Electoral Studies 68 (2020) 102219
Available online 30 September 2020
0261-3794/© 2020 Elsevier Ltd. All rights reserved.
Rejecting representation? Party systems and popular support for
referendums in Europe
Brett R. Bessen
Department of Political Science, University of Colorado Boulder, UCB 333, Boulder, CO 80309-0333, USA
A R T I C L E I N F O
Keywords:
Direct democracy
External efficacy
Party systems
Political trust
Popular vote processes
Process preferences
Referendums
A B S T R A C T
Under what conditions do citizens favor deciding political issues by popular vote? Models of support for popular
vote processes usually consider the influence of individual attitudes such as political trust and interest in politics.
But much less is known about the effect of institutional variables on support for popular vote processes. This
article builds on research showing that disaffection with elected officials shapes support for referendums by
considering the influence of the party system. First, an analysis of multilevel data from twenty-four European
democracies indicates that individuals are more supportive of referendums in countries with fewer effective
political parties. Second, a mediation analysis provides evidence that the number of parties influences referendum
support through individual-level political trust and external efficacy. Where there are fewer viable parties,
feelings that elected officials are unresponsive tend to increase popular support for referendums. These findings
suggest a trade-off between available representation by political parties and support for direct influence over
public policy.
Under what conditions do citizens support the use of referendums?
Citizens’ views of popular vote processes—a term that includes referendums
as well as other processes that allow citizens to vote directly on
policy issues (el-Wakil and McKay, 2019, 1)—are usually attributed to
more foundational attitudes about politics. For instance, interest in
politics, trust in elected officials, and populist attitudes have been shown
to shape support for processes that increase popular control over policy
(e.g., Bowler et al., 2007; Bengtsson; Mattila, 2009; Mohrenberg et al.,
2019). Yet referendums occur alongside parties, legislative elections,
and other institutions that—to varying degrees—encourage responsive
policy-making by elected officials. As such, differences in countries’
representative institutions are likely to shape popular demand for the
use of referendums. Here, I extend research on attitudes associated with
referendum support by considering the contribution of institutional
variables. Specifically, I argue that party systems shape popular support
for referendums through citizens’ perceptions of system responsiveness.
Referendum support is an important attitude about the process by
which voters’ preferences are incorporated into public policy. While
many democratic theorists are wary of processes seen to promote “direct
democracy” (Madison, 1787; Gutmann and Thompson, 2009), others
argue that, when used alongside other representative institutions,
referendums will increase popular engagement in politics (Smith and
Tolbert, 2009) and provide a check on self-interested elites (Matsusaka,
2005, p. 204). Additionally, most citizens in Western democracies express
relatively favorable views of referendums (Bowler et al., 2007;
Dalton et al., 2001). This paper contends that popular attitudes about
referendum use are responsive to differences in countries’ representative
political institutions.
I argue that high levels of referendum support follow from limited
representation available to voters in countries with fewer viable political
parties. With fewer parties, voters are less likely to identify parties that
capture their full range of views and are therefore prone to view elected
officials as unresponsive. Further, in more majoritarian systems, limited
representation for those who vote for electorally unsuccessful parties
increases perceptions that authorities are unresponsive or untrustworthy.
In turn, views that authorities are unresponsive lead citizens to
support referendums as a perceived remedy to unresponsive government.
Citizens respond to limited party representation by supporting the
use of referendums as a way to constrain elected representatives or
directly influence public policy.
This article contributes to the literature examining citizens’ process
preferences by providing evidence that these preferences are shaped by
E-mail address: brett.bessen@colorado.edu.
Contents lists available at ScienceDirect
Electoral Studies
journal homepage: http://www.elsevier.com/locate/electstud
https://doi.org/10.1016/j.electstud.2020.102219
Received 6 February 2020; Received in revised form 30 August 2020; Accepted 8 September 2020
Electoral Studies 68 (2020) 102219
2
the political context. Most research on support for popular vote processes
emphasizes the attitudes that lead voters to prefer popular control
over public policy.1
As such, we know little about how, if at all, institutional
conditions shape support for popular vote processes. I build on
research showing that referendum support is associated with feelings
that government is unresponsive (e.g., Schuck and de Vreese, 2015;
Bengtsson; Mattila, 2009; Coff´e; Michels, 2014) by linking these feelings
with variation in the party system. Namely, I argue that cross-national
variation in the number of political parties shapes referendum support
through individuals’ attitudes toward their political representatives. The
analysis indicates that citizens are sensitive to the availability of representation
by political parties; they respond to limited party representation
with increased support for referendums.
In the following section, I conceptualize referendum support and
discuss prior research on support for popular vote processes. Next, I
extend this research by developing the argument that decreases in the
number of parties contribute to referendum support through citizens’
perceptions of system responsiveness. I then test this proposition using
data from twenty-four European democracies. In combination, hierarchical
linear models and a causal mediation analysis indicate that the
number of parties influences referendum support through political trust
and external efficacy. As discussed below, external efficacy involves
feelings about government’s responsiveness to people’s interests. The
analysis indicates that party systems shape citizens’ attitudes about the
role of “the people” in political decision-making. Where party representation
is limited, citizens are more likely to prefer processes
perceived to create popular control over public policy.
1. Direct democracy and referendums
Direct democracy refers to a model of democratic practice where
citizens play an immediate role in deciding political issues. It is usually
understood in contrast to representative democracy, where decisions are
made by elected officials who are supposed to embody or interpret citizens’
interests (Held, 2006). In contemporary democracies, however,
elements of direct democracy co-exist with representative institutions
including elections, political parties, and legislatures. el-Wakil and
McKay (2019) therefore recommend a “democratic systems approach”
that recognizes that popular votes processes—i.e., initiatives and referendums—interact
with and potentially complement other democratic
institutions. In line with this premise, I examine how the party system
shapes citizen support for referendums.
Referendums are a type of popular vote process that allow citizens to
vote on policies that have already been proposed or approved by elected
officials. Accordingly, referendums provide citizens with “a device of
control and issue-based accountability” over the legislature or executive
(Cheneval and el Wakil, 2018, 295). Put differently, referendums
introduce majority opinion as a potential veto player in democratic
systems (Hug and Tsebelis, 2002). Successful referendums can alter the
status quo by overturning or otherwise altering existing policies. Other
properties of referendums vary. Among other differences, they can be
initiated by citizens (from the bottom up) or by political elites (from the
top down) and their outcomes can be non-binding or legally binding.2
2. Explaining referendum support
Citizens can be said to support referendums if they favor referendums
as decision-making processes for important political issues. As such,
referendum support is a preference about the processes that governments
use to decide on policy.3
Most citizens in Western democracies approve
of the occasional use of referendums and other popular vote processes
(Dalton et al., 2001; Bowler et al., 2007). However, there is still significant
individual-level and cross-national variation in this preference.
Previous studies have focused on explaining why some individuals are
highly supportive of referendum use while others are not.
The framework created by Dalton et al. (2001) parses between
participatory and disaffected attitudes as drivers of support for popular
vote processes. On one hand, referendum support might be explained by
cognitive mobilization within the electorate. The cognitive mobilization
hypothesis predicts that referendum support comes from politically
engaged citizens who value more direct participation in political
decision-making. On the other, support for referendums may stem from
dissatisfaction with the conventional institutions of representative democracy.
The dissatisfaction hypothesis predicts that citizens who are
distrustful of political elites or dissatisfied with institutional outputs will
support referendums. According to this logic, dissatisfied citizens favor
popular vote processes because they seem to allow voters to steer or
bypass the legislature.
The cognitive mobilization hypothesis rests on the idea that engaged
citizens favor processes that allow them participate in political decisions.
In theory, economic development and widespread education
promote participatory values that contribute to support for referendums
(Dalton et al., 2001, 146–47). Several studies find that interest in
politics—used as an indicator of political engagement—is associated
with referendum support (Bowler and Donovan, 2019; Donovan and
Karp, 2006; Bowler et al., 2007). Relatedly, Schuck and de Vreese
(2015) find that a sense of civic duty is associated with support for
holding referendums on E.U. membership. But other studies suggest that
political engagement is less important for referendum support. Some
studies find weak or negative relationships between political interest
and referendum support (Werner, 2020; Bengtsson; Mattila, 2009).
Recent work also suggests that, rather than participatory values, more
instrumental concerns over policy outcomes shape attitudes around
referendums (Werner, 2020; Brummel, 2020). In short, there is mixed
evidence for the idea that political engagement shapes referendum
support.
Political dissatisfaction is more consistently associated with referendum
support. Feelings that elected officials are untrustworthy or unresponsive
can motivate support for processes that appear to place
decision-making power in the hands of voters (Dalton, 2004, 184).
Empirically, low levels of external efficacy have been shown to shape
support for referendums. Low external efficacy indicates perceptions
that “government institutions and authorities” are insufficiently
responsive “to citizen demands” (Craig et al., 1990, 290). Separate
Dutch and Finnish samples indicate that low external efficacy is associated
with support for popular vote processes (Bengtsson and Mattila,
2009; Coff´e; Michels, 2014). Similarly, Schuck and de Vreese (2015)
find that political cynicism—an attitude that involves a “gap between
voters and their political representatives” (pg. 151)—is related to support
for using referendums to decide on E.U. integration.
Low levels of political trust also correspond with support for popular
vote processes. While related to the concept of external efficacy, political
trust is a somewhat broader concept that involves feelings about political
authorities’ capacity and commitment to act in voters’ interests
1
Among others, see Dalton et al. (2001); Donovan et al. (2009); Coff´e and
Michels (2014); Bowler et al. (2007); Font et al. (2015); Bengtsson and Mattila
(2009). For exceptions see Schuck and de Vreese (2015) and Bowler and
Donovan (2019).
2
For a discussion of these properties see Cheneval and el Wakil (2018).
Altman (2018) refers to elite-initiated popular vote processes as plebiscites,
while Qvortrup (2018) refers to them as ad hoc referendums. I use the broader
term, referendums, to refer to both citizen and elite-initiated processes that
allow citizens to vote on existing legislation.
3
See Hibbing and Theiss-Morse (2001) on process preferences. Brummel
(2020) uses the phrase “referendum support” while Bowler and Donovan
(2019) refer to this variable as “support for referendums.” I use the two phrases
interchangeably.
B.R. Bessen
Electoral Studies 68 (2020) 102219
3
(Craig et al., 1990). Several analyses indicate that low levels of political
trust are associated with increased support for referendum use (Bowler
and Donovan, 2019; Werner, 2020; Dalton, 2004). Distrust in political
parties, in particular, is strongly related to referendum support (Dalton,
2004, 183). This is likely because referendums are seen to provide a
popular check on parties in the legislature. In a similar vein, Bauer and
Fatke (2014) show that frequent popular vote processes are associated
with lower levels of political trust in Swiss cantons. When officials are
seen as untrustworthy, voters seem to favor popular vote processes as a
ways to constrain their representatives or more directly influence policy.
3. Party systems and referendum support
I argue that referendum support reflects differences in countries’
representative political institutions. Namely, I expect that referendum
support varies with the opportunities for representation provided by
political parties: In countries with fewer successful parties, individuals
should, on average, be more supportive of referendums. This argument
follows from the idea that voter representation suffers as the number of
relevant parties decreases. First, fewer parties contribute to perceptions
that the legislature is untrustworthy (low political trust) or unresponsive
to citizens’ interests (low external efficacy). Second, perceptions that
government is untrustworthy or unresponsive increase referendum
support. Decreases in the number of relevant parties thereby contribute
to referendum support through political trust and external efficacy. This
causal sequence is illustrated in Fig. 1.4
In systems with fewer parties, misalignment between voters and
parties in government decreases political trust and external efficacy
(arrow A in Fig. 1). Primarily, voters are less likely to identify parties
that correspond with their full range of policy views in countries where
there are fewer parties.5
Decreases in the number of parties therefore
reduce congruence between citizens’ views and party platforms (Weβels,
1999, 2007; Dalton, 1985). Secondly, countries with fewer parties are
more likely to be governed by single-party majority cabinets (Taagepera,
2002; Lijphart, 2012). The number of parties is logically connected to
cabinet type: Single-party majority cabinets are “always possible” when
there are fewer than two legislative parties and “always impossible”
when there are more than four legislative parties (Taagepera, 2003, 5).
Systems with fewer parties tend to produce cabinets that exclude minority
parties and therefore provide poor representation to individuals
who support smaller parties.
In turn, Europeans’ feelings toward elected officials reflect the
availability of party representation. Representatives are regarded as less
trustworthy (Marien, 2011; Miller and Listhaug, 1990) and less
responsive (Karp and Banducci, 2008) in countries with fewer relevant
parties. External efficacy and political trust are related attitudes that tap
into the more general concept of system responsiveness (Miller and
Listhaug, 1990, 359). As noted above, external efficacy refers to “beliefs
about the responsiveness of governmental authorities and institutions to
citizen demands” (Craig et al., 1990, 290). In comparison, political trust
involves judgments about government actors’ commitment and capacity
to act in voters’ interests. People tend to trust institutions that appear
faithful to their interests and capable of enacting them (Levi and Stoker,
2000). These two attitudes—i.e., political trust and external efficacy—mediate
the effect of the party system on referendum support.
Through these attitudes, lower numbers of parties lead citizens to
support referendums as a way to more directly influence policy (arrow B
in Fig. 1). Why do political distrust and low external efficacy increase
referendum support? Feelings that elected officials are unresponsive
create support for processes perceived to take decisions out of the hands
of politicians and create popular control over policy (Dalton, 2004,
181–184). First, voters likely consider referendums a way to bypass poor
representation and alter the legal status quo. Citizen-initiated referendums,
in particular, allow voters to challenge existing laws that no
longer conform with their policy preferences (see Cheneval and el Wakil,
2018). Second, voters may understand referendums as a way to
constrain their representatives. Once voters signal their preferences in
referendums, it becomes costly for party elites to ignore their policy
preferences (Hug and Tsebelis, 2002; Matsusaka, 2005). Even
non-binding referendums can considerably limit policy-makers who find
it untenable to oppose decisions made by popular vote. Voters can use
referendums to signal their dissatisfaction with existing policies and
thereby check officials’ legislative power. This makes referendums
attractive to individuals who find their representatives untrustworthy or
unresponsive.
A related argument for the influence of the party system concerns
electoral losers. In systems with fewer parties, referendums are
appealing to electoral losers who face limited representation in government
coalitions. Those who vote for electorally unsuccessful parties—i.e.,
electoral losers—receive comparatively poor representation in
systems with fewer parties (Anderson and Guillory, 1997; Lijphart,
2012). More proportional systems support more parties and, consequently,
give political minorities greater say in the government
decision-making. Electoral losers in majoritarian systems are therefore
“less satisfied with the way democracy works” than their counterparts in
more proportional systems (Anderson and Guillory, 1997, 68). Several
studies suggest that support for out-of-government parties increases
support for referendum use (Karp and Banducci, 2008; Smith et al.,
2010). Along these lines, Bernauer and Vatter (2012) find that electoral
losers are more satisfied with democracy in more proportional systems
and in countries that employ popular vote processes. In countries with
fewer legislative parties, electoral losers likely see referendums as a way
to influence policy and offset limited representation in government.
Finally, note that the argument focuses on the number of relevant
parties rather than the broader electoral system. While components of
the electoral system, most importantly district magnitude, shape the
number of relevant parties (Shugart and Taagepera, 2017; Cox, 1997),
the number of parties most directly reflects citizens’ options for political
representation. The effective number of legislative parties is particularly
critical since party members can only provide representation if they are
elected to parliament. Relatedly, greater numbers of legislative parties
increase the likelihood of multi-party cabinets (Taagepera, 2002, 2003).
There is also significant variation in the number of parties within systems
under PR (proportional representation) (Lijphart, 2012). This
variation is partly a product of the electoral system but also reflects
other factors such as the number of social cleavages at key historical
junctures (B´ertoa, 2014; Lipset and Rokkan, 1967). Increases in the
number of viable parties make PR systems relatively more proportional.
Even across PR systems, decreases in the number of relevant parties
should lower political trust and external efficacy and, consequently,
increase referendum support.
4. Data and methods
The analysis uses survey data from Wave 6 of the European Social
Fig. 1. Proposed pathway linking the number of parties with referendum
support.
4
The argument allows for the possibility that—alongside external efficacy
and political trust—other attitudes might link the number of parties with referendum
support. I do not assume that the effect of the number of parties is
entirely mediated by efficacy and political trust.
5
See Dalton (1985): “… as the number and variety of parties increases, it
becomes easier for voters to locate a party that simultaneously represents their
views on all issue dimensions” (287).
B.R. Bessen
Electoral Studies 68 (2020) 102219
4
Survey (2012) alongside different country-level variables. Wave 6 is
particularly useful since its democracy module asks respondents for
their opinions about various democratic procedures, including referendums.
In addition, there were relatively few high-profile referendums in
2012.6
This is conducive to the current study because, in the short-term,
referendum support is likely influenced by the results of symbolically
important referendums. For instance, the United Kingdom’s 2016 Brexit
referendum likely shaped referendum support in the U.K. and, possibly,
elsewhere in the European Union.7
The absence of high-profile referendums
makes this year well-suited to study the influence of the party
system. The analysis includes individuals located in 24 European
democracies.8
4.1. Dependent variable: referendum support
Referendum support is indicated by a single item measure that
ranges from 0 to 10. This survey question asks respondents: “How
important do you think it is for democracy in general that citizens have the
final say on the most important political issues by voting on them directly in
referendums.” Possible responses range from “not at all important” to
“extremely important,” with larger values indicating greater importance.
The survey item asks about referendums; a common type of popular
vote process that allows citizens to vote on policies that have been
proposed or approved by elected officials. It further specifies that citizens
should “have the final say” and therefore suggests a preference for
deciding important political issues through binding popular votes.
Fig. 2 presents country means of referendum support sorted from
lowest (in the Netherlands, 7.21) to highest (in Cyprus, 9.00). The crosscountry
mean is 8.27. Consistent with prior research, most people say
they support referendum use (Donovan and Karp, 2006; Bowler et al.,
2007). The relative position of each country also roughly corresponds
with other research on referendum support. In Switzerland, where
popular vote processes are highly institutionalized, referendum support
is comparatively high. Likewise, Spanish samples generally show high
levels of referendum support (Bowler et al., 2007, 352). This consistency
supports the validity of the measure and suggests that country-level
variables shape referendum support.
While referendum support is high on average, it is significantly more
concentrated in some countries than others. Figure S1 in the supplemental
materials shows the distribution of referendum support in each
country included in the analysis. In some countries, most individuals
strongly endorse referendum use. For example, in Cyprus 60% of respondents
indicate the highest level of referendum support—i.e., that
referendums are “extremely important for democracy”—and 84% of
individuals fall within the highest three values. Referendum support is
similarly concentrated in Spain, Poland, and Bulgaria. In comparison,
other countries have a larger base of moderate support for referendum
use. Only 20% of Belgians, 30% of Portuguese, and 13% of Dutch answer
that having the “final say” via referendums is “extremely important for
democracy.” More moderate referendum support indicates that referendums
are seen as less essential for democratic decision-making in
these countries.
4.2. Independent variables
4.2.1. Country level
The primary country-level variable is the effective number of legislative
parties. The effective number of legislative parties is a “size-weighted
count” of the parties that obtain seats in the legislature (Laakso and
Taagepera, 1979; Shugart and Taagepera, 2017, 64).9
This is a standard
measure and is particularly appropriate here, since parties with
comparatively few seats are less capable of translating party-voters’
preferences into policy. Those who vote for parties that receive fewer
seats are likely to view authorities as unresponsive and to support referendum
use. The effective number of legislative parties matters because
it affects voters’ opportunities for representation in the national
assembly.10
The models also include a dummy variable indicating majoritarian
electoral systems. This allows us to examine the influence of the number
of parties while holding PR versus majoritarian rule constant. The
majoritarian variable is coded as one for countries with majority or
plurality electoral districts at the first electoral tier in the previous national
election. Coding was based on the National Level Party Systems
dataset, used by Shugart and Taagepera (2017).11
The majoritarian
systems in the sample include; Britain (first-past-the-post), France
(majority-plurality), Bulgaria, Hungary, and Lithuania (mixed-member
majoritarian).12
The remaining countries in the sample use some form of
PR.
The frequency that countries employ popular vote processes might
also shape support for referendums. While Schuck and de Vreese (2015)
find no effect of previous referendum use, Bowler and Donovan (2019)
find that more frequent referendums are associated with referendum
support. In theory, frequent usage creates familiarity with popular vote
processes and could bolster support for referendums. The analysis includes
a count of the number of popular votes held in each country between
2002 and 2012. This variable is from the Varieties of Democracy dataset
and does not distinguish between different popular vote processes (e.g.,
Fig. 2. Mean popular support for referendums in 24 countries.
6
Five countries in the sample had some type popular vote processes in 2012:
Iceland, Ireland, Lithuania, Slovenia, and Switzerland.
7
See Brummel (2020) and Marien and Kern (2018) on the effect of losing in
referendums.
8
Table S3 in the supplemental materials presents the list of countries.
9
The effective number of legislative parties “is the inverse sum of squared
fractional [seat] shares:This measure is from the Comparative Political Dataset
Set (see Armingeon et al. 2017).“ Ns =
∑
1/(si)2
, where s is the seat share of
party i (Shugart and Taagepera, 2017, 64).
10
I show a model with the effective number of electoral parties in Table S10 in
the supplemental materials. This measure captures the number of parties
weighted by the proportion of votes received by each party (Shugart and Taagepera,
2017; Laakso and Taagepera, 1979).
11
See Struthers et al 2018.
12
Bulgaria reformed its electoral system in 2011, scrapping single-member
districts. But as of 2012 representation was based on the 2009 elections
which were conducted using a mixed-member majoritarian system (Hardman,
2011). Under mixed-member majoritarianism, the upper tier only partially
compensates for single-member districts at the first tier, making these systems
closer to majoritarianism than PR (Shugart and Taagepera, 2017, 59).
B.R. Bessen
Electoral Studies 68 (2020) 102219
5
citizen initiatives or referendums). To account for the diminishing
impact of each additional referendum, this variable is logged in the main
analysis. Logging the variable also reduces the influence of outliers, such
as Switzerland, where popular votes are uncommonly frequent.13
Half
the countries in the sample held at least one popular vote during this ten
year period.14
Finally, the analysis accounts for the size of the population and
democratic quality in each country. The population measure indicates
the total population (logged) during 2012.15
Small populations may
increase support for referendums. Given a smaller population, people
may perceive their votes in referendums or initiatives as more impactful
(c.f. Ladner, 2002, 826). Democratic quality is measured using the
electoral democracy index from the Varieties of Democracy dataset,
where larger values indicate higher quality democratic institutions.16
All else equal, high quality democratic institutions allow citizens to hold
politicians accountable and may reduce demand for referendums.
4.2.2. Individual level
At the individual level, the primary independent variables are
external efficacy and political trust. The political trust measure is an additive
index of three items: trust in politicians, trust in political parties,
and trust in parliament (α = 0.92) and is scaled from (0) “no trust at all”
to (10) “complete trust.” The index captures confidence in the actors
most responsible for political representation. External efficacy is indicated
by an item that asks respondents ‘‘how often do you think the
government in [country] today changes its planned policies in response
to what most people think?” The measure is scaled from 0 to 10 with
highest value indicating that the government “always” responds to
people’s preferences.
Alongside these variables, I include a measure indicating the importance
of living in democracy, an indicator of interest in politics, and a party
alternatives item which asks respondents whether political parties offer
“clear alternatives to one another.” The party alternatives item relates to
the opportunities voters have for substantive representation by political
parties. The importance of democracy item helps parse between disaffected
and pro-system attitudes associated with referendum support. As
with civic duty with respect to voting (Schuck and de Vreese, 2015), I
expect that referendum support will increase with beliefs in the
importance of democracy.17
The political interest indicator ranges from
(1) “not at all interested in politics” to (4) “very interested in politics.”
Existing research suggests that political interest is associated with support
for referendums (Schuck and de Vreese, 2015; Donovan and Karp,
2006).
Finally, I include control variables indicated by the existing literature.
These include gender, age, political ideology (left-right), years of
education, and household income. Increases in income and education
are usually associated with decreased support for referendums (Bowler
et al., 2007; Coff´e and Michels, 2014). Findings are mixed with respect
to ideology, though there is some indication that liberalism corresponds
with approval of referendums (Bengtsson and Mattila, 2009). Interestingly,
a number of studies indicate that women are more likely to favor
referendums (Donovan and Karp, 2006; Bengtsson; Mattila, 2009;
Bowler et al., 2007). One possibility is that women are more supportive
of referendums in response to their comparatively poor descriptive
representation in parliament.
4.3. Empirical strategy
I have argued that referendum support is shaped by the number of
political parties and, in turn, by individuals’ feelings about government
responsiveness. First, I graphically present the relationships between the
key independent variables and referendum support. These plots provide
a sense of the data and indicate countries that deviate from the expected
relationship. Second, I analyze the effects of country-level and
individual-level variables with a series of hierarchical linear models. The
analysis employs random intercept models with two levels; i individuals
nested within j countries. Standard errors are adjusted for the number of
observations at the country-level, permitting appropriate significance
testing of the country-level variables (Steenbergen and Jones, 2002;
Krull and MacKinnon, 1999). Third, I incorporate these models into a
mediation analysis in order to evaluate external efficacy and political
trust as mediators of the party system.
5. Results
Simple scatter plots indicate that fewer parties, low trust, and low
external efficacy are associated with increased referendum support.
First, Fig. 3 plots the effective number of legislative parties against
average support for referendums in each country. As expected, referendum
support decreases as the effective number of parties increases (n
= 24, r = − 0.52). A number of country cases are noteworthy here.
Belgium (BE), for example, falls close to the trend line but has a
remarkably high number of parties (8.43). Cyprus (CY) has a moderate
number of parties (3.60) but is uncharacteristically high in referendum
support (9.00). The Netherlands (NL) has a relatively large number of
parties (5.72) but is uncommonly low in support for referendums. It is
worth noting that the Netherlands has a highly permissive electoral
system with nationwide proportional representation and a large number
of parties for a country of it size (Shugart and Taagepera, 2017, 33).
Overall, Fig. 3 clearly illustrates the broader relationship: fewer effective
legislative parties are associated with increased support for
referendums.
The number of parties likely shapes referendum support through
perceptions of system responsiveness. Fig. 4 presents the relationships
between country means of political trust, external efficacy, and average
referendum support in two scatter plots. The left-hand panel plots political
trust against average referendum support in each country (n = 24,
r = − 0.36). The right-hand panel shows the relationship between
Fig. 3. Number of legislative parties and mean referendum support in
24 countries.
13
Switzerland held 81 popular votes between 2002 and 2012.
14
While referendum support is generally high, popular vote processes are
relatively infrequent in most countries in the sample. The number of referendums
in the previous ten years and individual-level referendum support are
only weakly correlated (r = 0 : 04).
15
The population measure is from the Comparative Political Data Set.
16
The electoral democracy index includes components capturing freedom of
association and expression, clean elections, and the extent of suffrage (see
Coppedge et al., 2017, 49).
17
The importance of democracy item may resemble the dependent variable. In
fact, the items are only moderately correlated (r = 0.198). Different model
specifications also show that the results are robust to the exclusion of this item.
B.R. Bessen
Electoral Studies 68 (2020) 102219
6
external efficacy and the same dependent variable (n = 24, r = − .19).
Since political trust and external efficacy both capture views of system
responsiveness, the relative position of each country roughly corresponds
across the two panels. While there is considerable spread around
the line of best fit, the trend indicates that as political trust decreases,
referendum support increases. Switzerland (CH) and Denmark (DK)
present notable exceptions: they are uncommonly high in trust and
support for referendums. The right-hand panel illustrates a similar trend
with respect to external efficacy: increases in external efficacy correspond
with decreases in referendum support.
5.1. Hierarchical models
Table 1 presents five random intercept models that explain referendum
support as a function of country and individual-level variables. All
five models include the same country-level variables: the effective
number of legislative parties, the number of popular votes held between
2002 and 2012 (logged), an index of democratic quality, and total
population (logged). Model 1 only includes the country-level variables.
Model 2 includes the individual-level control variables but excludes the
proposed mediators, i.e., political trust and external efficacy. The proposed
mediation process requires that the estimated effect of number of
parties is diminished when political trust and external efficacy are added
to the model (Krull and MacKinnon, 1999; Baron and Kenny, 1986).
Models 3 and 4 iteratively add political trust and external efficacy.
Finally, Model 5 incorporates the full set of variables alongside the
proposed mediators.
5.2. Results: country-level
5.2.1. Number of parties
Lower numbers of legislative parties are associated with increased
referendum support. The coefficient on the effective number of legislative
parties is negatively signed across the models and reaches statistical
significance in Models 1 and 2 (p < 0.05). The diminished effect of the
number of parties after the introduction of political trust (Model 3) and
external efficacy (Model 4) is consistent with the mediating effect of
these variables (see Baron and Kenny, 1986; Krull and MacKinnon,
1999). Based on Model 1, which omits the individual-level variables, a
one-SD (1.44 party) increase in the number of parties is associated with a
0.20 unit decrease in the country mean of referendum support. Once
individual-level controls are included, as in Model 2, an identical increase
is associated with a 0.17 unit decrease in referendum support.
Fig. 5 presents predicted values of referendum support as the effective
number of legislative parties increases from the minimum (1.98) to
maximum (8.43) observed values. Moving from the minimum to
maximum number of parties is associated with a 0.89 unit decrease in
referendum support. While some caution is warranted since few countries
exceed six effective legislative parties, Fig. 5 demonstrates the
general relationship: an increase in number of legislative parties is
associated with a decrease in referendum support.18
This general relationship is robust to changes in the sample and
model specification. To address the concern that outlying countries bias
the results, I regressed support for direct democracy on the country-level
variables while iteratively omitting each country-case from the model.
This analysis indicates that no particular country-case drives the relationship
between fewer parties and increased referendum support.19
The
estimate associated with the number of parties is also robust to variety of
alternative model specifications. In the supplemental materials, I
include alternative measures of democratic quality and popular vote
frequency, and add controls for federalism, unemployment, and economic
growth. Increases in the number of legislative parties are significantly
associated with decreased referendum support in each model
specification.20
5.2.2. Other country-level variables
First, referendum support is not significantly associated with PR or
Fig. 4. Country means of political trust, external efficacy, and referendum support.
18
An increase in the number of electoral parties also corresponds with a
decrease in referendum support. See Table S10 in the supplemental materials.
19
Table S7 in the supplemental materials shows the results of this analysis
omitting three outliers: The Netherlands, Switzerland, and Belgium. As shown
here, p = 0.07 when Belgium is omitted from the model. In all other models, the
coefficient on the effective number of parties is statistically significant at p <
0.01. The full analysis includes 24 identical models, each of which excludes one
country-case.
20
These models are presented in Tables S8 and S10 in the supplemental
materials.
B.R. Bessen
Electoral Studies 68 (2020) 102219
7
Table 1
Models of support for referendums in 24 countries. Random effects specified at the country-level. Higher values of the dependent variable indicate increasing
importance of deciding policy by referendums.
(1) (2) (3) (4) (5)
Country-level
No. Parties (seats) − 0.139***
(0.054)
− 0.117**
(0.053)
− 0.072
(0.052)
− 0.080*
(0.046)
− 0.050
(0.046)
Majoritarian − 0.136
(0.206)
− 0.008
(0.203)
0.052
(0.199)
0.014
(0.177)
0.049
(0.177)
Ln Pop. Votes, 02 - 12 0.050
(0.066)
0.066
(0.065)
0.067
(0.064)
0.072
(0.057)
0.069
(0.056)
Electoral Dem. Index − 2.202
(1.363)
− 2.181
(1.346)
− 1.335
(1.321)
− 1.490
(1.172)
− 0.943
(1.171)
Ln Population − 0.012
(0.056)
0.004
(0.056)
− 0.005
(0.055)
0.009
(0.049)
0.001
(0.049)
Individual-level
Political Trust − 0.131***
(0.006)
− 0.100***
(0.007)
External Efficacy − 0.076***
(0.005)
− 0.054***
(0.005)
Political Interest − 0.027*
(0.014)
0.025*
(0.014)
0.041***
(0.015)
0.076***
(0.015)
Party Alternatives 0.061***
(0.005)
0.086***
(0.005)
0.061***
(0.005)
0.076***
(0.005)
Imp. Democracy 0.205***
(0.006)
0.216***
(0.006)
0.192***
(0.007)
0.200***
(0.007)
Ideology (LR) − 0.030***
(0.005)
− 0.021***
(0.005)
− 0.021***
(0.005)
− 0.015***
(0.005)
Education (years) − 0.024***
(0.003)
− 0.021***
(0.003)
− 0.011***
(0.003)
− 0.009***
(0.003)
Income − 0.027***
(0.004)
− 0.023***
(0.004)
− 0.021***
(0.005)
− 0.018***
(0.005)
Female 0.110***
(0.022)
0.129***
(0.022)
0.036
(0.023)
0.052**
(0.023)
Age − 0.002***
(0.001)
− 0.002***
(0.001)
− 0.002***
(0.001)
− 0.002***
(0.001)
Variance Components
Individual-level 0.112***
(0.033)
0.109***
(0.032)
0.105***
(0.031)
0.081***
(0.025)
0.081***
(0.025)
Country-level 3.927***
(0.027)
3.661***
(0.029)
3.604***
(0.029)
3.180***
(0.029)
3.149***
(0.029)
Observations 43,415 31,049 30,645 23,875 23,583
Log Likelihood − 91,344 − 64,245 − 63,172 − 47,727 − 47,025
*p<0.1; **p<0.05; ***p<0.01.
Fig. 5. Effective number of legislative parties and predicted values of referendum support. Predictions are based on the fixed portion of the model.
B.R. Bessen
Electoral Studies 68 (2020) 102219
8
majoritarian rule. Notably, increases in the number of parties still
correspond with decreases in referendum support when holding PR
versus majoritarian rule constant. Second, I find that more frequent
popular vote processes are not significantly associated with referendum
support. As literature on the Swiss experience suggests, more frequent
popular vote processes might signal dissatisfaction with political leaders
rather than enthusiasm for referendums (Bauer and Fatke, 2014).21
Third, the electoral democracy index is negatively signed but not statistically
significant. In part, this may be due to limited variation in
democratic quality in this sample. Finally, logged population is not
significantly associated with referendum support.
5.2.3. Results: individual-level
Political trust and external efficacy are associated with decreases in
referendum support across the models in Table 1. Fig. 6 presents the
individual-level coefficient estimates from Model 5. Political trust is
significantly associated with decreases in referendum support. According
to model 5, a one-SD increase on the political trust item corresponds
with a 0.22 unit decrease in referendum support. Likewise, external
efficacy is associated with decreases in support for referendums. Model 5
indicates that a one-SD increase on the external efficacy variable corresponds
with a 0.13 unit increase in referendum support. These two
variables indicate that feelings that elected officials are untrustworthy
or unresponsive increase popular support for referendums.
The importance of democracy and political interest items are associated
with increased support for referendums. The effect of the
importance of living in a democracy variable is relatively substantial.
Model 5 indicates that a one-SD increase on the importance of democracy
measure corresponds with a 0.38 unit increase in referendum
support. Political interest is associated with a more modest increase in
referendum support.22
General approval of democratic government and,
to a lesser extent, political are associated with support for deciding issues
through referendums.
Against expectations, the party alternatives variable is associated
with increased support for referendums. We would expect that views
that parties are similar, and therefore offer limited options to voters,
would be associated with referendum support. But Model 5 indicates
that a one-SD increase on the party alternatives measure corresponds
with a 0.18 unit increase in referendum support. This is probably
because the party alternatives variable captures political engagement
rather than congruence (or incongruence) between voters and parties.23
The political trust and external efficacy variables, in comparison, more
directly measure perceived congruence between citizens and their representatives
(Craig and Maggiotto, 1982; Levi and Stoker, 2000). That is
to say, the party alternatives variable likely captures pro-system attitudes
related to referendum support.
The remaining individual-level variables suggest that, in general,
those who benefit less from the status quo are more likely to favor referendums.
Decreasing income and education are associated with support
for referendums. The analysis also indicates women are slightly
more supportive of referendum use. All else equal, female respondents
score 0.05 units higher on the referendum support measure.24
Liberal
ideology is associated with modest increases in referendum support
across the models in Table 1. Finally, increasing age is associated with
modest decreases in referendum support.25
5.3. Mediation analysis
The main contention of this paper is that the party system influences
referendum support through individual-level political efficacy and political
trust.26
To evaluate this sequence, I conduct a causal mediation
analysis using the multilevel data structure described above. The causal
mediation framework builds on conventional mediation procedures (e.
g., Baron and Kenny, 1986) and is compatible with the multilevel data.27
Causal mediation analysis decomposes a proposed causal sequence into
the direct effect of a treatment variable (Ti) and the effect occurring
thorough specified mediators (Mi) on some outcome of interest (Imai
et al., 2011). Here, I evaluate the effect of a four party increase (Ti) on
support for direct democracy through the mediators external efficacy
and political trust (Mi).
Table 2 presents the models employed in the mediation analysis.
Models 6 and 7 regress the mediators, political trust and external efficacy,
on the effective number of parties and the independent variables
from the preceding analysis.28
Note that the outcome model, i.e., Model
8, tests for mediating effects of political trust while controlling for
external efficacy (and vice versa) and therefore provides a tough test of
the proposed mechanism. As shown in Table 2, the number of parties
varies with the mediating variables in the expected direction; increases
in the number of parties are associated with higher levels of political
Fig. 6. Coefficient estimates from Model 5. Higher values of the dependent
variable indicate the increasing importance of referendums.
21
Table S8 in the supplemental materials includes controls for the availability
of popular vote processes.
22
Political trust is negatively signed only when the model omits external efficacy
or political trust. Models 3–5 control for these variables and should
therefore produce more reliable estimates of the relationship between political
interest and referendum support.
23
The party alternatives item is positively correlated with political interest,
political trust, and the importance of democracy items. A simple random
intercept model indicates that the party alternatives item is not significantly
associated with the number of parties.
24
Estimate based on Model 5.
25
For similarities with previous research see: Coff´e and Michels (2014) on
education, Rose and Weβels (2020) on income, Bengtsson and Mattila (2009) on
education, Donovan and Karp (2006) on gender, and Schuck and de Vreese
(2015) on age.
26
Interactions are appropriate for moderation effects. Appropriately, crosslevel
interactions between the number of parties and the attitudinal variables
are not statistically significant.
27
Retaining the multi-level structure allows for correct calculations of the
standard errors and therefore reduces the likelihood of type-I error (Krull and
MacKinnon, 1999). See Imai et al. (2011) for a full discussion of causal mediation
analysis. I use their R package, mediation, for the following analysis
(Tingley et al., 2014).
28
These controls reduce the likelihood that the relationship between the
mediator and outcome variable is confounded by another omitted variable.
Models with the full set of control variables are presented in the appendix. The
model predicting political trust does not include external efficacy as predictor
(and vice versa).
B.R. Bessen
Electoral Studies 68 (2020) 102219
9
trust and external efficacy. Model 8 then regresses referendum support
on the number of parties, the mediating variables (external efficacy and
political trust), and the controls. Consistent with the proposed mechanism,
the estimated effect of the number of parties decreases in
magnitude once the mediators, external efficacy and political trust, are
included in the model. This provides initial evidence that the effect of
the party system is mediated by external efficacy and political trust.
I then use the method proposed by Imai et al. (2011) to estimate the
effect of an increase from two to six parties on support for direct democracy
through the mediators external efficacy and political trust. This
procedure first estimates values of the mediators, efficacy and trust, with
the effective number of parties held at values of two and six, respectively.
Second, values of the outcome variable, support for direct democracy,
are estimated using the predicted values of efficacy and trust
obtained from the previous set of models. The average causal mediation
effect (ACME), i.e., the effect of the party system mediated by efficacy
and trust, is calculated as the average difference between estimates of
support for direct democracy using the values of the mediators from the
two and six-party conditions. Thus the ACME associated with political
trust (or external efficacy) captures the average difference in support for
direct democracy occurring through political trust (or external efficacy)
when the number of parties increases from two to six. Finally, confidence
intervals around the ACME and other of interest quantities are
estimated by quasi-Bayesian Monte Carlo using 1000 simulations (King
et al., 2000; Imai et al., 2010).
This analysis indicates that political trust and, to a lesser extent,
external efficacy mediate the effect of the available representation by
political parties. Fig. 7 graphically presents the results of the mediation
analysis. The average causal mediation effect (ACME) indicates the
Table 2
Mediation Models. Control variables presented in the appendix.
(6) (7) (8)
Dependent variable:
Political Trust External Efficacy Referendum Support
No. Parties 0.377***
(0.103)
0.249***
(0.084)
− 0.050
(0.046)
Political Trust − 0.100***
(0.007)
External Efficacy − 0.054***
(0.005)
Observations 23,583 23,583 23,583
Log Likelihood − 47,748 − 52,703 − 47,025
*p<0.1; **p<0.05; ***p<0.01.
Fig. 7. Results of the mediation analysis. Bars indicate 95% confidence intervals around the average causal mediation effect (ACME), average direct effect (ADE),
and total effect.
B.R. Bessen
Electoral Studies 68 (2020) 102219
10
average effect of the number of parties occurring through external efficacy
or political trust, while the average direct effect (ADE) represents
the unmediated effect of a four unit increase in the number of parties on
referendum support. The total effect captures both the direct effect of the
party system and the effect mediated by external efficacy or political
trust. The total effect is therefore equal to the sum of the ACME and ADE.
As illustrated in the left-hand panel of Fig. 7, the mediation effect
(ACME) associated with political trust is negative and statistically
different from zero while the direct effect is not. This suggests that increases
in the number of parties attenuate referendum support through
the mediating variable, political trust. The ACME associated with political
trust is − 0.15 while the total effect is − 0.35, indicating that political
trust mediates about 41% of the effect of the number of parties on
referendum support. As shown in the right-hand panel, the ACME
associated with external efficacy is also statistically different from zero
but more modest in magnitude. Holding political trust constant, external
efficacy is estimated to mediate about 18% of the effect associated with
the number of parties. Meanwhile the direct effect of the party system is
not statistically different from zero. This indicates that the effect of the
party system is likely mediated by external efficacy as well as political
trust.
The mediation analysis supports the contention that attitudinal
variables, political trust and external efficacy, mediate the relationship
between the number of parties and individual-level support for referendums.
Nevertheless, mediation analysis cannot alleviate concerns
about feedback effects between individual-level attitudes and referendum
support.29
As such, the sequence tested here significantly depends
on the theoretically stipulated relationship between the party system
and individual-level attitudes. Recall, then, that the causal sequence is
logically motivated: First, the effect of system-level variation on support
for direct democracy is necessarily mediated by individual-level attitudes.
Second, because parties are the main avenue for political representation,
political trust and external efficacy covary with country-level
differences in the number of parties (Miller and Listhaug, 1990;
Anderson and Guillory, 1997). Finally, because referendums allow
voters to alter existing policies or constrain their representatives, people
who view political authorities as unresponsive are likely to support
referendums.
6. Discussion and conclusion
The analysis indicates that popular support for referendums is
responsive to variation in the party system. Primarily, the multilevel
analysis indicates that decreases in the number of relevant parties
correspond with increases in referendum support. The general relationship
between the number of parties and referendum support holds
when the analysis omits any one country-case and is robust to a variety
of alternative model specifications. Further, the mediation analysis
provides evidence that the number of parties shapes referendum support
through individual-level political trust and external efficacy. First, decreases
in the number of parties contribute to feelings that authorities
are untrustworthy or unresponsive. In turn, feelings that elected officials
are unresponsive generate support for referendums. This sequence indicates
that referendum support results from limited representation by
political parties.
Several qualifications are worth noting. First, the extent to which the
argument applies to other types of popular vote processes is open to
question. While the dependent variable refers to referendums, limited
party representation might generate support for other types of popular
vote processes. For example, initiatives allow for citizens to propose new
policies (Altman, 2018) and therefore might be seen as a better
corrective to poor representation.30
Future studies might address
whether the institutional context shapes support for other types of
popular votes processes. Second, though the argument emphasizes political
dissatisfaction, the analysis provides evidence that pro-system
attitudes also contribute to referendum support. Notably, the importance
of living in a democracy and, more modestly, political interest
were associated with support for referendums. Majority
decision-making procedures are likely to be appealing to individuals
with strong democratic commitments.
Even so, the idea that referendum support is responsive to party
representation is consistent with research indicating that citizens are
concerned with policy outputs (Strebel et al., 2019; Esaiasson et al.,
2019). That is, citizens are more supportive of referendums when they
believe that party representatives are inattentive to their policy interests.
This is not to suggest that referendums will, in fact, generate
more representative policies, particularly for individuals with minority
views. Support for referendums might rest on unfounded beliefs that
public opinion—and the likely outcome of referendums—align with
individuals’ preferences (Werner, 2020). Rather, these findings indicate
that voters perceive referendums as a potential remedy to poor substantive
representation by political parties.
Citizens’ responsiveness to the party system challenges the idea that
support for referendums—and other popular vote processes—is necessarily
part of a ‘populist’ view, where representative institutions are seen
as “unfortunate substitutes for direct democracy” (Achen and Bartels,
2017, 88). The idea that referendum support constitutes a naive vision of
democratic practice suggests that citizens will endorse referendums at
similar levels regardless of the availability of party representation.
Against this view, this analysis shows that citizens respond to increases
in the number of relevant parties with decreased referendum support.
Popular support for referendums does not mean that citizens are
rejecting representation, but it does suggest that the party system is seen
as insufficiently representative.
An important implication is that electoral reforms that increase the
number of parties can decrease referendum support. Because the number
of parties is subject to institutional design (Lijphart, 2012; Shugart
and Taagepera, 2017), reforms that increase the number of parties are
likely to decrease popular demand for referendums.31
Citizens are sensitive
to the trade-offs between referendums and conventional party
representation, even if they do not see the two as completely interchangeable.
More broadly, this suggests that process preferences are
responsive to interactions between different democratic institutions.
Future research might examine how support for other decision-making
procedures, including more deliberative and technocratic varieties, is
shaped by the larger institutional context. Better understanding the
29
See the supplemental materials for sensitivity analyses (Fig. S2 and S3) and a discussion of the sequential ignorability assumption.
30
See Bowler et al. (2007): Among Californians who think ballot initiatives are a “good thing,” 76.9% believe that they “make government more responsive” (pg.
359).
31
See Table S5 for models that control for change in district magnitude between 1990 and 2012. While increases in district magnitude are associated with modest
decreases in referendum support, the contemporaneous effect of the number of parties is robust to the inclusion of this variable.
29
See the supplemental materials for sensitivity analyses (Fig. S2 and S3) and
a discussion of the sequential ignorability assumption.
30
See Bowler et al. (2007): Among Californians who think ballot initiatives
are a “good thing,” 76.9% believe that they “make government more responsive”
(pg. 359).
31
See Table S5 for models that control for change in district magnitude between
1990 and 2012. While increases in district magnitude are associated with
modest decreases in referendum support, the contemporaneous effect of the
number of parties is robust to the inclusion of this variable.
B.R. Bessen
Electoral Studies 68 (2020) 102219
11
perceived trade-offs between different processes would provide insight
into how democratic systems can better respond to citizens’ preferences.
Acknowledgment
For their helpful comments and suggestions, I thank Carew Boulding,
Brendan Connell, Jennifer Fitzgerald, Joseph Jupille, and Hannah Paul
as well as two anonymous reviewers and the editors of Electoral Studies.
Appendix
Table A.1
Mediation models.
(6) (7) (8)
Dependent variable:
Political Trust External Efficacy Referendum Support
Mediation Components
No. Parties 0.377***
(0.103)
0.249***
(0.084)
− 0.050
(0.046)
Political Trust − 0.100***
(0.007)
External Efficacy − 0.054***
(0.005)
Controls
Country-level
Majoritarian 0.503
(0.395)
0.473
(0.322)
0.049
(0.177)
Ln Pop. Votes, 02-12 0.024
(0.126)
0.212
(0.103)
0.069
(0.056)
Electoral Dem. Index 6.178**
(2.615)
4.329**
(2.133)
− 0.943
(1.171)
Population (logged) − 0.061
(0.108)
0.017
(0.088)
0.001
(0.049)
Individual-level
Political Interest 0.371***
(0.015)
0.058***
(0.019)
0.076***
(0.015)
Party Alternatives 0.199***
(0.005)
0.210***
(0.006)
0.076***
(0.005)
Imp. Democracy 0.088***
(0.007)
0.008
(0.008)
0.200***
(0.007)
Ideology 0.052***
(0.005)
0.063***
(0.007)
− 0.015***
(0.005)
Edu (years) 0.016***
(0.003)
− 0.001
(0.004)
− 0.009***
(0.003)
Income 0.036***
(0.005)
0.014**
(0.006)
− 0.018***
(0.005)
Female 0.153***
(0.024)
0.010
(0.030)
0.052**
(0.023)
Age − 0.003***
(0.001)
0.0002
(0.001)
− 0.002***
(0.001)
Observations 23,583 23,583 23,583
Log Likelihood − 47,748.330 − 52,703.480 − 47,025.390
*p<0.1; **p<0.05; ***p<0.01.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi.org/10.1016/j.electstud.2020.102219.
References
Achen, C.H., Bartels, L.M., 2017. Democracy for Realists: Why Elections Do Not Produce
Responsive Government. Princeton University Press.
Altman, D., 2018. Citizenship and Contemporary Direct Democracy. Cambridge
University Press.
Anderson, C.J., Guillory, C.A., 1997. Political institutions and satisfaction with
democracy: A cross-national analysis of consensus and majoritarian systems. Am.
Polit. Sci. Rev. 91 (1), 66–81.
Armingeon, K., Gerber, M., Leimgruber, P., Beyeler, M., Menegale, S., 2017. Comparative
Political Data Set I, 1960-2005. Institute of Political Science, University of Berne.
Baron, R.M., Kenny, D.A., 1986. The moderator–mediator variable distinction in social
psychological research: Conceptual, strategic, and statistical considerations. J. Pers.
Soc. Psychol. 51 (6), 1173.
Bauer, P.C., Fatke, M., 2014. Direct democracy and political trust: Enhancing trust,
initiating distrust–or both? Swiss Polit. Sci. Rev. 20 (1), 49–69.
Bengtsson, Å., Mattila, M., 2009. Direct democracy and its critics: Support for direct
democracy and ‘stealth’democracy in Finland. W. Eur. Polit. 32 (5), 1031–1048.
Bernauer, J., Vatter, A., 2012. Can’t get no satisfaction with the Westminster model?
Winners, losers and the effects of consensual and direct democratic institutions on
satisfaction with democracy. Eur. J. Polit. Res. 51 (4), 435–468.
B´ertoa, F.C., 2014. Party systems and cleavage structures revisited: A sociological
explanation of party system institutionalization in East Central Europe. Party Polit.
20 (1), 16–36.
B.R. Bessen
Electoral Studies 68 (2020) 102219
12
Bowler, S., Donovan, T., 2019. Perceptions of referendums and democracy: The
referendum disappointment gap. Polit. Govern. 7 (2), 227–241.
Bowler, S., Donovan, T., Karp, J.A., 2007. Enraged or engaged? Preferences for direct
citizen participation in affluent democracies. Polit. Res. Q. 60 (3), 351–362.
Brummel, L., 2020. ‘You can’t always get what you want’: The effects of winning and
losing in a referendum on citizens’ referendum support. Elect. Stud. 65 (102155),
1–11.
Cheneval, F., el Wakil, A., 2018. The Institutional Design of Referendums: Bottom-Up
and Binding. Swiss Polit. Sci. Rev. 24 (3), 294–304.
Coff´e, H., Michels, A., 2014. Education and support for representative, direct and stealth
democracy. Elect. Stud. 35, 1–11.
Coppedge, M., Gerring, J., Lindberg, S.I., Skaaning, S.-E., Teorell, J., Altman, D.,
Andersson, F., Bernhard, M., Fish, M.S., Glynn, A., et al., 2017. V-dem, version 7.
Varieties of Democracy (V-Dem) Project.
Cox, G.W., 1997. Making Votes Count: Strategic Coordination in the World’s Electoral
Systems. Cambridge University Press.
Craig, S.C., Maggiotto, M.A., 1982. Measuring Political Efficacy. Political Methodology,
pp. 85–109.
Craig, S.C., Niemi, R.G., Silver, G.E., 1990. Political efficacy and trust: A report on the
NES pilot study items. Polit. Behav. 12 (3), 289–314.
Dalton, R.J., 1985. Political parties and political representation: Party supporters and
party elites in nine nations. Comp. Polit. Stud. 18 (3), 267–299.
Dalton, R.J., 2004. Democratic Challenges, Democratic Choices. Oxford Univ. Press.
Dalton, R.J., Burklin, W.P., Drummond, A., 2001. Public opinion and direct democracy.
J. Democr. 12 (4), 141–153.
Donovan, T., Karp, J.A., 2006. Popular support for direct democracy. Party Polit. 12 (5),
671–688.
Donovan, T., Tolbert, C.J., Smith, D.A., 2009. Political engagement, mobilization, and
direct democracy. Publ. Opin. Q. 73 (1), 98–118.
el-Wakil, A., McKay, S., 2019. Disentangling Referendums and Direct Democracy: A
Defence of the Systemic Approach to Popular Vote Processes, pp. 1–18.
Representation.
Esaiasson, P., Persson, M., Gilljam, M., Lindholm, T., 2019. Reconsidering the role of
procedures for decision acceptance. Br. J. Polit. Sci. 49 (1), 291–314.
Font, J., Wojcieszak, M., Navarro, C.J., 2015. Participation, representation and expertise:
Citizen preferences for political decision-making processes. Polit. Stud. 63 (S1),
153–172.
Gutmann, A., Thompson, D.F., 2009. Why Deliberative Democracy? Princeton University
Press.
Hardman, H., 2011. Electoral System Change in Europe since 1945: Bulgaria. www.elec
toralsystemchanges.eu.
Held, D., 2006. Models of Democracy. Stanford University Press.
Hibbing, J.R., Theiss-Morse, E., 2001. Process preferences and American politics: what
the people want government to be. Am. Polit. Sci. Rev. 95 (1), 145–153.
Hug, S., Tsebelis, G., 2002. Veto players and referendums around the world. J. Theor.
Polit. 14 (4), 465–515.
Imai, K., Keele, L., Tingley, D., 2010. A general approach to causal mediation analysis.
Psychol. Methods 15 (4), 309.
Imai, K., Keele, L., Tingley, D., Yamamoto, T., 2011. Unpacking the black box of
causality: Learning about causal mechanisms from experimental and observational
studies. Am. Polit. Sci. Rev. 105 (4), 765–789.
Karp, J.A., Banducci, S.A., 2008. Political efficacy and participation in twenty-seven
democracies: How electoral systems shape political behaviour. Br. J. Polit. Sci. 38
(2), 311–334.
King, G., Tomz, M., Wittenberg, J., 2000. Making the Most of Statistical Analyses:
Improving Interpretation and Presentation. Available at: SSRN 1083738.
Krull, J.L., MacKinnon, D.P., 1999. Multilevel mediation modeling in group-based
intervention studies. Eval. Rev. 23 (4), 418–444.
Laakso, M., Taagepera, R., 1979. “Effective” number of parties: a measure with
application to West Europe. Comp. Polit. Stud. 12 (1), 3–27.
Ladner, A., 2002. Size and direct democracy at the local level: the case of Switzerland.
Environ. Plann. C Govern. Pol. 20 (6), 813–828.
Levi, M., Stoker, L., 2000. Political trust and trustworthiness. Annu. Rev. Polit. Sci. 3 (1),
475–507.
Lijphart, A., 2012. Patterns of Democracy: Government Forms and Performance in
Thirty-Six Countries. Yale University Press.
Lipset, S.M., Rokkan, S., et al., 1967. Cleavage Structures, Party Systems, and Voter
Alignments: An Introduction. Free Press, New York.
Madison, J., 1787. The Federalist Papers : No. 10.
Marien, S., 2011. The effect of electoral outcomes on political trust: A multi–level
analysis of 23 countries. Elect. Stud. 30 (4), 712–726.
Marien, S., Kern, A., 2018. The winner takes it all: revisiting the effect of direct
democracy on citizens’ political support. Polit. Behav. 40 (4), 857–882.
Matsusaka, J.G., 2005. Direct democracy works. J. Econ. Perspect. 19 (2), 185–206.
Miller, A.H., Listhaug, O., 1990. Political parties and confidence in government: A
comparison of Norway, Sweden and the United States. Br. J. Polit. Sci. 20 (3),
357–386.
Mohrenberg, S., Huber, R.A., Freyburg, T., 2019. Love at First Sight? Populist Attitudes
and Support for Direct Democracy. Party Politics. In press.
Qvortrup, M., 2018. Referendums Around the World. Springer.
Rose, R., Weßels, B., 2020. Do populist values or civic values drive support for
referendums in europe? Eur. J. Political Res. https://doi.org/10.1111/1475-
6765.12399. In press.
Schuck, A.R., de Vreese, C.H., 2015. Public support for referendums in Europe: A crossnational
comparison in 21 countries. Elect. Stud. 38, 149–158.
Shugart, M.S., Taagepera, R., 2017. Votes from Seats: Logical Models of Electoral
Systems. Cambridge University Press.
Smith, D.A., Tolbert, C., 2009. Educated by Initiative: The Effects of Direct Democracy on
Citizens and Political Organizations in the American States. University of Michigan
Press.
Smith, D.A., Tolbert, C.J., Keller, A.M., 2010. Electoral and structural losers and support
for a national referendum in the U.S. Elect. Stud. 29 (3), 509–520.
Steenbergen, M.R., Jones, B.S., 2002. Modeling multilevel data structures. Am. J. Polit.
Sci. 218–237.
Strebel, M.A., Kübler, D., Marcinkowski, F., 2019. The importance of input and output
legitimacy in democratic governance: Evidence from a population-based survey
experiment in four West European countries. Eur. J. Polit. Res. 58 (2), 488–513.
Struthers, C., Li, Y., Shugart, M., 2018. National and District Level Party Systems
Datasets.
Taagepera, R., 2002. Implications of the effective number of parties for cabinet
formation. Party Polit. 8 (2), 227–236.
Taagepera, R., 2003. Arend Lijphart’s dimensions of democracy: Logical connections and
institutional design. Polit. Stud. 51 (1), 1–19.
Tingley, D., Yamamoto, T., Hirose, K., Keele, L., Imai, K., 2014. Mediation: R Package for
Causal Mediation Analysis.
Werner, H., 2020. If I’ll win it, I want it: The role of instrumental considerations in
explaining public support for referendums. Eur. J. Polit. Res. 59 (2), 312–330.
Weßels, B., 1999. System characteristics matter: Empirical evidence from ten
representation studies. In: Policy Representation in Western Democracies. Oxford
University Press on Demand.
Weßels, B., 2007. Political representation and democracy. In: The Oxford Handbook of
Political Behavior.
B.R. Bessen