Content uploaded by Dominic Nyhuis
Author content
All content in this area was uploaded by Dominic Nyhuis on Jul 07, 2020
Content may be subject to copyright.
Political Science Research and Methods
http://journals.cambridge.org/RAM
Additional services for Political Science Research and Methods:
Email alerts: Click here
Subscriptions: Click here
Commercial reprints: Click here
Terms of use : Click here
Separating Candidate Valence and Proximity Voting: Determinants of Competitors’
Non-Policy Appeal
Dominic Nyhuis
Political Science Research and Methods / FirstView Article / February 2016, pp 1 - 17
DOI: 10.1017/psrm.2016.7, Published online: 17 February 2016
Link to this article: http://journals.cambridge.org/abstract_S2049847016000078
How to cite this article:
Dominic Nyhuis Separating Candidate Valence and Proximity Voting: Determinants of Competitors’
Non-Policy Appeal. Political Science Research and Methods, Available on CJO 2016 doi:10.1017/
psrm.2016.7
Request Permissions : Click here
Downloaded from http://journals.cambridge.org/RAM, IP address: 141.2.134.65 on 18 Feb 2016
Political Science Research and Methods Page 1 of 17
©The European Political Science Association, 2016 doi:10.1017/psrm.2016.7
Separating Candidate Valence and Proximity Voting:
Determinants of Competitors’Non-Policy Appeal*
DOMINIC NYHUIS
Previous scholarship has provided ample evidence that non-spatial considerations can
trump voters’policy preferences in candidate selections. The literature has been less suc-
cessful, however, in providing a sense of the factors that raise candidates’non-policy
appeal. Faced with the challenging task of separating policy and non-policy aspects of indivi-
dual vote choices, empirical research has frequently relied on shorthand measures like candi-
date incumbency. This paper separates the valence component from policy-based candidate
selections by explicitly supplying voters with information on the policy agreement between
themselves and their district candidates. Relying on the distinction between campaign valence
and character valence by Stone and Simas, it is shown that candidate valence is driven by can-
didate visibility in a party-dominated political system.
Electoral decisions are not exhaustively described by the ideological proximity between
voters and candidates. A Downsian-type model of party competition holds that parties or
candidates take positions in a policy space and that voters choose the alternative that is
ideologically closest to them (Downs 1957). Scholars have noted that this view does not provide
a comprehensive description of party competition—neither conceptually nor in terms of the
empirical implications’accuracy (Adams and Merrill 2003, 161–2). In an early critique of the
Downsian model, Stokes (1963) points out that there are numerous issues like the economic
prosperity of a country, where voters—by and large—do not diverge with respect to the desired
policy outcomes. Therefore, competitors do not campaign by staking out policy positions on
these scales. Instead, competition is driven by public perceptions of competence or actual
quality differentials (Aragones and Palfrey 2004; Buttice and Stone 2012).
Recent research including the quality aspect of political competition has typically assumed
that there are two factors that determine the electoral success of competitors, a policy- and
a non-policy-related component, i.e. ideological proximity and candidate valence. These
candidate evaluation factors are typically treated as orthogonal in the formal contributions on
the subject (Ashworth and Bueno de Mesquita 2009; Serra 2010). Empirically, however, it is
difficult to separate the two factors and to capture their relative importance for individual vote
choices due to the well-known rationalization effects when trying to capture candidate quality
and policy proximity from surveys (Conover and Feldman 1986; Bartels 2002a; Lebo and
Cassino 2007). Specifically, it is challenging to determine the familiarity of voters with
the policy profiles of candidates and, hence, whether they are able to correctly assess their
ideological proximity to the candidates.
* Dominic Nyhuis, Department of Political Science, Goethe University Frankfurt, Theodor-W.-Adorno-Platz 6,
60323 Frankfurt (dominic.nyhuis@soz.uni-frankfurt.de). The author is grateful for feedback on the ideas that are
put forth in this paper by Jens Brandenburg, Sean Carey, and Thomas Gschwend. The author is particularly
indebted to Martin Elff, Lukas Stötzer, and Steffen Zittlau for invaluable suggestions that have greatly improved
the paper. The author also likes to express the gratitude to the curators of the platform http://www.abgeordnet
enwatch.de, specifically Martin Reyher and Roman Ebener, for allowing the author to run the additional
questionnaire on their site.
In this paper, I propose to estimate a comprehensive indicator of the valences of candidates
competing in the German federal election of 2013 while controlling for the effect of ideological
proximity. This is done by explicitly familiarizing voters with the policy positions of district
candidates and making vote recommendations based on a comparison between the individual
policy preferences and the candidates’profiles. After informing voters about their correct vote
choice from a spatial perspective (cf. Lau and Redlawsk 1997; Lau, Andersen and Redlawsk
2008), voters are invited to indicate their prospective vote choice. The systematic residual
between the correct vote and the actual prospective vote choice provides a comprehensive
estimate of the candidates’valences. The research thus makes two important contributions. One,
it introduces a novel technique for estimating candidate valences and also provides a data set of
candidates’valence advantages in the German federal election of 2013. Two, the research
investigates determinants of variations in the size of candidate valences.
The remainder of this article begins by briefly outlining the valence factor of vote choices as
well as components of candidate valence. Third section describes the data and method that is
applied in the empirical investigation. The association of the explicit valence indicators and
candidate status is the subject of fourth section. Fifth section provides some additional
robustness checks. Sixth section concludes.
THE VALENCE COMPONENT OF VOTE CHOICES
In his seminal critique of the dimensional model of politics, Stokes (1963) argues that numerous
issues structure political competition where neither voters nor political actors differ with regard
to preferred outcomes (cf. Stokes 1992). Faced with such valence-issues, political competition
shifts to perceptions of competence or the ability to deliver upon promises to create the
universally desired outcomes. Since the proposition of the valence concept by Stokes, numerous
contributions have been put forth that investigate how non-policy factors are related to
candidate behavior and electoral success. Several formal contributions on the subject have
introduced valence in Downsian-type models to explore variation in candidate position-taking
induced by valence differentials (Groseclose 2001; Aragones 2002; Schofield 2004; Schofield
2007; Ashworth and Bueno de Mesquita 2009; Hummel 2010; Bernahrdt, Camara and
Squintani 2011) and to derive equilibria in multi-party or multi-dimensional competition set-
tings (Ansolabehere and Snyder 2000; Schofield 2003).
Empirical contributions on the valence theory of politics have investigated how specific non-
policy characteristics of candidates are related to electoral outcomes. The most well-known
factor that alters vote choices is the incumbency status of the candidate (Cox and Katz 1996;
Berry, Berkman and Schneiderman 2000; Burden 2004; Hogan 2008; Stone et al. 2010). But
candidate valence is a multi-dimensional concept that has been taken to mean a variety of things
in the literature, such as “incumbency, greater campaign funds, better name recognition,
superior charisma, superior intelligence”(Groseclose 2001, 862). A useful distinction to classify
aspects of candidate valence was put forth by Stone and Simas (2010). The authors assert that it
is possible to differentiate between campaign valence and personal character valence. While
the former encompasses characteristics like name recognition and campaign funds, personal
character valence covers factors such as “integrity, competence, and dedication to public
service”(Stone and Simas 2010, 373).
1
1
A similar argument was proposed by Adams et al. (2011) who differentiate between strategic valence and
character valence. Although different in name, the two dimensions of valence refer to the same aspects as the
campaign valence and personal character valence in the sense of Stone and Simas (2010).
2NYHUIS
The formal contributions on the valence theory of voting have dealt with the
multi-dimensionality of the concept by defining comprehensive valence parameters to capture
the range of factors contained in the concept. Conversely, the empirical literature has struggled
to come up with comprehensive measures of candidate valence. The challenge of generating
comprehensive empirical estimates of valence is related to the difficulty of capturing valence
from survey evidence. Faced with systematic distortions in the candidates’perceived policy
profile in survey data (Conover and Feldman 1986; Granberg and Brown 1992; Merrill,
Grofman and Adams 2001), it is unclear to what degree proximity voting and candidate valence
drive individual vote choices.
Consequently, the empirical research has frequently relied on shorthand indicators for can-
didate valence such as the incumbency status of candidates. While the research has consistently
shown that this indicator is systematically and positively related to electoral outcomes, it cannot
provide a comprehensive image of the valence advantage of candidates. Moreover, it has the
implausible feature to not produce variation in valence among the group of incumbents. One
alternative measure to estimate the effect of valence on electoral outcomes was recently
proposed by Clark (2009). In an extensive hand coding of several Western European countries,
the author records political events that are related to public perceptions of parties’“competence,
integrity and unity”(Clark 2009, 112). As expected, the measure is negatively related to
the vote share of parties in subsequent electoral cycles. While the work is a commendable
contribution, the effect of events on party images only provides evidence at the level of the
character valence in the terminology of Stone and Simas (2010).
2
Campaign valence factors of
incumbents like increased funds or access to staff should be largely unaffected by political
scandals. Another two recent contributions grapple with the difficulty of generating empirical
and comprehensive measures of candidate valence by employing expert ratings of candidate
valence. Stone and Simas (2010) investigate the long-standing proposition in the formal
literature that there are position-taking incentives of valence advantages while Buttice and Stone
(2012) show that candidate quality alters individual vote choices and that the size of the effect is
conditional on the position-taking of candidates.
The present research project intends to provide a comprehensive estimate of the valence
advantages of candidates for the German Bundestag. The subsequent investigation of the
measure assesses how it is related to determinants of candidate valence. Drawing on the
distinction between campaign valence and personal character valence by Stone and Simas
(2010), I suggest that campaign valence and candidate visibility in particular should be most
closely related to the estimated candidate valence. Stone and Simas propose that voters
intrinsically care only about the personal character valence of candidates, whereas campaign
valence is rather an accidental by-product of the nature of the competition. While the assertion
is plausible that voters care most about quality of character, elements such as integrity and
honesty are comparatively difficult to assess for voters. Conversely, factors that drive the
visibility of candidates should be more closely related to the comprehensive indicator as voters
need to be familiar with candidates in order to consider them viable electoral alternatives.
The dominant factor of campaign valence in the comprehensive valence measure is
particularly true for the German electoral system that is dominated by parties. In the two-tiered
German electoral system, a list vote determines the overall composition of the Bundestag,
although voters can personalize their choices by selecting a nominal district candidate (Saalfeld
2009). As the electoral law is governed by the proportional component and therefore by the
2
As Clark (2009) focuses on the images of parties rather than individual candidates, personal character
valence refers to perceptions of the character of collective actors in this case.
Separating Candidate Valence and Proximity Voting 3
behavior of parties, the nominal candidates are frequently relatively unknown quantities.
Consequently, the valence component of candidates should primarily be driven by the need
to make themselves visible (campaign valence) before hoping to showcase their integrity
(character valence). Stone and Simas suggest in their outline of the two dimensions of valence
that there is a systematic relationship between character valence and campaign valence where
candidates with a character valence advantage might be more successful in attracting campaign
funds (Stone and Simas 2010, 373). Conversely, it seems unlikely that there is a pronounced
effect of character valence on campaign valence in the German case where individual
candidacies are generally less dependent on third-party funds for their campaign activities.
For example, nominal candidates in the 2013 German federal election only covered about
20 percent of their campaign expenses with third-party donations on average (Rattinger et al.
2014). Put differently, political systems that are dominated by parties are unlikely to exhibit
substantial effects of personal character valence on either campaign valence or electoral
outcomes. In an extension of the arguments put forth by Stone and Simas (2010), I suggest that
it is possible to differentiate between candidate visibility at the national level and at the local
level. There are elements that increase the valence of candidates such as holding a governmental
office (national level) and elements that are more strictly tied to the district competition such as
the number of campaign appearances in the electoral district (local level).
A COMPREHENSIVE MEASURE OF CANDIDATE VALENCE
One of the principal challenges for investigating candidate valence from surveys is the difficulty
to assess whether voters are familiar with the policy positions of candidates. Candidate
placements on policy scales are subject to distortions and rationalizations (Bartels 1988;
Granberg and Brown 1992; Merrill, Grofman and Adams 2001), impeding a differentiation
between the valence and proximity component of vote choices. To circumvent this limitation,
this paper employs evidence from a vote advice application of German nominal candidates
during the 2013 federal election campaign. The online tool explicitly provided voters with
information on the policy positions of their district candidates and made vote recommendations
based on voters’policy preferences. Subsequently, voters were invited to indicate their
prospective vote choices and in numerous instances voters did not select the candidate that was
indicated as ideologically closest by the vote advice application. This sequence of the data
collection allows explicitly ruling out the possibility that voters’selections are driven by their
unawareness of the candidates’policy positions. Instead, candidate valence trumped proximity
considerations. In the remainder of this section, I sketch the data collection and outline
the generation of the valence estimates. This section closes with an overview of the
operationalizations of the variables that should be related to the valence parameters.
During the German federal election campaign of 2013, the curators of the platform (http://www.
abgeordnetenwatch.de) assembled a survey with a binary response format—the Kandidatencheck—
and invited all nominal candidates to participate. The questionnaire covered both core issues of the
electoral campaign (e.g., the retirement age, austerity policies, and internet surveillance) as well as
long-term concerns of the German political system (e.g., minimum wages, temporary staff policies,
and renewable energies). Moreover, the items were selected to cover most major policy areas and
the two ideological dimensions that structure electoral competition in Germany—an economic
left-right dimension on the one hand and a cultural dimension on the other (Bräuninger and
Debus 2008; Pappi and Brandenburg 2009). To alleviate concerns regarding potential effects of
biased issue selection on the vote recommendations, the items were assembled by the platform’s
4NYHUIS
curators in collaboration with the political editors of those public and private media outlets that
featured the survey on their web appearances to generate publicity for the platform and to inform
voters about the available alternatives.
3
The candidate responses were made publicly accessible in the form of a vote advice
application in the weeks before the election (Cedroni and Garzia 2010; Schultze and Marschall
2012), allowing voters to log on to the platform and take an identical survey. After inputting
their responses, the system calculated an agreement score between the voters and their nominal
candidates and output the list of candidates—ordered by the proximity of the stated policy
preferences.
4
Having been provided a vote recommendation, users of the platform were invited
to respond to a supplemental questionnaire. The additional questionnaire collected, inter alia,
information on the users’prospective vote choices in the upcoming federal election.
5
To estimate the candidate valences, I model the individual vote choices and control for policy
distance as provided by the vote advice application, thus capturing the valences in candidate-specific
residuals. First, the distance between the users’policy preferences and the candidates’policy profiles
is calculated for each user.
6
The choice for one of the five district candidates is then included in a
mixed conditional logit model where the policy distance is the main explanatory variable.
7
Ifurther
include party fixed effects in order to control for selections that are driven by party label rather than
candidate-specific factors.
8
The model has the following form—the vote choice of individual ifor
competitor jfrom the choice set S
i
of the five district competitors is given as
πij =
expðηij Þ
P
k2Si
expðηik Þ;
and
ηij =βdistance distanceij +βspd spdj+βgruene gruenej+βfdp fdpj+βlinke linkej+Uj;
where U
j
are random effects with a normal distribution, estimated to be equal for all individuals faced
with the same choice set. The candidate-specific random effects are treated as estimates of the
comprehensive candidate valence.
9
There are two caveats regarding the underlying data. One, the
platform has not provided users with the option of weighting the issues before calculating the policy
distances. The model applied by the vote advice application therefore effectively assumes that every
3
The information in this paragraph is partly based on personal communication with the curators of the platform.
4
The calculation of the agreement scores relies on an unweighted proximity model (Wagner and Ruusuvirta 2012).
5
Approximately 36,000 users of the Kandidatencheck also provided information on the supplemental
questionnaire.
6
As there were 24 items in the survey, the agreement scores could range between 0 and 24. Empirically, the
differences range between 0 and 23.5; neutral selections were calculated as a difference of 0.5.
7
I make the simplifying assumption that only the five competitors from the parties with a parliamentary
representation at the time of the election are a viable alternative for the district vote. In reality, there are more
nominal candidates from fringe parties and occasionally there are even non-party competitors. Voters that
selected a fringe party competitor were discarded from the data set.
8
I do not model additional socio-demographic indicators which are known to influence vote choices, e.g.
denomination (Elff and Roßteutscher 2009; Ackermann and Traunmüller 2014), as the interest of this contribution is
not the accuracy of the individual candidate selection, but rather the aggregate deviations from recommended vote
choices. This is to say that voter-specific idiosyncracies are of little consequence as long as they do not systematically
favor candidates from particular parties. To control for such factors, the model contains party-specificfixed effects.
Moreover, I do not control for party proximity as an additional variable to potentially shape individual vote choices as
candidates were supplied model responses from party headquarters to ensure a consistent party image. Therefore,
candidate responses and party preferences are, on the whole, well aligned.
9
Estimations were performed using version 0.3-1 of the mclogit package in R. A comprehensive introduction
to the model is provided by Elff (2009).
Separating Candidate Valence and Proximity Voting 5
agreement/disagreement is equally important, whereas in fact the influence of issues on vote choice
might vary by voter. The effect of unweighted vote recommendations should be that voters more
frequently select candidates who were not labeled as ideologically closest by the platform. By the
same token, voters might have based their candidate selections on additional policy considerations if
they were irritated by the vote recommendation. To be sure, the ideal solution for this imprecision in
the estimates would have been to offer voters the possibility to weigh the issues. However, absent this
state of affairs, it seems unlikely that making unweighted vote recommendations would greatly affect
the valence estimates. First of all, the unweighted policy distances should be strongly related to the
weighted policy distances, thus creating imprecision only in edge cases. What is more, it is plausible
to assume that non-weighted recommendations do not systematically favor one candidate over
another, i.e. if the estimated candidate valences were elevated, they would be elevated across the
board. Put differently, it would be of greater concern if voters systematically considered aspects in
their candidate selections that favored candidates from one party over those from another. For
example, it is plausible that a subset of voters would be more inclined to select candidates from one
of the two main parties, regardless of policy proximity. This effect should, however, be controlled for
by the party-specificfixed effects.
A second concern regarding the underlying data is related to the non-random sample of vote
advice application users. As Marschall (2014) shows in his contribution on the subject, there is
clear evidence that users’socio-demographic statuses typically do not match the population
baseline. The most concerning factor among the various deviations from the baseline is a higher
average educational attainment (Marschall 2014, 100–1). It thus needs to be considered whether
the self-selected user sample might bias the estimated candidate valences. Previous research has
shown that better educated voters are more likely to take personal characteristics of candidates into
account when making vote choices (Glass 1985; Lau 1986; Miller, Wattenberg and Malanchuk
1986). It can thus be assumed that more voters in the sample apply valence-based voting compared
with an unbiased sample. However, as this contribution is interested in candidate-specific valences
rather than individual vote choices, there is no reason to expect biases due to an over-educated
sample as there is no fixed underlying metric of the valence estimates. More frequent valence
voting, therefore, still generates systematic variation in the valence indicator.
There were 299 electoral districts in the German federal election of 2013. For the estimation
of the indicator I drop all districts where one of the five competitors has not participated in the
Kandidatencheck as voters need to be familiarized with the policy positions of all viable
candidates in order for the argument to hold.
10
Despite a sample size of ~36,000 respondents,
there is fairly little data for each electoral district. Consequently, I discard all districts with a
coverage of fewer than 50 respondents overall. Both restrictions—a full candidate sample and at
least 50 observations—limit the data set to 114 electoral districts and 570 nominal candidates.
Table 1 presents the estimates from the candidate selection model. In line with expectations,
the policy distance is negatively and significantly associated with the individual vote choice.
The party fixed effects are also highly significant and in line with prior expectations. The
Christian Democrats (baseline) have the highest party valence, the Social Democrats a little less
so, the three smaller parties are trailing behind. The majoritarian electoral system on the district
tier generates a strong tendency for two parties—the CDU/CSU and SPD—to dominate the
district vote. Even though Die Linke has won several East German electoral districts in recent
10
The overall participation rate of candidates is ~90 percent, thus discarding districts with incomplete
candidate coverage does not thin the data set dramatically. Nevertheless, the candidates are not missing at
random. In particular, several high-ranking members of government are missing which might affect the results in
Explaining Valence Advantages of German Candidates section.
6NYHUIS
years, this is not evident in the party fixed effects, possibly suggesting that the electoral results
of successful Die Linke candidates are more strictly tied to candidate quality.
Figure 1 displays the distribution of the valence indicators—the candidate-specific random
effects U
j
. The candidates with the four highest and four lowest values are printed at the top of
the figure. Both Stefan Liebich (Die Linke) and Gesine Lötzsch (Die Linke), party chairwoman
in the years 2010–2012, have been able to win district mandates in previous electoral cycles—
Lötzsch in the years 2002, 2005, and 2009, Liebich in 2009—despite their membership in one
of the minor competing parties. Their successful nominal campaigns and the position of party
chairwoman for Gesine Lötzsch suggest a valence surplus due to visibility that is reflected in the
estimates and in their subsequent successful electoral bids in the federal election of 2013.
Similarly, Steffen Bockhahn (Die Linke) was party chairman of the state-level party
organization in Mecklenburg-Vorpommern in the years 2009–2012 and has won the mandate in
the district Rostock—Landkreis Rostock II in the federal election of 2009. In addition,
TABLE 1Conditional Logit Model: Candidate Selection
Distance −0.228***
(0.004)
SPD −0.409***
(0.061)
Grüne −0.995***
(0.063)
FDP −1.611***
(0.074)
Die Linke −1.667***
(0.066)
Observations 11,874
Deviance 27,947
Note:*p<0.10, **p <0.05, ***p <0.01.
−1.5 −1.0 −0.5 0.0 0.5 1.0 1.5
0.0
0.5
1.0
1.5
Estimated candidate valence
Density
Ströbele (GRÜNE) +0.99
Bockhahn (LINKE) +0.92
Liebich (LINKE) +0.92
Lötzsch (LINKE) +0.87
Vogel (SPD) −0.71
Vogt (SPD) −0.64
Müller (CDUCSU) −0.64
Volz (SPD) −0.63
Fig. 1. Distribution of candidate valence parameters U
j
from the Conditional Logit Model in Table 1. The
candidates with the four highest and four lowest values are shown at the top of the figure.
Separating Candidate Valence and Proximity Voting 7
Bockhahn was party chairman in the municipal council of Rostock. Therefore, despite coming
in second in the race for the district mandate in 2013, he was nonetheless quite visible in the
district which is plausibly echoed in a high valence indicator. Finally, Hans-Christian Ströbele
(Grüne) has also won a nominal mandate since 2002 as the first and only candidate from
Bündnis 90/Die Grünen, suggesting a similar high valence advantage for him.
Turning to the candidates with the lowest valence estimates, all four competitors have
achieved electoral results in the bottom quartiles of their respective parties.
11
Among the four
candidates, only Ute Vogt (SPD) was able to win a mandate for the Bundestag via the
state-level party list. In fact, after a particularly poor showing during the previous federal
election, where she came in third after the Christian Democrat and the Green candidate with a
mere 18.0 percent of the district vote in Stuttgart I, she resigned from the position of state-level
party chairwoman in Baden-Württemberg. The unusual state of affairs for a West German
Social Democrat candidate to come in third continued in 2013 when Ute Vogt was placed third
yet again with only 16.6 percent of the district vote. Götz Müller (CDU), the fourth candidate
with a particularly low valence indicator also had a plausibly poor showing in his district where
he scored an exceptionally low 13.7 percent of the votes. He was placed fourth, being surpassed
by all three competitors from the left-wing parties. This fourth place is even more astounding as
the 2013 federal election concluded with a comparatively strong result for the CDU due to the
popularity of chancellor Merkel, which gave a boost to most Christian Democratic nominal
candidates, prompting one news outlet to label Müller as “Merkel’s loneliest fighter.”
12
These
first, unsystematic observations provide some indication that the estimated valence parameters
do reflect the candidates’non-policy advantages and disadvantages.
It was argued that the visibility of candidates in the district and at the national stage should be
positively related to the valence parameters. Regarding national visibility, candidate valence should
increase if candidates have won the district mandate in the previous electoral cycle. Candidates can
also increase their visibility and hence their valence by holding an office at the national stage.
To capture this empirically, information on all governmental (Minister and parlamentarische
Staatssekretäre) and parliamentary offices (Parlamentspräsidium and Fraktionsvorstände)was
collected. As indicators of local candidate valence, the number of campaign events a candidate has
attended were selected as well as the number of personal posters a candidate has put up. Both
variables were gathered from self-reported information in the candidate survey that was conducted
by the project “Making electoral democracy work”(Blais 2010).
Character valence is more challenging to capture. The best available indicators of
character valence are the widely applied survey ratings of candidates’character traits (cf. Miller,
Wattenberg and Malanchuk 1986; Funk 1996; Bartels 2002b; Hayes 2005). Such trait ratings are
frequently collected for top candidates. However, as no ratings are available for the rank-and-file
candidates that ran in the German electoral campaign of 2013, I rely on three structured variables
instead that should be related to candidate character and perceptions of character valence.
One aspect that features prominently in German candidates’self-presentation are academic titles.
Most candidates with a PhD will display their title on campaign posters to indicate superior
intelligence and competence (Manow and Flemming 2011). Accordingly, data on whether can-
didates have a PhD was collected. Moreover, information was assembled on the candidates’
professions as indicated on the electoral ballot. As before, more prestigious professions might
11
Ines Vogel (SPD) received 14.9 percent of the district votes in Dresden I, Ute Vogt (SPD) 16.6 percent in
Stuttgart I, Götz Müller (CDU) 13.7 percent in Berlin-Friedrichshain—Kreuzberg—Prenzlauer Berg Ost, and
Tobias Volz (SPD) 19.2 percent in Konstanz.
12
http://www.sueddeutsche.de/politik/wahlkreis-atlas-merkels-einsamster-kaempfer-1.1744497 (March 7, 2015).
8NYHUIS
signal a greater electoral viability to voters. The information of the candidates’professions was
systematized by the Bundeswahlleiter
13
and put into a structured scheme of professions. Based on
this coding, all candidates were assigned a value of the job prestige that is associated with their
profession (Frietsch and Wirth 2001). One final aspect of candidate quality that will be included in
the analysis is physical attractiveness, which has consistently shown to be associated with can-
didate selections (Rosar and Klein 2005; Rosar and Klein 2013; Rosar and Klein 2014).
14
Table A1 in the Appendix provides summary statistics of the independent variables.
EXPLAINING VALENCE ADVANTAGES OF GERMAN CANDIDATES
The estimated valence indicator should be systematically associated with the proposed
components of candidate valence. Table 2 presents the results from three ordinary least square
models that regress the factors on the estimated valence parameters. Model 1 contains indicators
of candidate visibility at the national stage—the candidates’incumbency status and whether or
not candidates were office-holders in the previous electoral cycle. In line with expectations, the
incumbency status is positively and significantly associated with the estimated candidate
valence. The same is true for office-holders. The size of the office-holder variable is surprisingly
small, however, which might be explained by missing observations in this variable. Some of the
most well-known office-holders have not participated in the Kandidatencheck, such that several
incumbent ministers are missing in the data set. Therefore, the office variable contains
comparatively many office-holders with somewhat minor offices.
Model 2 considers indicators of local candidate visibility—the logged number of campaign
events a candidate participated in as well as the logged number of personal campaign posters
candidates had in their electoral districts.
15
Both variables are positively and significantly
associated with the estimated candidate valence. The incumbency status is still positive in
model 2, but no longer meets the ordinary criterion for statistical significance. Needless to
say that particularly the number of personal campaign posters a candidate can put up is
systematically related to the incumbency status of candidates.
Model 3 presents the estimates from a model that regresses candidate character aspects on the
valence indicator, specifically whether or not candidates have a doctorate, their job prestige, and
physical attractiveness. While the title variable is plausibly signed positive, there is no
systematic evidence that either a doctorate or the candidates’job prestige is systematically
related to the estimated candidate valence. Conversely and in line with previous research, the
candidates’attractiveness is positively and significantly associated with the estimated valence
indicator. Nonetheless, the model fails to explain much of the variation in the valence
parameters. It thus needs to be concluded that—at least in the party-dominated German political
system—candidate visibility is the more important condition for candidates to appear a viable
electoral alternative to voters. Aspects of the candidate character, on the other hand, bear little
relation to electability.
13
The Bundeswahlleiter is the administrative head of the electoral process.
14
I am grateful to Ulrich Rosar and Markus Klein for providing a data set of candidates’attractiveness during
the 2013 federal election campaign. Incidentally, although physical attractiveness is a consistent non-policy
factor in candidate selections, it is somewhat at odds with the two dimensions of valence proposed by Stone and
Simas (2010). Nevertheless, it will be considered as an element of character valence here as physical attrac-
tiveness is a strictly personal attribute rather than a quality that stems from the campaign context.
15
Note that the number of observations shrinks drastically due to missing observations in the anonymous
candidate survey. The office-holder variable was discarded in model 2 due to an insufficient number of office-
holders that have participated in the anonymous candidate survey.
Separating Candidate Valence and Proximity Voting 9
NON-PROXIMITY AND NON-PARTY PREFERENCES
To capture candidate valences, a model was estimated that explains individual prospective vote
choices with the policy distance to the district candidates. The model contained party fixed
effects to control for the party label. It is possible to go beyond this model specification and
even control for voters’long-term party preferences. Table 3 presents the results from a model
similar to the one proposed in A Comprehensive Measure of Candidate Valence section with the
TABLE 2Ordinary Least Square Regressions: Distance-
Based Valence
(1) (2) (3)
National Local Character
District incumbent 0.185*** 0.112
(0.030) (0.084)
Office-holder 0.128***
(0.044)
log(Campaign events) 0.046**
(0.021)
log(Number of posters) 0.024***
(0.009)
Doctorate 0.042
(0.032)
Job prestige −0.000
(0.000)
Attractiveness 0.044***
(0.013)
Constant −0.118*** −0.433*** −0.073
(0.030) (0.121) (0.055)
Observations 570 163 537
Adjusted R
2
0.061 0.092 0.012
Residual SE 0.243 0.234 0.250
F-statistic 7.156*** 3.347*** 1.903
Note: Party fixed effects not displayed.
*p <0.10, **p <0.05, ***p <0.01.
TABLE 3Conditional Logit Model: Candidate Selection
(1) (2)
Distance −0.228*** −0.113***
(0.004) (0.004)
Feeling thermometer 0.351***
(0.005)
SPD −0.409*** −1.64***
(0.061) (0.051)
Grüne −0.995*** −0.746***
(0.063) (0.054)
FDP −1.611*** −1.111***
(0.074) (0.066)
Die Linke −1.667*** −1.011***
(0.066) (0.058)
Observations 11,874 11,874
Deviance 27,947 22,430
Note:*p<0.10, **p <0.05, ***p <0.01.
10 NYHUIS
addition of party feeling thermometers. The first model in the table is identical to the model in
Table 1. It is reprinted for reference. Model 2 includes the feeling thermometers in addition to
the policy distance and the party fixed effects. In line with expectations, there is a positive and
highly significant association between the party feeling thermometer and the candidate
selection, above and beyond the party fixed effects. Nevertheless, the estimated candidate
valences in the two models are well aligned, as shown in the top left panel of Figure 2. The
figure shows the candidate-specific random effects from models 1 (x-axis) and 2 (y-axis) in
Table 3. Despite controlling for the party feeling thermometers in the latter model, the estimated
candidate valences are close to the 45ºline, suggesting no dependence of the estimates on
whether or not the additional control variable is included.
So far, the spatial component of the vote choice was controlled for by employing the distance
between the voter preferences and the candidates’policy profiles. One might argue that the
theoretical account requires modeling the actual vote recommendation rather than the policy
distance between candidates and voters. Therefore, a binary indicator was generated that is set
to 1 if a candidate was recommended by the platform and 0 otherwise. The two models from the
previous table were re-run, where the policy distance was replaced with the binary vote
recommendation. Table 4 presents the results from this analysis. Again, there is a strong and
positive effect of the vote recommendation on the candidate selection. Not surprisingly, the
parameter estimate shrinks when the party feeling thermometer is added in model 2. Figure 2
displays the candidate random effects from the four models in all possible combinations.
16
There is a high agreement between the estimated candidate valences, regardless of the model
specification. Thus, there is good evidence that the estimated parameters are not driven by the
specific modeling choice.
CONCLUSION
Voters do not always choose the candidate or party that is ideologically closest to them.
Scholars have provided ample evidence that non-policy factors such as the visibility, the
perceived integrity, or decisiveness of candidates can tilt vote choices in favor of candidates that
do not advocate the most compelling policy profile for individual voters. Despite the plausibility
of the proposition and the evidence to back it up, it has proven difficult to capture candidates’
valence advantages from voters directly in the form of surveys. On the one hand, this is due to
the uncertainty regarding the awareness of voters of the candidates’policy profiles and to the
well-known distortion effects in the minds of voters on the other.
This contribution has attempted to capture a comprehensive valence indicator of German
candidates from data generated by a vote advice application. By making vote recommendations
based on the overlap between individual policy preferences and the policy profile of district
candidates, it was possible to take proximity out of the equation and to capture the valence
advantages of candidates. I have provided evidence that the estimated valence indicator is
associated with ordinary measures of candidate valence, particularly variables that increase the
visibility of candidates are strongly associated with the proposed measure. Conversely, there is
little indication that character valence is associated with the estimated candidate valence.
As a party-dominated political system, Germany is a least likely case for discovering
character valence effects. Therefore, finding little systematic relation between the character
16
The alternative are the pure distance model (Table 1), the distance model with feeling thermometer (model 2
in Table 3), the vote recommendation model (model 1 in Table 4), and the vote recommendation model with feeling
thermometer (model 2 in Table 4).
Separating Candidate Valence and Proximity Voting 11
−0.5 0.0 0.5 1.0
0.0
Candidate valence (Distance)
−0.6
−0.4
−0.2
0.2
0.4
0.6
0.8
Candidate valence (Distance and party
feeling thermometer)
−0.5 0.0 0.5 1.0
−0.5
0.0
0.5
Candidate valence (Distance)
Candidate valence (Vote
recommendation)
−0.5 0.0 0.5 1.0
−0.4
−0.2
0.0
0.2
0.4
0.6
Candidate valence (Distance)
Candidate valence (Vote recommendation
and party feeling thermometer)
−0.6 −0.4 −0.2 0.0 0.2 0.4 0.6 0.8
−0.5
0.0
0.5
Candidate valence (Distance and feeling thermometer)
Candidate valence (Vote
recommendation)
−0.6 −0.4 −0.2 0.0 0.2 0.4 0.6 0.8
−0.4
−0.2
0.0
0.2
0.4
0.6
Candidate valence (Distance and feeling thermometer)
Candidate valence (Vote recommendation
and feeling thermometer)
−0.5 0.0 0.5
−0.4
−0.2
0.0
0.2
0.4
0.6
Candidate valence (Vote recommendation)
Candidate valence (Vote recommendation
and feeling thermometer)
(a) (b)
(c) (d)
(e) (f)
Fig. 2. Dependence of the estimated candidate valences on the various model specifications. The alternatives
are the Pure Distance Model (Table 1), the Distance Model With Feeling Thermometer (model 2 in Table 3),
the Vote Recommendation Model (model 1 in Table 4), and the Vote Recommendation Model With Feeling
Thermometer (model 2 in Table 4). The valence estimates are stable across the various model specifications.
12 NYHUIS
valence factors and the comprehensive candidate valence indicator is strictly tied to the specific
German electoral system with its strong non-majoritarian component—the mixed-member
proportional system (Shugart and Wattenberg 2001; Gschwend, Johnston and Pattie 2003;
Ferrara, Herron and Nishikawa 2005). The candidate as an independent political entity is of
much greater importance in the US context where the valence concept originates. Consequently,
the substantive results cannot generalize beyond the specific political system under investigation
here. This is all the more true, as more candidate-centered electoral systems should engender a
more systematic link between character valence and campaign valence (Stone and Simas 2010,
373), where candidates with a character valence advantage might be able to raise their
campaign valence. What should generalize, however, is the technique for estimating candidate
valences by explicitly supplying voters with information on the candidates’policy profiles.
As vote advice applications are an increasingly common tool to help citizens make informed
vote choices, scholars should take the opportunity to apply evidence from such tools in similar
analyses to capture the non-spatial component of voting.
One upshot of this research program is a data set on the valences of German nominal
candidates. The data are well suited to find applications beyond the substantive interest of the
present contribution. For instance, several publications have recently considered the behavioral
incentives of German candidates—both within the parliamentary assembly (Stratmann and Baur
2002; Sieberer 2010; Stoffel 2013; Sieberer 2015) and on the campaign trail (Wüst et al. 2006;
Zittel and Gschwend 2008). It seems likely that a high valence advantage would affect position-
taking in both cases as a higher valence translates to a greater electoral safety. Future research
should strive to combine the research interests of these literatures and assess how the proposed
valence parameters are related to candidate and MP behavior.
This article has considered prospective vote choices as an instrument for calculating the
valence advantages of German candidates. One aspect that has not been covered is the question
which voters are more likely to stray from the vote recommendations and to select candidates
with a valence advantage. Future research could apply similar evidence to consider factors that
increase the likelihood of valence voting. I would suggest that politically more sophisticated
voters are more likely to disregard the vote recommendations for two reasons. One, more
sophisticated voters have a better intuition of the valence advantages of candidates. It requires a
TABLE 4Conditional Logit Model: Candidate Selection
(1) (2)
Distance 1.161*** 0.637***
(0.020) (0.020)
Feeling thermometer 0.373***
(0.005)
SPD 0.489*** 0.244***
(0.057) (0.046)
Grüne −0.013 −0.292***
(0.058) (0.048)
FDP −1.553*** 1.058***
(0.072) (0.640)
Die Linke −0.683*** −0.522***
(0.061) (0.051)
Observations 11,874 11,874
Deviance 29,444 22,408
Note:*p<0.10, **p <0.05, ***p <0.01.
Separating Candidate Valence and Proximity Voting 13
minimum amount of political sophistication to make the connection between candidates
in an electoral race and their campaign appearances. Two, some previous research has
shown that more sophisticated voters are more likely to take personal characteristics of
candidates into account when casting their vote (Glass 1985; Lau 1986; Miller, Wattenberg
and Malanchuk 1986; for an opposing point of view see Pierce 1993). Glass (1985) argues
that sophisticated voters are aware that they cannot foresee the issues that will come up
throughout the electoral cycle. Therefore, their best option is to select a candidate with
commendable character traits which could be observed as higher levels of valence voting
by political sophisticates.
REFERENCES
Ackermann, Kathrin, and Richard Traunmüller. 2014. ‘Jenseits von Schwerkraft und Höllenfeuer:
Nicht-lineare Kontexteffekte auf den Zusammenhang von religiöser Gruppenzugehörigkeit
und individuellem Wahlverhalten bei fünf Bundestagswahlen’.Politische Vierteljahresschrift 55
(1):33–66.
Adams, James, and Samuel Merrill III. 2003. ‘Voter Turnout and Candidate Strategies in American
Elections’.Journal of Politics 65(1):161–89.
Adams, James, Samuel Merrill III, Elizabeth N. Simas, and Walter J. Stone. 2011. ‘When Candidates
Value Good Character. A Spatial Model with Applications to Congressional Elections’.Journal of
Politics 73(1):17–30.
Ansolabehere, Stehen, and James M. Snyder Jr. 2000. ‘Valence Politics and Equilibrium in Spatial
Election Models’.Public Choice 103(3/4):327–36.
Aragones, Enriqueta. 2002. ‘Mixed equilibrium in a Downsian model with a favoured candidate’.Journal
of Economic Theory 103(1):131–61.
Aragones, Enriqueta, and Thomas R. Palfrey. 2004. ‘The Effect of Candidate Quality on Electoral
Equilibrium. An Experimental Study’.American Political Science Review 98(1):77–90.
Ashworth, Scott, and Ethan Bueno de Mesquita. 2009. ‘Elections with Platform and Valence
Competition’.Games and Economic Behavior 67(1):191–216.
Bartels, Larry M. 1988. Presidential Primaries and the Dynamics of Public Choice. Princeton, NJ:
Princeton University Press.
Bartels, Larry M. 2002a, ‘Beyond the Running Tally. Partisan Bias in Political Perceptions’.Political
Behavior 24(2):117–50.
Bartels, Larry M. 2002b, ‘The Impact of Candidate Traits in American Presidential Elections’. In Anthony
King (ed.), Leaders’Personalities and the Outcome of Democratic Elections.44–69. Oxford, Great
Britain: Oxford University Press.
Bernhardt, Dan, Odilon Camara, and Francesco Squintani. 2011. ‘Competence and Ideology’.Review of
Economic Studies 78(2):487–522.
Berry, William D., Michael B. Berkman, and Stuart Schneiderman. 2000. ‘Legislative Professionalism and
Incumbent Reelection. The Development of Institutional Boundaries’.American Political Science
Review 94(4):859–74.
Blais, André. 2010. ‘Making Electoral Democracy Work’.Electoral Studies 29(1):169–70.
Bräuninger, Thomas, and Marc Debus. 2008. ‘Der Einfluss von Koalitionsaussagen, programmatischen
Standpunkten und der Bundespolitik auf die Regierungsbildung in den deutschen Ländern’.
Politische Vierteljahresschrift 49(2):309–38.
Burden, Barry C. 2004. ‘Candidate Positioning in US Congressional Elections’.British Journal of
Political Science 34(2):211–27.
Buttice, Matthew K., and Walter J. Stone. 2012. ‘Candidates Matter. Policy and Quality Differences in
Congressional Elections’.Journal of Politics 74(3):870–87.
Cedroni, Lorella, and Diego Garzia (eds) 2010. Voting Advice Applications in Europe. The State of the Art.
Naples, Italy: ScriptaWeb.
14 NYHUIS
Clark, Michael. 2009. ‘Valence and Electoral Outcomes in Western Europe, 1976–1998’.Electoral Studies
28(1):111–22.
Conover, Pamela J., and Stanley Feldman. 1986. ‘The Role of Inference in the Perception of Political
Candidates’. In Richard R. Lau and David O. Sears (eds), Political Cognition. 127–58. Hillsdale,
MI: Lawrence Erlbaum Associates.
Cox, Gary W., and Johnathan N. Katz. 1996. ‘Why Did the Incumbency Advantage in U.S. House
Elections Grow?’American Journal of Political Science 40(2):478–97.
Downs, Anthony. 1957. An Economic Theory of Democracy. Cambridge, Great Britain: Cambridge
University Press.
Elff, Martin. 2009. ‘Social Divisions, Party Positions, and Electoral Behavior’.Electoral Studies
28(2):297–308.
Elff, Martin, and Sigrid Roßteutscher. 2009. ‘Die Entwicklung Sozialer Konfliktlinien in den Wahlen von
1994 bis 2005’. In Oscar W. Gabriel, Bernhard Weßels and Jürgen W. Falter (eds), Wahlen und
Wähler. Analysen aus Anlass der Bundestagswahl 2005. 307–25. Wiesbaden, Germany: VS Verlag
für Sozialwissenschaften.
Ferrara, Federico, Erik S. Herron, and Misa Nishikawa. 2005. Mixed Electoral Systems. Contamination
and its Consequences. New York, NY: Palgrave Macmillan.
Frietsch, Rainer, and Heike Wirth. 2001. ‘Die Übertragung der Magnitude-Prestigeskala von Wegener auf
die Klassifizierung der Berufe’.ZUMA-Nachrichten 48(25):139–63.
Funk, Carolyn L. 1996. ‘The Impact of Scandal on Candidate Evaluations. An Experimental Test of the
Role of Candidate Traits’.Political Behavior 18(1):1–24.
Glass, David P. 1985. ‘Evaluating Presidential Candidates. Who Focuses on Their Personal Attributes?’.
Public Opinion Quarterly 49(4):517–34.
Granberg, Donald, and Thad A. Brown. 1992. ‘The Perception of Ideological Distance’.Western Political
Quarterly 45(3):727–50.
Groseclose, Tim. 2001. ‘A Model of Candidate Location When One Candidate has a Valence Advantage’.
American Journal of Political Science 45(4):862–86.
Gschwend, Thomas, Ron Johnston, and Charles Pattie. 2003. ‘Split-Ticket Patterns in Mixed-Member
Proportional Election Systems. Estimates and Analyses of Their Spatial Variation at the German
Federal Election, 1998’.British Journal of Political Science 33(1):109–27.
Hayes, Danny. 2005. ‘Candidate Qualities Through a Partisan Lens. A Theory of Trait Ownership’.
American Journal of Political Science 49(4):908–23.
Hogan, Robert E. 2008. ‘Policy Responsiveness and Incumbent Reelection in State Legislatures’.American
Journal of Political Science 52(4):858–73.
Hummel, Patrick. 2010. ‘On the Nature of Equilibria in a Downsian Model with Candidate Valence’.
Games and Economic Behavior 70(2):425–45.
Lau, Richard R. 1986. ‘Political Schemata, Candidate Evaluations, and Voting Behavior’. In Richard R.
Lau and David O. Sears (eds), Political Cognition.95–126. Hillsdale, MI: Lawrence Earlbaum
Associates.
Lau, Richard R., and David P. Redlawsk. 1997. ‘Voting Correctly’.American Political Science Review
91(3):585–98.
Lau, Richard R., David Andersen, and David P. Redlawsk. 2008. ‘An exploration of correct
voting in recent U.S. Presidential elections’.American Journal of Political Science 52(2):
395–411.
Lebo, Matthew J., and Daniel Cassino. 2007. ‘The Aggregated Consequences of Motivated Reasoning and
the Dynamics of Partisan Presidential Approval’.Political Psychology 28(6):719–46.
Manow, Philip, and Peter Flemming. 2001. ‘Der Titel Als Politisches Distinktionsmerkmal.
Eine Untersuchung Akademischer und Adeliger Wahlbewerber Zum Deutschen Bundestag
1949–2009’.Zeitschrift für Politikwissenschaft 21(4):531–51.
Marschall, Stefan. 2014. ‘Profiling Users’. In Diego Garzia and Stefan Marschall (eds), Matching Voters
with Parties and Candidates. Voting Advice Applications in a Comparative Perspective.93–104.
Colchester, Great Britain: ECPR Press.
Separating Candidate Valence and Proximity Voting 15
Merrill, Samuel III, Bernard Grofman, and James Adams. 2001. ‘Assimilation and Contrast Effects in
Voter Projections of Party Locations. Evidence from Norway, France, and the USA’.European
Journal of Political Research 40(2):199–221.
Miller, Arthur H., Martin P. Wattenberg, and Oksana Malanchuk. 1986. ‘Schematic Assessments of
Presidential Candidates’.American Political Science Review 80(2):521–40.
Pappi, Franz U., and Jens Brandenburg. 2009. Wähler im Politikraum Des Deutschen Parteiensystems
1980 bis 2005. Working Paper 124. Mannheim, Germany: Mannheimer Zentrum für Europäische
Sozialforschung.
Pierce, Patrick A. 1993. ‘Political Sophistication and the Use of Candidate Traits in Candidate Evaluation’.
Political Psychology 14(1):21–35.
Rattinger, Hans, Sigrid Roßteutscher, Rüdiger Schmitt-Beck, Bernhard Weßels, and Christof Wolf.
2014. ‘Kandidatenstudie 2013, Befragung, Wahlergebnisse und Strukturdaten (GLES)’. GESIS
Datenarchiv, ZA5716 Datenfile Version 2.0.0, Köln, Germany.
Rosar, Ulrich, and Markus Klein. 2005. ‘Physische Attraktivität und Wahlerfolg. Eine empirische Analyse
am Beispiel der Wahlkreiskandidaten bei der Bundestagswahl 2002’.Politische Vierteljahresschrift
46(2):263–87.
Rosar, Ulrich, and Markus Klein. 2013. ‘Pretty Politicians. Die Physische Attrakvität von Spitzenkandidaten,
ihr Einfluss bei Wahlen und die These der Personalisierung des Wahlverhaltens’.InThorstenFaas,
Kai Arzheimer, Sigrid Roßteutscher and Bernhard Weßels (eds), Koalitionen, Kandidaten,
Kommunikation. Analysen zur Bundestagswahl 2009.149–70. Wiesbaden, Germany: VS Verlag für
Sozialwissenschaften.
Rosar, Ulrich, and Markus Klein. 2014. ‘The Physical Attractiveness of Front-Runners and Electoral Success.
An Empirical Analysis of the 2004 European Parliament Elections’.Zeitschrift Für Vergleichende
Politikwissenschaft 8(2):197–209.
Saalfeld, Thomas. 2009. ‘Germany: Stability and Strategy in a Mixed-Member Proportional System’.In
Michael Gallagher and Paul Mitchell (eds), The Politics of Electoral Systems. 209–29. Oxford:
Oxford University Press.
Schofield, Norman. 2003. ‘Valence Competition in the Spatial Stochastic Model’.Journal of Theoretical
Politics 15(4):371–83.
Schofield, Norman. 2004. ‘Equilibrium in the Spatial “Valence”Model of Politics’.Journal of Theoretical
Politics 16(4):447–81.
Schofield, Norman. 2007. ‘The Mean Voter Theorem. Necessary and Sufficient Conditions for Convergent
Equilibrium’.Review of Economic Studies 74(3):965–80.
Schultze, Martin, and Stefan Marschall. 2012. ‘Voting Advice Applications and Their Effect on Voter
Turnout. The Case of the German Wahl-O-Mat’.International Journal of Electronic Governance
5(3/4):349–66.
Serra, Gilles. 2010. ‘Polarization of What? A Model of Elections with Endogenous Valence’.Journal of
Politics 72(2):426–37.
Shugart, Matthew S., and Martin P. Wattenberg (eds) 2001. Mixed-Member Electoral Systems. The Best of
Both Worlds? Oxford, Great Britain: Oxford University Press.
Sieberer, Ulrich. 2010. ‘Behavioral Consequences of Mixed Electoral Systems. Deviating Voting Behavior
of District and List MPs in the German Bundestag’.Electoral Studies 29(3):484–96.
Sieberer, Ulrich. 2015. ‘Using MP Statements to Explain Voting Behavior in the German
Bundestag. An Individual Level Test of the Competing Principals Theory’.Party Politics
21(2):284–94.
Stoffel, Michael. 2013. ‘MPs with Competing Principals. A Theoretical and Empirical Contribution
to the Study of Representation in Mixed-Member Electoral Systems’. Dissertation, University of
Mannheim, Mannheim, Germany.
Stokes, Donald E. 1963. ‘Spatial Models of Party Competition’.American Political Science Review
57(2):368–77.
Stokes, Donald E. 1992. ‘Valence Politics’. In Dennis Kavanagh (ed.), Electoral Politics. 141–64. Oxford,
Great Britain: Clarendon Press.
16 NYHUIS
Stone, Walter J., and Elizabeth N. Simas. 2010. ‘Candidate Valence and Ideological Positions in U.S.
House Elections’.American Journal of Political Science 54(2):371–88.
Stone, Walter J., Sarah A. Fulton, Cherie D. Maestas, and L. Sandy Maisel. 2010. ‘Incumbency
Reconsidered. Prospects, Strategic Retirement, and Incumbent Quality in U.S. House Elections’.
Journal of Politics 72(1):178–90.
Stratmann, Thomas, and Martin Baur. 2002. ‘Plurality Rule, Proportional Representation, and the German
Bundestag. How Incentives to Pork-Barrel Differ Across Electoral Systems’.American Journal of
Political Science 46(3):506–14.
Wagner, Markus, and Outi Ruusuvirta. 2012. ‘Matching Voters to Parties. Voting Advice Applications and
Models of Party Choice’.Acta Politica 47(4):400–22.
Wüst, Andreas W., Herrmann Schmitt, Thomas Gschwend, and Thomas Zittel. 2006. ‘Candidates in the 2005
Bundestag Election. Mode of Candidacy, Campaigning and Issues’.German Politics 15(4):420–38.
Zittel, Thomas, and Thomas Gschwend. 2008. ‘Individualised Constituency Campaigns in
Mixed-Member Electoral Systems. Candidates in the 2005 German Elections’.West European Politics
31(5):978–1003.
APPENDIX
TABLE A1Summary Statistics of Independent Variables
District incumbent
Yes 94 0.16
No 476 0.84
Office-holder
Yes 34 0.06
No 536 0.94
log(Campaign events)
Mean 5.25
SD 2.54
Minimum 0
Maximum 8.70
log(Personal campaign posters)
Mean 3.72
SD 1.06
Minimum 0
Maximum 6.91
Doctorate
Yes 75 0.13
No 495 0.87
Job prestige
Mean 95.80
SD 22.68
Minimum 33.73
Maximum 188.35
Candidate attractiveness
Mean 1.95
SD 0.89
Minimum 0.21
Maximum 5.33
Separating Candidate Valence and Proximity Voting 17
