Threat or corrective?
The causal eﬀect of challenger party entrance on political trust ∗
Final Paper of the Course:
’Political Representation: Elites and Masses in the 21st century’
Lecturer: Assoc. Prof. Dr. Armen Hakhverdian
September 1, 2019
Change is an elementary feature of democracies. The possibility that voters
oust those in power and replace them with others of their choice is already a
minimum deﬁnition of democracy. However, in developed democracies a dis-
tinction can be made between the replacement of governments and changes to
the make-up of legislative bodies. In Europe and beyond, parliaments and party
systems are subject to fundamental shifts. With the waning importance of es-
tablished cleavages (Best 2011), challenger parties of diverse backgrounds have
emerged (Akkerman, De Lange, and Rooduijn 2016; Dolezal 2010). So far, it
is unclear whether these changes represent a threat to the fundamental proper-
ties of western democracies or a healthy corrective to an increasingly detached
When the German radical-right party ’Alternative f¨ur Deutschland’ (AfD)
overcame the electoral threshold of ﬁve percent and won 94 seats in the Ger-
man Bundestag in 2017, pundits had already identiﬁed the main reason why
12.6 percent of the electorate supported a radical right populist party: not the
actual issues put forward by the party mattered, but its ability to mobilise dis-
satisfaction among the voters. Voters turned to the AfD as a means to express
distrust in the current political system, their vote was a ’protest-vote’ (Sch¨afer
2017). It is an established ﬁnding that challenger parties draw support from
politically distrustful voters (Hooghe and Dassonneville 2018; Schumacher and
Rooduijn 2013; Van der Brug 2003). However, the development of political sup-
port after a challenger party gained access to the legislature and represents its
voters’ interests is less clear.
∗An ealier version of this paper was presented at the Summer School of the ECPR Standing
Group Political Parties
This study is interested in the causal eﬀect of challenger parties’ representa-
tion in national parliaments on political trust. Scholars oﬀer conﬂicting theories
of this link: Miller and Listhaug (1990) argue that challenger parties function as
some kind of pressure valve that - once they gain parliamentary status - chan-
nels distrust back into the electoral system. Hooghe and Dassonneville (2018)
criticise this view and claim that protest voters become more distrusting in the
long run, since they subsequently identify with the party (and its policies) they
initially voted for only out of protest (Dinas 2014).
This study sets out to address this theoretical conundrum. As political trust
aﬀects challenger party support (Van der Brug 2003; Schumacher and Rooduijn
2013), endogeneity problems emerge in the study of the link between chanllenger
party representation and political trust. Hence a regression discontinuity design
(RDD) is employed. The RDD is able to deal with issues of endogeneity and
can give a precise estimate of the local eﬀect of challenger parties’ overcoming of
the electoral threshold. It will contribute to the literature on political trust and
provide critical evidence to break the theoretical deadlock the current debate
The paper proceeds as follows: after a theoretical section that formulates two
main hypotheses, the main variables are explained, before the research design
and data are described. Lastly, the main ﬁndings including some robustness
checks are presented, and their implications discussed.
2 Party system change and political trust
2.1 Parties as ’pressure valve’
In their 1990 paper, Arthur H. Miller and Ola Listhaug compare levels of po-
litical trust in the period 1964-86. They observe that while all countries were
subject to similar factors aﬀecting political trust (most importantly stagﬂation
and growing criticism of the welfare state), it plummeted in Sweden and the
US, while - after a short decrease in the mid-seventies - political trust returned
to initial levels in Norway (1990: 361f). The authors argue that ”political dis-
content in Norway was reduced because new parties provided the disaﬀected
with a means of representation, thus channelling dissatisfaction back into the
electoral arena” (1990: 357). While the more ﬂexible party system of Norway
allowed voters to express their discontent and gain representation through a
new party, the more stable party systems of Sweden (four percent threshold)
and the US (ﬁrst-past-the-post voting) did not oﬀer this option so easily (1990:
382f). Generalizing on this logic, one can hypothesise that decreasing levels of
trust make political competition increasingly centrifugal as discontented voters
demand new alternatives. Once these alternatives gain representation, trust in
political institutions should resurge, as formerly distrustful voters’ interests are
represented. On the other hand, if voters’ discontent does not produce electoral
changes, it remains unaddressed and should thus grow further (1990: 364f).
A diﬀerent theoretical approach yields similar expectations about this link.
Following the vast literature on ’winners’ and ’losers’ of elections building on
Anderson & Guilleroy (1997) allows to hypothesise about the relationship be-
tween satisfaction with the political status quo and the representation of chal-
lenger parties. This approach conceptualises voters as ’winners’ and ’losers’ of
an election, dependent on whether they supported the parties that formed the
government subsequently (Anderson and Guillory 1997: 68). Voters of subse-
quently governing parties are more inclined to believe that democracy functions
properly in their country for (either or both of) two reasons: First, in a process
of motivated reasoning, voters who already hold favourable attitudes towards
the governing parties (and hence voted for them) should be more inclined to
believe that things go well once these parties are in power (Taber and Lodge
2006). Second, as citizens in favour of a given party should also share political
views alike to the parties’ platform, the political system should be a ”friendlier
place” for these citizens once these parties are in power (Anderson and Guil-
lory 1997: 68). And indeed, the positive eﬀect of being an election ’winner’ is
reduced the more distant a ’winning’ voter is from the parties she or he voted
for (Curini, Jou, and Memoli 2014: 258, Campbell 2015).
How does this apply to challenger party representation? Although parlia-
mentary status is irrelevant for the deﬁnition of ’winners’ and ’losers’ in the
classic literature, I believe the theory can be extended to apply to challenger
parties’ representation. Even though their parties did not necessarily win oﬃce,
voters of formerly unrepresented challenger parties that do gain parliamentary
status can be conceived of as ’winners’ in the sense that their views are now
represented in parliament. Hence issues important to these voters are more vis-
ible and are more likely to ﬁnd their way into politics compared to the formerly
unrepresented status. Earlier ﬁndings indicate that winning parliamentary seats
can produce similar eﬀects for voters’ support of political institutions as winning
oﬃce, and especially so among formerly out-of-parliament opposition parties.
Those voters that support a party that does not win seats experience a decline
in satisfaction. This eﬀect seems to be of similar size as the eﬀect of winning
an election compared to losing it (Singh, Karako¸c, and Blais 2012).
Summarizing, if a party formerly not holding legislative seats enters parlia-
ment, I expect a subsequent increase in the level of political trust, compared to
cases were new parties fail to gain representation. This is my ﬁrst hypothesis:
’Pressure valve’ hypothesis:
When an extra-parliamentary opposition party wins parliamentary seats, politi-
cal trust should increase subsequently, compared to cases in which new parties
fail to gain representation.
2.2 A spiral of distrust
There is, however, another possibility to conceptualise the eﬀect of representa-
tion of challenger parties on political trust. Marc Hooghe and Ruth Dasson-
neville counterargue Miller and Listhaug and show that in the aftermath of the
2009 Belgian election, voters of the protest parties Vlaams Belang (VB) and
Nieuw-Vlaamse Alliantie (N-VA) have experienced a more substantial decrease
in political trust than other voters (2018). The causal mechanism they imagine
is that casting a vote for a protest party is an act of group selection, that in
turn shapes attitudes. Actors interpret their own behaviour subsequently and
develop ﬁtting preferences and attitudes (Dinas 2014; Hooghe and Dassonneville
2018: 109). Voting for protest parties hence makes voters hold more distrusting
attitudes towards the political system. Hooghe and Dassonneville argue subse-
quently that Miller and Listhaug are wrong. Instead of addressing discontent
with the party system, the emergence of protest parties will in fact lead to a
spiral of distrust, a threat rather than a chance to return to stability for party
systems (2018: 124ﬀ).
Hooghe and Dassonneville’s ﬁndings do not directly contradict Miller and
Listhaug, but indicate that voting for a protest party further decreases trust in
political institutions. Additionally, Miller and Listhaug used political cynicism,
external eﬃcacy, and trust in the government rather than institutional trust.
Also, Hooghe and Dassonneville’s argument only applies to these protest par-
ties’ voters rather than the entire electorate. Even though Miller and Listhaug
employ the term ’protest party’, it seems they refer to any party challenging
the existing party system, as an instrument for voters unhappy with the cur-
rent supply to express their discontent (Miller and Listhaug 1990: 364f). What
Hooghe and Dassonneville rightly question is the mechanism that should revive
political trust - they show that a party running on a protest platform can lead
to a ’spiral of distrust’ rather than a return to former levels.
I argue that this is dependent on a parties’ platform. If the rhetoric of par-
ties is anti-elitist, it is likely that an increase in resources and attention through
parliamentary representation further decreases political trust. A broad litera-
ture has been concerned with the eﬀect of populist parties’ rhetoric on political
satisfaction. Authors have found that voting for populist parties aﬀects polit-
ical eﬃcacy and trust in the elites negatively (Rooduijn, Van der Brug, and
De Lange 2016) and that political messages blaming individuals’ or groups’ for
social problems (Busby, Gubler, and Hawkins 2019) and populist news fram-
ing (Rooduijn, Van der Brug, De Lange, and Parlevliet 2017) increase political
cynicism. It is therefor straightforward to expect that political representation
of such parties might decrease satisfaction with the political system, the gov-
ernment, and the parliament and decrease political eﬃcacy.
This leads me to my second hypothesis:
’Populist threat’ hypothesis:
When an extra-parliamentary opposition party winning parliamentary seats is
running on a populist platform, political trust should decrease subsequently,
compared to cases in which a new populist party is not represented.
3.1 Parliamentary status
Winning parliamentary seats has several implications for political parties. Usu-
ally, parliamentary status entails ﬁnancial support through state funding, the
opportunity to participate in policy making, as well as increased media atten-
tion. Hence, a party winning seats will become more present in the public
discourse, compared to parties that do not gain access to parliament (Dinas,
Riera, and Roussias 2015: 189f). This can theoretically aﬀect political trust in
several ways: challenger parties visibly represent formerly discontented voters
and address their distrust. Protest parties can also be increasingly visible in a
society and make the ’issue’ of political distrust more salient, reaching more vot-
ers through additional means or undermining their own supporters’ trust in the
political system further. Lastly, the legislative entrance of the challenger party
could produce electoral ’winners’, which subsequently develop a more favourable
perspective on the political system.
3.2 Political Trust
Political trust denotes a relationship between a subject (in our case: the voter)
and an object (the institution that is trusted): ’A trusts B to do X’ (Hardin
1999: 26). It can be conceived in diﬀerent ways, but usually means an eval-
uation of how much one trusts a certain political entity. Sometimes it refers
to diﬀuse trust in regimes, such as democracy (Van der Meer and Hakhverdian
2017), sometimes it refers to trust in political institutions, such as the parlia-
ment, the judiciary, the police or politicians (Hakhverdian and Mayne 2012;
Marien and Hooghe 2011). These institutional trust indicators are correlated
and describe the same dimension (Hooghe, Marien, and Pauwels 2011: 258).
The most common use - especially by US-based scholars - is to deﬁne political
trust as ”a basic evaluative orientation toward the government” (Hetherington
1998: 791). While the ambiguous use of the term ”political trust” in the litera-
ture is highly problematic, all of these indicators are interesting for the present
analysis: The most diﬀuse indicator, satisfaction with the way democracy works
in one’s country, is probably the most important for democratic stability and
is arguably most related to the Miller-Listhaug model: when voters see that
new parties at odds with established parties can gain access to parliament, that
should signal to them that the democratic system is working. However, Miller
and Listhaug themselves use a variety of indicators that are best described by
political cynicism and trust in the government (1990: 360). Hooghe and Dasson-
neville use questionnaires asking directly for evaluations of trust in institutions
and the government (2018: 112). To address concerns that diﬀerent indicators
might yield diﬀerent results, I will include four diﬀerent dependent variables in
my analysis: trust in the parliament (institutional), trust in the government
(speciﬁc), satisfaction with democracy (diﬀuse), and political cynicism (Easton
1975). Table 5 in the appendix shows how these indicators are asked in the
4 Research Design
To estimate the causal eﬀect of parliamentary representation, I propose a regres-
sion discontinuity design (RDD). It is a well-established ﬁnding and assumption
of both models that lower levels of political trust lead to higher levels of electoral
support for challenger parties (Hetherington 1998; Hooghe, Marien, and Pauwels
2011; Hooghe and Dassonneville 2018: 107ﬀ). A solely observational technique
like a regression analysis would hence suﬀer from endogeneity, as those chal-
lenger parties which gain representation are already in environments of lower
political trust - the observation would be biased. The panel design employed
by Hooghe and Dassonneville, as well as Singh, Karako¸c, and Blais might allow
to observe attitude changes among diﬀerent groups of voters, but it does not
compare diﬀerent potential outcomes, nor is there any variation on the indepen-
dent variable (the parliamentary status of the opposition party). To observe the
actual causal eﬀect, one needs to look at subsequent changes and compare cases
that are as alike as possible , but diﬀer on the independent variable. The RDD
allows me to do this by exploiting a common institutional feature of democracies:
electoral thresholds. Discriminating between those parties that just managed
to overcome the electoral threshold and those that did not enables me to treat
the status of parliamentary representation as a ’treatment’. The fraction of the
electorate that makes the diﬀerence between these two cases is so small we can
assume the dependent variable to be neither aﬀected by voters or elites expec-
tations, nor by variation in other variables aﬀecting the independent variable
(opposition party’s parliamentary status). Hence, the treatment becomes ex-
ogenous. As a consequence, the parties just above and just below the threshold
are similar enough to allow for a ceteris paribus comparison estimating causal
eﬀects (Abou-Chadi and Krause 2018; Angrist and Pischke 2015; Dinas, Riera,
and Roussias 2015).
4.2 Case selection
The parties of interest in this analysis are the largest extra-parliamentary oppo-
sition parties on the national level within established democracies. These parties
have not held parliamentary seats prior to the election under study. This means
that these parties could have held seats at an earlier point in time, but not at the
time of the election under study. I focus on national elections, as they are most
salient to voters of most party systems. Additionally, this restriction improves
comparability. Tables 6 and 7 in the appendix show the cases included.
Cases where the largest party gained access to the parliament by winning
a constituency without meeting the threshold are excluded, as they cannot be
suﬃciently modelled with a strict regression discontinuity design. Employing a
fuzzy RDD is possible, however this would include parties that might be sub-
stantially diﬀerent from the challenger parties I observe. When parties gain
oﬃce below the threshold, it might be that they employed a diﬀerent strategy
(regional as opposed to national mobilisation) or that they represent a minor-
ity and were assigned seats to ensure representation - both could be related to
political trust. To make my analysis dependent on as few assumptions as pos-
sible, and to make the estimation of the causal eﬀect most valid, I will utilize
a strict RDD at the cost of dropping these cases. In cases where the largest
extra-parliamentary party did not gain a seat, but smaller parties did, I argue
that the Miller-Listhaug model would still expect dissatisfaction from the voters
of the largest party and hence include them.
To exemplify, the model would compare the change in political trust after the
election for the German Bundestag in 2013, when the AfD missed the 5 percent-
threshold, winning 4.7 percent of the votes, to the change after the National-
ratswahl in Austria 2017, where the Liste Pilz overcame the 4 percent-threshold
with 4.4 percent. For both parties this was the ﬁrst election to participate in
and they stayed close to the threshold with diverging outcomes.
The most eﬃcient way to obtain data on election results is the ParlGov-website
by Holger D¨oring and Philip Manow. The project collected election results from
”all EU and most OECD democracies (37 countries). The database combines
approximately 1600 parties, 950 elections (8700 results), and 1500 cabinets (3700
parties)” (D¨oring and Manow 2018).
The ParlGov data will be merged with survey data about political support
in national electorates. Most comparative studies employ the European Social
Survey’s (ESS) data (Hakhverdian and Mayne 2012; Marien and Hooghe 2011;
Van der Meer and Hakhverdian 2017). The ESS has several advantages: i)
it asks the same question in diﬀerent countries, making comparisons simple.
ii) It includes data from several European countries. iii) It covers an extended
period from 2002-2016. iv) it asks for diﬀerent indicators of political satisfaction
(satisfaction with democracy, satisfaction with government, trust in parliament)
in every wave (Norwegian Centre for Research Data 2018).
Compared to the ESS, the Comparative Study of Electoral Systems (CSES)
covers not as many variables related to political trust (Satisfaction with democ-
racy and political cynicism), but increases the number of cases and is matched
to the elections. Beyond that, the CSES asked the same questions in many
democracies globally from 1996 to 2016 and also This means that - once the
sample has been restricted to post-election surveys - the measurement happens
the same moment in the electoral cycle, namely right after an election. Since
I am interested in change in political support, the comparison becomes more
valid - all variables that change throughout the electoral cycle are controlled for
(The Comparative Study of Electoral Systems 2018). This study will use the
data from both surveys, so results can be compared.
4.4 The regression discontinuity approach
The RDD-design has three major components: the dependent variable (political
trust at electiont+1), the assignment variable (electoral support for challenger
party at electiont) and the cut-oﬀ variable (electoral threshold at electiont).
As I drop all cases where parties did not face a formal electoral threshold (e.g.
by winning a single constituency in a mixed-member proportional system), I
can make use of a so-called sharp RDD. This means that my cut-oﬀ is a strict
discriminator conditioning the treatment (parliamentary status):
DP S =(1 if Xi≥c
0 if Xi< c
where DP S denotes the treatment (whether the challenger party gained par-
liamentary status), Xistands for the assignment variable (electoral support of
challenger party i) and cdenotes the cut-oﬀ (electoral threshold). Parties above
the threshold gain parliamentary status and those below it do not. The assign-
ment variable sharply discriminates the treatment conditions. To enable the
comparison of party systems with diﬀerent electoral thresholds, a party’s vote
share is recoded as the distance to the electoral threshold. It is negative when
the challenger party lies below the threshold and positive when it lies above it.
Under the assumption that parties with a vote share just above the threshold
and those just below it are suﬃciently similar on other indicators to allow for
aceteris paribus comparison (because the diﬀerence in vote share is suﬃciently
small to assume randomness), we can estimate the causal eﬀect of the treatment:
ρ=E[∆Y(DP S = 1) −∆Y(DP S = 0)],
where ρis the causal eﬀect of challenger parties’ parliamentary status on
political trust and ∆Yindicates the change in political trust when a challenger
party won seats in parliament (DP S = 1) or when it did not (DP S = 0) (King,
Keohane, and Verba 1994: 94ﬀ; Morgan and Winship 2007: 2ﬀ).
4.5 Model speciﬁcations
As might be expected, this estimate is highly dependent on the speciﬁcations of
the model. Two factors are especially important: the bandwidth of the observed
variation of the assignment variable to both sides of the cut-oﬀ and the design
of the model estimator.
The correct deﬁnition of a bandwidth that deﬁnes which cases are included
and which are excluded is crucial for a meaningful application of the RDD.
While smaller bandwidths give a more precise estimation of the local average
treatment eﬀect (higher internal validity), broader bandwidths make the model
more robust. I will use the common optimal bandwidth algorithm proposed by
Imbens and Kalyanaraman (2010) to deﬁne the bandwidth for the main model.
For robustness checks, I will run the models again with half and double the
Following common practice, I will estimate a local linear regression estimator
with a triangular kernel. This estimator is more sensitive to observations closer
to the cut-oﬀ (which is the zone of interest) and allows for nonlinear estimation
of the local average treatment eﬀect (LATE; Abou-Chadi and Krause 2018: 8;
Dinas, Riera, and Roussias 2015: 191f; Imbens and Kalyanaraman 2010; Porter
4.6 Populism as a moderator
While the simple regression discontinuity design serves well to test whether
the representation of new parties increases political trust on average (’pressure
valve’ hypothesis), it is insuﬃcient to give any indication whether the eﬀect
might be diﬀerent for populist parties (’populist threat’ hypothesis). To test
this, I will split my sample into two, one of elections where the major challenging
party was a populist one and one where the major challenging party was not
populist. Similar to modelling an interaction, this allows me to assess whether
the impact of a populist party on political satisfaction diﬀers from that of a
non-populist party. To identify whether a party is populist or not, I will rely on
the PopuList, a collection of populist, radical right, radical left, and eurosceptic
parties (Rooduijn, Van Kessel, et al. 2019). Missing values were added by
hand based on the authors’ judgment, employing the deﬁnition of populism as a
’thin-centered’, anti-elitist ideology which constructs a dichotomous distinction
between ’the people’, whose interest populist parties claim to represent and ’the
elite’, a term that can describe many groups but is usually referring to those in
power, who - as populist rhetoric argues - do not represent the true ’will of the
people’ (Mudde 2004; Mudde 2013).
This section will present the ﬁndings of regression discontinuity models based on
CSES and ESS data, seperately for each of the dependent variables. Estimates
have been obtained using Drew Dimmery’s rdd package for R(Dimmery 2016;
R Core Team 2017). Note that these results are based on a limited data set and
do not include the full analysis yet. This limited number of cases is especially
important as the regression discontinuity design is highly responsive towards
individual cases close to the threshold.
5.1 Satisfaction with Democracy
As mentioned before, from the perspective of the Miller-Listhaug model, satis-
faction with democracy is probably the most interesting indicator of political
contentment. Miller and Listhaug argue that political trust should increase once
extra-parliamentary opposition parties win seats, as formerly unrepresented vot-
ers’ see their concerns represented. However, the visual representations of the
regression discontinuity models indicate otherwise (see ﬁgure 1): both the ESS
Figure 1: Change in satisfaction with the national democracy, plotted by
vote share relative to threshold with local linear regression curves. Dashed lines
and the CSES data show a negative eﬀect. However, as visible from the con-
ﬁdence intervals (dashed lines), as well as table 1, this eﬀect is not signiﬁcant.
Either way, this variable does not support the ’pressure valve’ hypothesis.
5.2 Satisfaction with government
Figure 2: Change in satisfaction with the way the government does
its job, plotted by vote share relative to threshold with local linear regression
curves. Dashed lines indicate 95%-CI’s.
Moving on to a more speciﬁc form of political support, the satisfaction with
government performance, I ﬁnd a more substantial negative eﬀect of legislative
entrance by challenger parties (see ﬁgure 2 and table 1). It is however not
signiﬁcant on any of the established levels. This is especially interesting since
this indicator is generally least likely to be aﬀected: while the democratic process
is directly impacted by the representation of challenger parties, the national
government is not. Again, this eﬀect points in the opposite direction of what was
expected: trust in the government decreases when new parties enter parliament.
One could speculate here that the large bandwidth weakens the endogeneity
assumption: as more cases away from the threshold are included, the likelihood
that weak government support is aﬀecting the electoral success of opposition
parties (rather than vice versa) increases. However, since a model with half
the recommended bandwidth produces a somewhat signiﬁcant negative eﬀect
(p < 0.1), this is unlikely (see table 4 in the appendix).
5.3 Trust in parliament
Figure 3: Change in trust in the national parliament, plotted by vote share
relative to threshold with local linear regression curves. Dashed lines indicate
Trust in parliament is arguably the most directly aﬀected indicator of polit-
ical support. And, indeed, this is the only indicator with somewhat signiﬁcant
results (see ﬁgure 3). The model shows a signiﬁcant negative eﬀect for both the
standard model and the model with reduced bandwidth (p < 0.1, p < 0.05; see
table 4 in the appendix). Note that this is one of the indicators that is used
by Hooghe and Dassonneville, who ﬁnd negative eﬀects of voting for a populist
extra-parliamentary opposition party that wins seats.
Figure 4: Change in political cynicism, plotted by vote share relative to
threshold with local linear regression curves. Dashed lines indicate 95%-CI’s.
5.4 Political cynicism
Lastly, Miller and Listhaug used an indicator of political cynicism to construct
their index of political trust. The regression discontinuity model for the two
related items in the CSES, namely ’who is in power makes a diﬀerence’ and ’who
people vote for makes a diﬀerence’ do not seem to be related to the electoral
success of opposition parties (table 1 and ﬁgure 4).
Variable Data BW LATE SE t N<c N>c
Democracy ESS 4.73 −0.33 0.32 -1.03 13 8
Democracy CSES 3.98 −0.26 0.26 -0.98 20 13
Parliament ESS 4.29 −0.89t0.46 -1.96 12 8
Government ESS 5.45 −0.73 0.46 -1.6 13 10
Who is in power CSES 3.46 0.13 0.22 0.63 20 12
Who people vote CSES 3.89 −0.08 0.2 -0.38 20 13
Table 1: Results of regression discontinuity estimates with Imbens-
Kalyanaraman optimal bandwidth using triangular kernel.
For results with double and half the bandwidths, consult Table 4 in the Appendix.
tp < 0.1
5.5 Robustness checks
5.5.1 Placebo tests
In order to check for the robustness of these (already mostly statistically non-
signiﬁcant) results, placebo tests are employed. These tests estimate regression
discontinuities for random thresholds other than the initial threshold. The fol-
lowing plots present these placebo tests at thresholds c= 1 and c=−2 for the
CSES data. Graph of ESS data are excluded for the sake of brevity, however
yield similar results.
Figure 5: Change in satisfaction with the national democracy, plotted by
vote share relative to threshold with local linear regression curves. Dashed lines
Figure 6: Change in ’Who is in power makes a diﬀerence’, plotted by
vote share relative to threshold with local linear regression curves. Dashed lines
As can be seen from ﬁgures 5, 6, and 7, the placebo tests show no clear
eﬀect for the chosen thresholds (as expected). This is especially true for the
measure of satisfaction with democracy, while the cynicism indicators somewhat
randomly but insigniﬁcantly change in their eﬀect with changing thresholds.
This latter ﬁnding points to the noisiness of the data and cautions against a
strong interpretation of the results.
5.5.2 Dropping outlier
One case in the CSES-sample aﬀects the interpretation very strongly: the leg-
islative election in Mexico in 2003 saw a steep decrease in political trust after
the new party ’Convergence’ entered the legislature. To test whether the results
are entirely driven by this case, I drop it from the sample and estimate the RDD
Again, robustness tests are only reported for CSES data. Figure 8 and table
2 show that, once the case of the 2003 legislative election in Mexico is dropped,
Figure 7: Change in ’Who people vote for makes a diﬀerence’, plotted by
vote share relative to threshold with local linear regression curves. Dashed lines
the LATE of challenger party representation on political support turns positive.
This eﬀect is surprisingly signiﬁcant for the cynicism indicators. It might hence
be that challenger party representation decreases political cynicism (note the
inverse coding; see table 5). However, this is again building on only a few cases.
Also note that the Mexican case is not covered by the ESS, which found a
statistically insigniﬁcant negative eﬀect for satisfaction with democracy. So far,
it can mainly be noted that the ﬁndings are not robust.
Variable Data BW LATE SE t N<c N>c
Democracy CSES 3.14 0.36 0.28 1.28 20 10
Who is in power CSES 2.81 0.82∗∗∗ 0.21 3.8 19 9
Who people vote CSES 3.14 0.49∗0.22 2.28 20 10
Table 2: Results of regression discontinuity estimates with Imbens-
Kalyanaraman optimal bandwidth using triangular kernel for robustness test
with exclusion of outlier Mexico 2003.
∗p < 0.05; ∗∗∗ p < 0.001
5.5.3 Populism as a mediator
To test the second, ’populist threat’ hypothesis, I present the visualisations of
the LATE for the sample of populist challenger parties (ﬁgure 9) in the ESS
Figures 9 and 10 indicate that for both populist and non-populist parties,
the eﬀect of parliamentary entrance on political trust is non-existent or nega-
tive. Given the weak reliability of the data, I refrain from making meaningful
inferences here, even though the eﬀect of populist party entrance on trust in
parliament is signiﬁcant (p < 0.05; see table 3).
Figure 8: Robustness tests for political support/cynicism excluding Mexico
2003, plotted by vote share relative to threshold with local linear regression
curves. Dashed lines indicate 95%-CI’s.
This paper studied the causal eﬀect of legislative entrance of challenger parties
on political trust. It did so by making use of electoral thresholds in proportional
systems to estimate a local average treatment eﬀect (LATE) through a regression
discontinuity design (RDD). The ﬁndings mainly indicate that the measures are
too noisy and the samples too small to give an indication about the LATE, even
though most estimates point to a non-robust, non-signiﬁcant negative eﬀect for
the entrance of both populist and non-populist challenger parties. The two
formulated hypotheses cannot be conﬁrmed because the data fail to address the
theoretical questions. From here, several pathways are available to study the
question in more detail.
The noisiness of the data is likely related to two factors: Firstly, the points
of measurement are several years apart. While the CSES-measure indicates the
change in political trust since the last election (usually four to ﬁve years), the
ESS measures unrelated to the election date every two years (see ﬁgure 11 in
the appendix). This issue could be solved by using data that measures politi-
cal support both before and after the election, such as the biannual Standard
Eurobarometer (Gesis 2018).
Secondly, the data looks at an aggregate measure of political support rather
Figure 9: Diﬀerent indicators of political support based on split sample of pop-
ulist parties, plotted by vote share relative to threshold with local linear re-
gression curves. Dashed lines indicate 95%-CI’s.
than speciﬁcally that of challenger parties’ voters. However, the theory behind
the hypotheses builds on mechanisms related to the speciﬁc voters, even though
Miller and Listhaug look at aggregate political trust. Additionally, former stud-
ies have shown that the impact of elections diﬀers for diﬀerent groups of voters
(Anderson and Guillory 1997; Hooghe and Dassonneville 2018; Singh, Karako¸c,
and Blais 2012). This issue can only be solved when employing panel data to
estimate the eﬀect of challenger party entrance on political trust. To make a
generalized statement beyond a single case would involve to bring together a
massive amount of panel data across polities and time.
A further line of inquiry might look into a multi-level design to interact
party system change with individual vote choice. The electoral ’winners/losers’
literature indicates that satisfaction of extra-parliamentary opposition parties’
voters should be most aﬀected when political parties win seats (Singh, Karako¸c,
and Blais 2012). It might well be that the aggregate eﬀects are diﬀerent from
the eﬀects found for diﬀerent groups of voters. A multi-level model could add
nuance to the overall picture, even though it cannot claim to study causality.
Summarizing, a more exhaustive analysis is certainly necessary to make a
claim about the impact of challenger party entrance into parliament on political
trust. While the overall eﬀects found draw a rather bleak sketch of electoral
change in modern democracies, this ﬁnding is driven by only a few cases and
neither robust nor statistically signiﬁcant. As low satisfaction with the gov-
ernment is not necessarily a sign for a weak democracy, but possibly with bad
Figure 10: Diﬀerent indicators of political support based on split sample of
non-populist parties, plotted by vote share relative to threshold with local
linear regression curves. Dashed lines indicate 95%-CI’s.
performance or a critical citizenry (Alexander and Welzel 2017); and as politi-
cal cynicism seems unaﬀected, the ﬁnal picture might be more optimistic about
the prospects of electoral change - at this point one can only guess. This is
especially true since one robustness test indicated that challenger party rep-
resentation might decrease political cynicism. However, the ﬁndings underline
that these diﬀerent indicators of political support should not be substituted for
each other. Further analyses will increase clarity to make more conﬁdent judg-
ments about the eﬀect of challenger parties’ parliamentary representation on
the state of democracies.
Variable Data BW LATE SE t N<c N>c
Government ESS 3.82 −1.27 0.9 -1.42 4 4
Democracy ESS 2.29 −0.70.71 -0.98 4 3
Parliament ESS 5.7 −1.39∗0.53 -2.63 4 4
Government ESS 5.77 −0.51 0.61 -1.29 9 7
Democracy ESS 5.68 −0.17 0.37 -0.47 9 7
Parliament ESS 4.94 −0.57 0.89 -0.637 9 4
Table 3: Results of regression discontinuity estimates with Imbens-
Kalyanaraman optimal bandwidth using triangular kernel for robustness test
with split samples (populist/non-populist parties).
∗p < 0.05
Abou-Chadi, Tarik and Werner Krause (2018). “The Causal Eﬀect of Radical
Right Success on Mainstream Parties’ Policy Positions: A Regression Discon-
tinuity Approach”. In: British Journal of Political Science, pp. 1–19. issn:
Akkerman, Tjitske, Sarah L. De Lange, and Matthijs Rooduijn (2016). Radical
right-wing political parties in Western Europe: into the mainstream? London
and New York: Routledge.
Alexander, Amy C . and Christian Welzel (2017). “The myth of deconsolidation:
Rising liberalism and the populist reaction”. In: Journal of Democracy: Web
Anderson, Christopher J. and Christine A . Guillory (1997). “Political Insti-
tutions and Satisfaction with Democracy : A Cross-National Analysis of
Consensus and Majoritarian Systems”. In: The American Political Science
Review 91.1, pp. 66–81.
Angrist, Joshua D. and J¨orn-Steﬀen Pischke (2015). Mastering ’Metrics. Prince-
ton University Press.
Best, Robin E. (2011). “The declining electoral relevance of traditional cleav-
age groups”. In: European Political Science Review 3.2, pp. 279–300. issn:
Busby, Ethan C, Joshua R Gubler, and Kirk A Hawkins (2019). “Framing and
Blame Attribution in Populist Rhetoric”. In: The Journal of Politics 81.2,
pp. 616–630. issn: 0022-3816. doi:10.1086/701832.
Campbell, Ross (2015). “Winners, losers and the Grand Coalition: Political
satisfaction in the Federal Republic of Germany”. In: International Po-
litical Science Review 36.2, pp. 168–184. issn: 1460373X. doi:10 . 1177/
Curini, Luigi, Willy Jou, and Vincenzo Memoli (2014). “Satisfaction with Democ-
racy and the Winner / Loser Debate : The Role of Policy Preferences and
Past Experience Satisfaction with Democracy and the Winner / Loser De-
bate : The Role of Policy Preferences and Past Experience”. In: 42.May,
pp. 241–261. doi:10.1017/S0007123411000275.
Dimmery, Drew (2016). rdd: Regression Discontinuity Estimation.url:https:
Dinas, Elias (2014). “Does Choice Bring Loyalty ? Electoral Participation and
the Development of Party Identiﬁcation”. In: American Journal of Political
Science 58.2, pp. 449–465. url:http://www.jstor.org/stable/24363496.
Dinas, Elias, Pedro Riera, and Nasos Roussias (2015). “Staying in the First
League: Parliamentary Representation and the Electoral Success of Small
Parties”. In: Political Science Research and Methods 3.2, pp. 187–204. doi:
Dolezal, Martin (2010). “Exploring the stabilization of a political force: The
social and attitudinal basis of green parties in the age of globalization”. In:
West European Politics 33.3, pp. 534–552. issn: 01402382. doi:10. 1080/
D¨oring, Holger and Philip Manow (2018). ParlGov - Parliaments and Govern-
Easton, David (1975). “A Re-Assessment of the Concept of Political Support”.
In: British Journal of Political Science 5.4, pp. 435–457.
Gesis (2018). Overview of Eurobarometer surveys with standard EU and trend
question modules. Tech. rep. url:https:// www . gesis.org/ fileadmin /
upload/ dienstleistung / daten/umfragedaten/eurobarometer / eb%7B%
Hakhverdian, Armen and Quinton Mayne (2012). “Institutional trust, educa-
tion, and corruption: A micro-macro interactive approach”. In: Journal of
Politics 74.3, pp. 739–750. issn: 00223816. doi:10.1017/S0022381612000412.
Hardin, Russell (1999). “Do we want trust in government?” In: Democracy and
Trust. Ed. by Mark E. Warren. Cambridge University Press. Chap. 2, pp. 22–
Hetherington, Marc J (1998). “The Political Relevance of Political Trust”. In:
American Political Science Review 92.4, pp. 791–808. issn: 01621459. doi:
Hooghe, Marc and Ruth Dassonneville (2018). “A Spiral of Distrust: A Panel
Study on the Relation between Political Distrust and Protest Voting in Bel-
gium”. In: Government and Opposition 53.1, pp. 104–130. issn: 14777053.
Hooghe, Marc, Soﬁe Marien, and Teun Pauwels (2011). “Where Do Distrusting
Voters Turn if There is No Viable Exit or Voice Option? The Impact of Polit-
ical Trust on Electoral Behaviour in the Belgian Regional Elections of June
2009”. In: Government and Opposition 46.2, pp. 245–273. issn: 0017257X.
Imbens, Guido and Karthik Kalyanaraman (2010). “Optimal bandwidth choice
for the regression discontinuity estimator”. London.
King, Gary, Robert O. Keohane, and Sidney Verba (1994). “Causality and
Causal Inference”. In: Designing Social Inquiry: Scientiﬁc Inference in Qual-
itative Research. Princeton University Press. isbn: 9781400821211.
Marien, Soﬁe and Marc Hooghe (2011). “Does political trust matter? An em-
pirical investigation into the relation between political trust and support for
law compliance”. In: European Journal of Political Research 50.2, pp. 267–
291. issn: 03044130. doi:10 .1111 /j . 1475- 6765 .2010 . 01930. x. arXiv:
Miller, Arthur H. and Ola Listhaug (1990). “Political Parties and Conﬁdence in
Government : A Comparison of Norway , Sweden and the United States”.
In: British Journal of Political Science 20.3, pp. 357–386.
Morgan, Stephen and Christopher Winship (2007). Conterfactuals and Causal
Inference. Methods and principles for social research.
Mudde, Cas (2004). “The Populist Zeitgeist”. In: Government and Opposition
39.4, pp. 541–563.
— (2013). “Populism”. In: Oxford Handbook of Political Ideologies, pp. 8–9.
Norwegian Centre for Research Data (2018). European Social Survey Cumulative
File, ESS 1-8.doi:10.21338/NSD-ESS-CUMULATIVE..
Porter, Jack (2003). “Estimation in the Regression Discontinuity Model”. In:
Department of Economics, pp. 1–66. url:http: / / www .ssc . wisc .edu /
R Core Team (2017). R: A Language and Environment for Statistical Comput-
ing. Vienna, Austria. url:https://www.r-project.org/.
Rooduijn, Matthijs, Wouter Van der Brug, and Sarah L. De Lange (2016).
“Expressing or fuelling discontent? The relationship between populist voting
and political discontent”. In: Electoral Studies 43. issn: 02613794. doi:10.
Rooduijn, Matthijs, Wouter Van der Brug, Sarah L. De Lange, and Jante Par-
levliet (2017). “Persuasive Populism? Estimating the Eﬀect of Populist Mes-
sages on Political Cynicism”. In: Politics and Governance 5.4, p. 136. doi:
Rooduijn, Matthijs, Stijn Van Kessel, et al. (2019). The PopuList: An Overview
of Populist, Far Right, Far Left and Eurosceptic Parties in Europe. url:
Sch¨afer, Armin (2017). Kultur statt ¨
Schumacher, Gijs and Matthijs Rooduijn (2013). “Sympathy for the ’Devil?’”
In: Electoral Studies 32.1, pp. 124–133.
Singh, Shane, Ekrem Karako¸c, and Andr´e Blais (2012). “Diﬀerentiating winners:
How elections aﬀect satisfaction with democracy”. In: Electoral Studies 31.1,
pp. 201–211. issn: 02613794. doi:10.1016 / j . electstud . 2011.11.001.
Taber, Charles S and Milton Lodge (2006). “Motivated Skepticism in the Eval-
uation of Political Beliefs”. In: American Journal of Political Science 50.3,
pp. 755–769. issn: 0092-5853. doi:10.1111/j.1540-5907.2006.00214.x.
arXiv: 9809069v1 [arXiv:gr-qc].
The Comparative Study of Electoral Systems (2018). CSES INTEGRATED
MODULE DATASET (IMD). December 4, 2018 version. doi:10. 7804 /
Van der Brug, Wouter (2003). “How the LPF Fuelled Discontent: Empirical
tests of explanations of LPF support”. In: Acta Politica 38.1, pp. 89–106.
issn: 0001-6810. doi:10.1057/palgrave.ap.5500005.url:http://link.
Van der Meer, Tom W.G. and Armen Hakhverdian (2017). “Political Trust as
the Evaluation of Process and Performance: A Cross-National Study of 42
European Countries”. In: Political Studies 65.1, pp. 81–102. issn: 14679248.
Variable Data Bandwidth LATE SE t N<c N>c
Satisfaction ESS 4.73 (IK) −0.33 0.32 -1.03 13 8
Democracy ESS 2.36 (I K
2)−0.68 0.48 -1.42 7 6
ESS 9.45 (IK*2) −0.37 0.27 -1.37 13 13
CSES 3.98 (IK) −0.26 0.26 -0.98 20 13
CSES 1.99 ( IK
2)−0.53 0.38 -1.41 11 8
CSES 7.95 (IK*2) −0.09 0.2 -0.46 25 18
Trust in ESS 4.29 (IK) −0.89t0.46 -1.96 12 8
Parliament ESS 2.15 (I K
2)−1.42∗0.59 -2.4 7 6
ESS 8.59 (IK*2) −0.63 0.41 -1.55 13 12
Satisfaction ESS 5.45 (IK) −0.73 0.46 -1.6 13 10
Government ESS 2.72 (I K
2)−1.72t0.7 -1.81 8 6
ESS 10.89 (IK*2) −0.61 0.43 -1.42 13 13
Who is in power CSES 3.46 (IK) 0.13 0.22 0.63 20 12
can make a CSES 1.73 ( IK
2)0.09 0.27 0.31 10 8
diﬀerence CSES 6.91 (IK*2) 0.11 0.18 -0.63 25 17
Who people vote CSES 3.89 (IK) −0.08 0.20 -0.38 20 13
for can make a CSES 1.95 ( I K
2)−0.05 0.28 -0.2 11 8
diﬀerence CSES 7.8 (IK*2) −0.04 0.16 -0.25 25 18
Table 4: Results of regression discontinuity estimates using triangular kernel.
’IK’ stands for Imbens-Kalyanaraman optimal bandwidth
tp < 0.1
∗p < 0.05
Satisfaction w. Democracy ESS How satisﬁed are you with the way democracy works in 0 - 10
Satisfaction w. Democracy CSES
On the whole, are you very satisﬁed, fairly satisﬁed, not
very satisﬁed, or not at all satisﬁed with the way
democracy works in [COUNTRY]?
1 (’very satisﬁed’)
- 5 (’not at all’)
Satisfaction w. Government ESS Now thinking about the [country] government, how satisﬁed
are you with the way it is doing its job? 0 - 10
0 - 10
Political Cynicism CSES
Some people say that it doesn’t make any diﬀerence who is in
power. Others say that it makes a big diﬀerence who is in power.
Using the scale on this card, (where ONE means that it
doesn’t make any diﬀerence who is in power and FIVE means
that it makes a big diﬀerence who is in power), where would
you place yourself?
Some people say that no matter who people vote for, it won’t
make any diﬀerence to what happens. Others say that who
people vote for can make a big diﬀerence to what happens.
Using the scale on this card, (where ONE means that voting
won’t make any diﬀerence to what happens and FIVE means that
voting can make a big diﬀerence), where would you place
1 (’No diﬀerence’)
- 5 (’big diﬀerence’)
Table 5: Political support variables in the diﬀerent surveys.
Figure 11: ESS integrated ﬁle ﬁeldwork dates
Country Election Party Vote Share Threshold
AT 2006-10-01 BZ ¨
O 4.110 4
AT 2013-09-29 FRANK 5.730 4
BG 2009-07-05 GERB 39.700 4
CZ 2006-06-02 Greens 6.290 5
CZ 2010-05-28 TOP 09 16.700 5
CZ 2013-10-25 ANO 11 18.770 5
DE 2005-09-18 NPD 1.600 5
DE 2009-09-27 Piraten 2 5
DE 2013-09-22 AfD 4.700 5
DK 2005-02-08 Centre Democrats 1 2
DK 2007-11-13 New Alliance 2.800 2
DK 2011-11-15 Christian Democrats 0.800 2
EE 2015-03-01 Free Party 8.700 5
HU 2006-04-09 MDF 5 5
HU 2010-04-11 Jobbik 16.400 5
HU 2014-04-06 Munkaspart 0.560 5
IS 2016-10-29 Reform 10.480 5
NL 2010-06-09 Trots 0.600 0.67
NO 2005-09-11 Red Electoral Alliance 1.200 4
NO 2009-09-13 Red Party 1.300 4
PL 2005-09-25 SD 3.900 5
PL 2007-10-21 Polish Labour Party 1.500 5
PL 2011-10-09 RP 10.020 5
PL 2015-10-25 Kukiz 15 8.810 5
SE 2006-09-17 Sweden Democrats 2.930 4
SE 2010-09-19 Sweden Democrats 5.700 4
SI 2008-09-21 Zares 9.370 4
SI 2011-12-04 Positive Slovenia 28.510 4
SI 2014-07-13 SMC 34.490 4
SK 2010-06-12 Freedom and Solidarity 12.140 5
SK 2012-03-10 SNS 4.550 5
UA 2007-09-30 Lytyn Bloc 3.970 3
UA 2012-10-28 UDAR 13.970 5
Table 6: Cases included: ESS
Country Election Party Vote Share Threshold
AUT 2013-09-29 Team Stronach 5.70 4
BGR 2014-10-05 Reformatorski Blok 8.89 4
CZE 2002-06-15 SNK 2.78 5
CZE 2006-06-03 Strana zelenych 6.29 5
CZE 2010-05-29 TOP 09 16.70 5
CZE 2013-10-25 ANO 2011 18.65 5
DEU 2002-09-22 Die Republikaner 0.60 5
DEU 2009-09-27 Piratenpartei Deutschland 2 5
DEU 2013-09-22 Alternative f¨ur Deutschland 4.70 5
DNK 2001-11-20 Independents 0 2
DNK 2007-11-13 Ny-Liberal Alliance 2.80 2
GRC 2012-06-17 Dimiourgia Xana 1.59 3
GRC 2015-09-20 Enosi Kentroon 3.43 3
ISL 2003-05-10 New Force 0.98 5
ISL 2007-05-12 Islandshreyﬁngin - lifandi land 3.30 5
ISL 2009-04-25 Borgarahreyﬁngin - Hreyﬁngin 7.20 5
ISL 2013-04-27 Bjort framtjo 8.25 5
ISR 2003-01-28 Ale Yarok 1.20 1.50
ISR 2006-03-28 Kadima 22 2
ISR 2013-01-22 Yesh Atid 14.33 2
MEX 2000-07-02 Democracia Social 1.90 2
MEX 2003-06-06 Convergence 2.30 2
MEX 2006-07-02 New Alliance Party 4.70 2
MEX 2009-07-05 NA 0 2
MEX 2012-07-01 NA 0.11 2
MEX 2015-07-07 Social Encounter Party 3.32 2
NLD 2006-11-22 Partij voor de Vrijheid 5.89 0.67
NLD 2010-06-09 Trots 0.56 0.67
NOR 2001-09-10 Rod Valgallianse 1.20 4
NOR 2005-09-12 Rod Valgallianse 1.20 4
NOR 2009-09-14 Rod Valgallianse 1.40 4
NZL 2002-07-27 United Future New Zealand 6.69 5
NZL 2008-11-08 NA 0 5
NZL 2011-11-26 New Zealand First Party 6.59 5
NZL 2014-09-20 Conservative Party of New Zealand 3.97 5
POL 2001-09-23 Platforma Obywatelska 12.70 5
POL 2007-10-19 Lewica i Demokraci 13.15 5
POL 2011-10-09 Ruch Palikota 10.02 5
ROU 2004-11-28 PNG 2.20 5
ROU 2012-12-09 Partidul Poporului - Dan Diaconescu 13.99 5
SVK 2016-03-06 Slovenska narodna strana 8.64 5
SVN 2004-10-03 Aktivna Slovenija 3.97 4
SVN 2008-09-21 Zares 9.37 4
SVN 2011-12-04 Lista Zorana Jankovi - Pozitivna Slovenija 28.51 4
SWE 2002-09-15 Sverigedemokraterna 1.44 4
SWE 2006-09-17 Sverigedemokraterna 2.90 4
SWE 2014-09-14 Feministiskt initiativ 3.12 4
TUR 2015-06-07 Halklarn Demokratik Partisi 13.12 10
TWN 2001-12-01 NA 0 5
TWN 2012-01-14 NA 0 5
Table 7: Cases included: CSES
’NA’ indicates that the candidate/list had no party aﬃliation.