonia - a comparative analysis of
four elections since 2005. Report for the Council of Europe. Council of Europe.
Trechsel, H. A., & Vassil, K. (2011). Internet voting in Estonia. A comparative analysis of
five elections since 2005. Report to the Estonian Na tional Electoral Comm ittee. (January 12, 2014)
Vassil, K., & Weber, T. (2011). A bottleneck model of e-voting: Why technology fails to
boost turnout. New Media & Society,13(8), 13361354.
Wejnert, B. (2002). Integrating models of diffusion of innovations: A conceptual frame-
work. Annual Review of Sociology,28,297326.
Wilhelm, A. (2000). Democracy in the digital age: Challenges to political life in cyberspace.
Kristjan Vassil is a senior research fellow at the Johan Skytte Institute of Political Studies,
University of Tartu. He defended his doctoral degree in the European University Institute
in January 2012 and his thesis focused on the effect of information and communication
technologies on voting behavior.
Mihkel Solvak is a senior research fellow at the Johan Skytte Institute of Political Studies,
University of Tar tu. He defended his doct oral dissertation in the University o f Tartu,
Estonia.His dissertationfocused on parliamentary roll-callbehavior. Mihkelhas published
a wide rangeof academic peer-reviewedjournal focusingon voting behavior,politicalpar-
ticipation and technology.
Alexander H. Trechsel (PhD), a professor at the European University Institute, Florence,
Italy; faculty fellow at Berkman Center for Internet & Society, Harvard University, Cam-
bridge, MA, USA.
R. MichaelAlvarez (PhD), a professor of political scienceat California Institute of Technol-
ogy, Pasadena, California, US. For more than a decade, R. Michael Alvarez has been one of
the most prominent academic researchers studying electiontechnologies, especially elec-
tronic and Internet voting.
7K. Vassil et al. / Government Information Quarterly xxx (2016) xxxxxx
Please cite this article as: Vassil, K., et al., The diffusion of internet voting. Usage patterns of internet voting in Estonia between 2005 and 2015,
Government Information Quarterly (2016),
... The proposed scheme is then utilized to implement an escrowfree Identity-Based Encryption (IBE) scheme that ensures confidentiality, and an escrow-free Identity-Based Signature (IBS) scheme that is secure against forgery. The Electronic Voting System (EVS) [161] is undoubtedly more efficient, participatory, and fair than the traditional postal voting system. However, inconsistencies have been observed in the EVS system, including ballot privacy, voter anonymity, transparency, and auditing. ...
Full-text available
Elliptic curve cryptography (ECC) is a remarkable mathematical tool that offers the same level of security as traditional public-key cryptography (PKC) with a significantly smaller key size and lower computational requirements. The use of pairing on elliptic curves has emerged as a vibrant field of research that provides enhanced security measures for the next generation of cryptographic systems. This thesis explores using ECC and Pairing-Based Cryptosystems (PBC) as effective mathematical tools for achieving high-security levels with minimal key size and computation cost. Specifically, the research aims to analyze Pairing-Friendly Elliptic Curves (PF-EC) and their practicality in resource-constrained environments. It proposes solutions to some of the limitations of existing applications of pairing-based cryptography. The thesis begins by presenting a comprehensive framework for constructing PF-EC and evaluating the practical security of several families of pairing-friendly curves. The study then explores the limitations of Identity-Based Encryption (IBE), a recognized application of pairing-based cryptography. It proposes mechanisms to address issues such as key escrow, secure key issuing, user slandering, and key abusing problems. The proposed solutions include an Escrow-Free Identity-Based Encryption (EF-IBE) scheme secured against confidentiality and an Escrow-Free Identity-Based Signature (EF-IBS) scheme that is forgeable and secure.
... No Russian speakers were included. Since different ethnic groups have different migration patterns (Mooses et al. 2020) as well as connection with digital society solutions (Vassil et al. 2016), only Estonian speakers were approached for the research. Suitable people were found from acquaintances of the authors, but also from the Facebook groups 'I'm an Estonian abroad,' "FinEst -Estonians in Finland," and 'Estonians in London.' Emigrated Estonians who do not maintain a connection with Estonia were not included in the study. ...
People who are simultaneously embedded into multiple countries are referred to as transnationals. We have investigated the influence of information and communications technology (ICT) on the transnational communities of Estonia, one of the world’s frontline countries when it comes to technological usage. We conducted interviews with Estonians who permanently live abroad to discover how ICT helps them stay connected with their country of origin. Transnational Estonians ensure the continuance of their connections via ICT. Thus, they directly participate in their home countries via ‘ways of being active,’ which strengthens their transnational identities.
... The research paper in [9]. This research looks to find out the changing of e-voting over time by using Rogers' theory of the diffusion of innovations. ...
Full-text available
Online voting is possible and can be a lot easier than face-to-face voting because it will save money by not having to pay people to manually count votes. It can also make voting easier for people with disabilities, people who live abroad, and people who live in rural areas. Unfortunately, because of the insincerity that is found in politics and politics, doing online voting will give the dubious politicians who always try to cheat during voting a perceived chance to cheat. That is why this research proposed an online voting system that would include features such as election administration, survey administration, and facial authentication. Laravel, a PHP framework, was used to create the online voting system. Deep learning in Python was used to perform face authentication. The system that has been developed can be used to properly conduct elections, conduct surveys, make announcements, and use face authentication for identification prior to casting a ballot, among other things.
Full-text available
One main objective of higher education is to provide quality education to its students. One way to achieve the highest level of quality in the higher education system is by discovering knowledge for prediction regarding enrolment of students in a particular course, alienation of traditional classroom teaching model, detection of unfair means used in online examination, detection of abnormal values in the result sheets of the students, and prediction about students’ performance. The knowledge is hidden among the educational data set and is extractable through data mining techniques. The present paper is designed to justify the capabilities of data mining techniques in the context of higher education by offering a data mining model for the higher education system in the university. In this research, the classification task is used to evaluate student’s performance, and as many approaches are used for data classification, the decision tree method is used here. By this, we extract data that describes students’ summative performance at semester’s end, helps to identify the dropouts and students who need special attention, and allows the teacher to provide appropriate advising/counseling.
A representative study came to the conclusion that more than 63% of German voters would have like to cast their vote for the federal election in 2021 online. In this paper, we aimed to investigate why Germans might be in favour or against online voting, conducting a online survey. We furthermore aimed to study the reactions of people being in favor of online voting if confronted with a verifiable remote voting system, as well as with interventions aimed at communicating that it is important to follow all the steps to verify. Our findings show that the majority of our participants were generally willing to vote online. Convenience emerged as the most popular reason for voting online. The reaction to the verifiable remote voting system was diverse, from our participants being irritated from the complexity, to very positive reactions due to high security level. Nonetheless, the majority of the participants did not change their willingness to vote online after seeing the proposed system. The different interventions had no effect. Furthermore, the majority agreed on the importance of verifiability being in place.Keywordsuser studyverifiable Internet votingtrust
This chapter overviews Estonia’s experiences in developing a highly digital society and discusses how this knowledge is transferable to other contexts in other countries. The cultural and social context that made Estonia’s digital transformation possible included the population’s high education level and good digital skills, the simplicity and clarity of ICT legislation, and public discourse that supported the adoption of ICT technologies in several public services for citizens. Nevertheless, the development of digital public services has led to digitalization, which also increases the digital divide in society. Estonia’s digital transformation practices cannot be transferred unanimously to other sociocultural contexts. Nevertheless, open collaboration between sectors, creating and nurturing a unifying narrative, and strategy-driven development is the key to successful digital transformation.
Election in any country is a very inflated and time-consuming process. Many countries have tried to implement numerous ways of conducting an election to make the process easy and time and cost-efficient. One such method is a blockchain-based secure voting system that has been in research for a long. However, it is still a challenge to implement it because of various reasons like security, availability of the internet to everyone, etc. Therefore, this paper presents a secure election system based on blockchain, tackling a few of these problems and making it more secure. The proposed system uses Aadhaar-based voter authentication and One-time password (OTP)-based verification. The proposed scheme uses a private blockchain that solves many security problems and even attacks like Sybil, Distributed denial of service (DDoS), etc. The proposed system can reduce the cost of elections to a large extent while increasing election security.KeywordsElectionBlockchainAadhaarAuthentication
Full-text available
The Routledge Handbook of Local Elections and Voting in Europe represents the standard reference text and practical resource for everybody who analyzes issues such as local electoral systems, voting behavior, or political representation in Europe. It provides comprehensive and expert coverage of 40 European countries – organized along the respective local state traditions – and in addressing a wide range of important questions related to local elections and voting, it broadens the scope of existing analyses quantitatively as well as qualitatively. Finally, it affords a more theoretically grounded typology of local elections and voting. Each country chapter is written by a leading expert and follows a rigorous conceptual framework for cross-national comparisons, providing an overview of the local government system, details on the place of local elections within the multilevel political system, specific features of the electoral system, analysis of the main electoral outcomes in recent decades, and, finally, reflective discussion. Representative democracy is as widespread at the local as at the national level, and as the significance of local authorities in Europe has increased in recent decades, local elections represent a crucial area of study. The Routledge Handbook of Local Elections and Voting in Europe is an authoritative and essential reference text for scholars and students interested in local electoral politics and, more broadly, European studies, public administration, and political science.
Full-text available
This volume contains papers presented at E-Vote-ID 2022, the Seventh International Joint Conference on Electronic Voting, held during October 4–7, 2022. This was the first in-person conference following the COVID-19 pandemic, and, as such, it was a very special event for the community since we returned to the traditional venue in Bregenz, Austria. The E-Vote-ID conference resulted from merging EVOTE and Vote-ID, and 18 years have now elapsed since the first EVOTE conference in Austria.
Logistic regression estimates do not behave like linear regression estimates in one important respect: They are affected by omitted variables, even when these variables are unrelated to the independent variables in the model. This fact has important implications that have gone largely unnoticed by sociologists. Importantly, we cannot straightforwardly interpret log-odds ratios or odds ratios as effect measures, because they also reflect the degree of unobserved heterogeneity in the model. In addition, we cannot compare log-odds ratios or odds ratios for similar models across groups, samples, or time points, or across models with different independent variables in a sample. This article discusses these problems and possible ways of overcoming them.
The Deepening Divide: Inequality in the Information Society explains why the digital divide is still widening and, in advanced high-tech societies, deepening. Taken from an international perspective, the book offers full coverage of the literature and research and a theoretical framework from which to analyze and approach the issue. Where most books on the digital divide only describe and analyze the issue, Jan van Dijk presents 26 policy perspectives and instruments designed to close the divide itself.
From the Publisher: Digital Divide examines access and use of the Internet in 179 nations world-wide. A global divide is evident between industrialized and developing societies. A social divide is apparent between rich and poor within each nation. Within the online community, evidence for a democratic divide is emerging between those who do and do not use Internet resources to engage and participate in public life. Part I outlines the theoretical debate between cyber-optimists who see the Internet as the great leveler. Part II examines the virtual political system and the way that representative institutions have responded to new opportunities on the Internet. Part III analyzes how the public has responded to these opportunities in Europe and the United States and develops the civic engagement model to explain patterns of participation via the Internet.
Recent years have seen increasing interest in internet voting in theory and practice. Proponents hope that modernizing the electoral process will boost turnout. Less optimistic scholars object that the new technology merely perpetuates existing patterns of participation. This study aims to arbitrate the controversy. New survey data from the 2007 general election in Estonia allow us to predict the usage of e-voting and its impact on electoral participation. We find that e-voting mostly affects ‘peripheral’ citizens (in a demographic and political sense), but only few of these citizens vote online in the first place. Conversely, the impact on typical e-voters is low. This bottleneck effect explains why e-voting has failed to boost turnout but also points to a role in reducing political inequality.