Explaining Deviant Voting in Subnational Elections: The Case of Local Council Elections in Germany

Abstract

Elections in multi-level systems are challenging for voters because they are supposed to hold political representatives accountable in a complex system of responsibilities. Therefore, when casting their votes in elections at different levels of government, level-specific factors should be of relevance for voters when making voting decisions. We compare deviant voting behaviour in two specific types of local democracy, namely those where institutional and political factors cluster differently. While local candidate-oriented democracies are characterised by candidate-centred voting systems and local party systems with low degrees of nationalisation, local party-oriented democracies are characterised by closed list systems and local party systems that show high degrees of nationalisation. Using individual and aggregate data on local elections in Germany from 2014, we show that voters in local candidate-oriented democracies seem to be more affected by local factors when casting their votes. However, we also show that our results depend on the respective measurement strategies for deviant voting, as local vote intention in free list systems may be measured in alternative ways. The paper makes a significant contribution to the understanding of deviant and local voting behaviour in multi-level democracies with different voting systems.

Introduction

Elections in multi-level systems are challenging for voters because they are supposed to hold political representatives accountable in a complex system of responsibilities. On what basis do voters make their voting decisions in elections at the local level? According to the social-psychological model of electoral research, local voting behaviour is generally based on political attitudes (e.g. Campbell et al. 1960; Campbell, Gurin, and Miller 1954). Analogous to national elections, voters who identify with a local party and/or rate its candidates or their political content positively are going to be more likely to vote for it. In reality, case studies reveal some evidence for different voting behaviour at local and national elections (e.g. Mjelde et al. 2016; Quinlivan 2020). So why do voters deviate from their national voting intention when they cast their ballots in other – e.g. local – elections (deviant voting behaviour)? Until now, most studies looking at deviant voting in local elections are solely based on aggregate data (Bolgherini, Grimaldi, and Paparo 2021; Kjær and Steyvers 2019). These studies show deviant voting behaviour to different degrees. But they cannot explain why voters cast different votes in national and local elections. Additionally, the few studies trying to explain deviant voting behaviour by using survey data show limited results: Marien, Dassonneville, and Hooghe (2015) analysed individualised voting behaviour at local elections in Belgium and affiliate deviant voting to knowing local candidates and being concerned about local policies. Lidström (2021) goes one step further and distinguishes informed same-party voters from split ticket voters at local elections in Sweden. Split ticket voters have weaker party identifications in comparison to same-party voters, and generally support smaller parties.

Both studies look at individual factors explaining deviant voting in local elections. However, first it remains unclear whether deviant voting in second-order elections (Reif and Schmitt 1980) is due to (a) voters realising their ‘true’ national party preferences, (b) dissatisfaction with the national government, or (c) whether deviant voting is a result from policies, candidates, or parties from the second-order arena affecting the individual voting decisions. Second, both studies do not consider the impact of context factors when explaining deviant voting. Kjær and Steyvers (2019) link the logic of contextual factors to the second-order election concept and postulate that certain local elections, due to different institutional or political contexts, are more likely to correspond to the ideal type of a second-order election. They expect more localised voting behaviour in smaller municipalities and municipalities with less nationalised local party systems.

Third, papers only analyse deviant voting only where deviant voting is ‘easy’ to measure, i.e. when there is only one vote allowed in the national and the local elections. They do not answer the question of how to operationalise and measure deviant voting in contexts where electoral systems differ within a multi-level political system.

This is why we first focus on the question of how to measure deviant voting in subnational elections with different electoral systems, and second on the impact of local institutional and political factors affecting variations in the extent and the explanation of local deviant voting. As Kjær and Steyvers (2019) argue, the focus on national politics when explaining local voting behaviour leads to an undervaluation of local heterogeneity and different local contexts corresponding to different degrees of second-orderness. Second, we therefore ask why voters in local elections deviate from their national voting intention and to what extent this deviance depends on the institutional and political setting of local democracy. We assume that deviant voting in local elections varies according to the institutional setting of local democracy and the competitive nature of local policy-making, especially looking at the local electoral system (either more candidate, or more party centred), and the nationalisation of the local party system. To answer these research questions and contribute to the debate about deviant voting behaviour in local elections we analyse both aggregate and individual level data at the 2014 local elections across nine states in Germany. This specific research design allows us to examine the influence of two distinct types of local democracy on deviant voting in local elections. Exit polls conducted by infratest dimap also provide the rare opportunity to analyse municipal voting behaviour at the individual level (Neu and Konrad-Adenauer-Stiftung 2018). According to our results, the degree of deviant voting depends on the measurement of deviant voting behaviour. In addition, there is some evidence for local factors shaping deviant voting, in particular in candidate-oriented democracies.

This paper is structured as follows. First, we refer to the debate on the nationalisation and localisation of local electoral behaviour, followed by a review of the current state of research to develop our assumptions. We then introduce our research design and present our empirical results. Last, we discuss our results and implications for future research.

Theory

Voting behaviour in local elections still represents a blind spot in empirical research. When explaining voting behaviour in local elections, two explanatory approaches can be derived from the literature: the unitarian and federalist approach (Gschwend 2008).

Unitarists – on the one hand - focus on the national level when analysing local elections and attribute subnational voting behaviour to national factors (Gschwend 2008, 231–232). In this context, the second-order concept developed by Reif and Schmitt (1980) has become a main feature when studying elections in multi-level democracies. Federalists – on the other hand - analyse subnational factors like local parties, issues and candidates to explain subnational electoral behaviour (Gschwend 2008, 232–234).

In sum, there are three rationales explaining deviant voting behaviour in local elections. The first two rationales result from Reif and Schmitt’s ‘unitarist’ view assuming that there is ‘less at stake’ in SOEs. First, deviant voting may result from voters’ realising their ‘true’ party preferences: in first-order elections, voters preferring small national parties might strategically vote for larger parties because they do not want to waste their votes. When there is ‘less at stake’, however, they might realise their ‘true’ party preferences. Therefore, smaller parties should be more successful in second-order contests (e.g. Mjelde et al. 2016; Rallings and Thrasher 2005; Tsirbas 2015).

Second, because there is ‘less at stake’, voting in second-order contests might also become more expressive. Voters may be tempted to punish the governing parties at the national level by explicitly voting for opposition parties, smaller national parties, or even local lists in second-order contests. Some empirical studies even argue that local elections are a kind of referendum about national governments and national policies (e.g. Chiaramonte and D’Alimonte 2012; Esen and Gumuscu 2019; Quinlivan 2020). In sum, by expressing their dissatisfaction with the incumbent national government, the share of votes for nationally governing parties in the SOE should be lower than in the latest FOE. According to Reif and Schmitt (1980, 9–10), this ‘punishment’ effect results in deviant voting and it should be strongest in the mid-term of the national electoral cycle.

Third, local voting behaviour should not only be a result of national conditions but also of local institutional and political characteristics. There is some evidence of a substantial amount of so-called localised voting behaviour, e.g. local electoral behaviour resulting from local politics (Dodeigne et al. 2022). Other studies provide empirical evidence for retrospective voting and incumbent advantage in local elections. Dassonneville, Claes, and Lewis-Beck (2016) attribute the re-election of local officials to the local economic situation, among other factors. Burnett and Kogan (2017) find a correlation between the quality of local infrastructure and the electoral success of incumbents. Overall, both national and local factors seem to shape local voting behaviour. However, the extent of localised voting behaviour remains uncertain.

In the following, both the extent and the causes of deviant voting behaviour in local elections are examined. Deviant voting is not the same as split-ticket voting, vote switching or dual voting because we do not compare voting behaviour at local and national elections (Burden and Helmke 2009, 2). When we talk about deviant voting behaviour, we refer to the difference between the reported voting behaviour in a local election and the reported voting intention in a hypothetical national election next Sunday. If a voter’s local voting decision deviates from his or her current national voting intention, this constitutes deviant voting behaviour (Marien, Dassonneville, and Hooghe 2015).¹

Our basic assumption is that the type of local democracy influences the degree of deviance in local elections. Depending on the type of local democracy, the influence of national and local factors on local elections and voting behaviour should vary. Two types of local democracies have during the last decades developed across the German states: The type of local candidate-oriented and local party-oriented democracies (Holtkamp 2006; Vetter 2009).

In local candidate-oriented democracies the electoral system is more personalised than in local party-oriented democracies. The electoral system is based on free lists, where voters can distribute a fixed number of votes among several candidates even on different lists (Van der Kolk 2007). Election campaigns are accordingly dominated by candidates and not by parties. In contrast, local party-oriented democracies are characterised by strong party politicisation where closed list systems are used. Voters only have one vote and thus elect only one single party. The varying impact of parties and candidates also affects the style of policy-making within the local councils with a higher level of party cohesion in party-oriented democracies (Holtkamp 2006, 644–646). In party-oriented democracies, local party systems are significantly more nationalised and competitive, while in candidate-oriented democracies, local party systems are much less dominated by national parties and policy-making is said to be far less competitive (Holtkamp 2006; Kuhn and Vetter 2013).

We expect that deviant voting is more common in local candidate-oriented democracies with free list systems, where voters have multiple votes and can vote for both lists and single candidates even from different lists: First, as free list systems are more personalised than closed list systems, votes should be cast rather for local candidates than for a local party (Gabriel and Westle 2012, 318; Cryns and Hembach 1987, 126). Voters might therefore deviate because they know candidates personally. Second, in candidate-oriented democracies, the local party systems are also less nationalised. Voters might therefore be more inclined to deviate from their national preference depending on the attribution of local problem-solving competences to local lists or reason their voting behaviour with local candidates (Table 1).

Table 1. Types of local democracies.

	Candidate-Oriented Democracies	Party-Oriented Democracies
Voting systems	Free lists	Closed lists
Key actors of electoral campaigns	Candidates	Parties
Nationalisation of local party system	Low	high

Source: Holtkamp (2006) and Vetter (2009).

In party-oriented local democracies with closed lists, on the contrary, the electoral system allows voters only to cast their ballot for one list or party and not for individual local candidates. Additionally, the election campaigns in party-oriented local democracies are more concentrated on and dominated by parties (Holtkamp 2006). So, the influence of national parties on local voting behaviour in systems with closed lists should be stronger and deviant voting less likely.

Overall, there are several arguments in favour of less deviance in local voting behaviour in local party-oriented democracies than in local candidate-oriented democracies. Local elections in party-oriented democracies should be more biased towards national parties due to the dominant position of the national parties and the local electoral system. Local electoral behaviour should therefore be more party-oriented and more nationalised and local context factors should be less relevant when explaining local voting behaviour. On the contrary, in local candidate-oriented democracies, deviant local electoral behaviour is expected to be more common, and its causes should more likely be located at the local level. The expected observable implications are therefore:

1. Deviant voting is more common in local candidate-oriented than in local party-oriented democracies.

2. The more important local factors (primarily knowledge of candidates) are perceived by the voters, the more they should deviate in their voting decision. This effect should be stronger in candidate- than in party-oriented democracies.

3. Voters with a party identification should show a lower probability for deviant voting behaviour

Methods

Data and Research Design

In Germany, election days (and years) for municipal council elections generally differ from state to state to guarantee the independence of local elections from political trends at higher levels of government. In 2014, however, local council elections in nine German states (Baden-Württemberg, Brandenburg, Mecklenburg-West Pomerania, North Rhine-Westphalia, Rhineland-Palatinate, Saarland, Saxony, Saxony-Anhalt, Thuringia) were held the same day and coincided with elections to the European Parliament.² This opportunity to study local elections within a federal system based on individual-level data offers the possibility for a comparative approach, and to test the hypotheses outlined above as the types of local democracy differ between the states.

Voter registration in Germany is automatic for all elections. Therefore, the number of eligible and the number of registered voters is more or less the same and turnout at the aggregate level is measured as the percentage of eligible voters who cast their vote(s). The active right to vote in federal elections is restricted to people with German citizenship and the minimum voting age is 18. These regulations differ for local elections: German citizens and citizens of other EU member states are entitled to vote if they have lived in the respective municipality for a certain time, which varies from state to state, e.g. three months in Baden-Wuerttemberg (§12(1) GemO BW) or 16 days in North Rhine-Westphalia (§7 KWG NRW). Voting age differs as well. In most states voting age at local elections is 16, whereas in Rhineland-Palatinate, Saarland, Saxony, and Thuringia in 2014 was 18 (Holtmann, Rademacher, and Reiser 2017).

Local electoral systems in Germany vary from state to state although all states use a system of proportional representation (Vetter 2022). Minor exceptions exist especially in very small municipalities in Rhineland-Palatinate (Holtmann, Rademacher, and Reiser 2017, 95; Meyer 2007, 432). Local elections are usually held at large and there is no threshold (Korte 2016, 22). But local elections differ regarding the formula used to translate votes into seats: the D’Hondt method favouring bigger parties is used in Saxony and Saarland. The other states either follow the Hare/Niemeyer or the Sainte-Laguë/Schepers method.

For local elections in North Rhine-Westphalia, a system of personalised proportional representation with closed lists is used. Voters only may cast one vote for one party. It is impossible to vote for a specific candidate on a list. In the Saarland, elections are held according to the principle of proportional representation with closed lists (Korte 2016, 22).

In contrast to North Rhine-Westphalia and Saarland, the system of proportional representation with free lists is used in all other states in our sample. Candidates are selected from lists of participating parties and voters’ associations. Voters may either vote for a complete list (Frech 2018; Holtmann, Rademacher, and Reiser 2017; Van der Kolk 2007), or they may distribute their votes among the different lists, and one candidate can be given up to three votes. The number of votes is three in Brandenburg, Mecklenburg-West Pomerania, Saxony, Saxony-Anhalt, and Thuringia. In Baden-Wuerttemberg and Rhineland-Palatinate, the number of votes corresponds to the number of seats in the local council. Council sizes differ according to the size of the municipalities. For example, the number of votes in Baden-Wuerttemberg is eight in municipalities with fewer than 1,000 inhabitants and 60 in municipalities with more than 400,000 inhabitants. In Rhineland-Palatinate the number varies between 6 votes in municipalities with fewer than 300 inhabitants and 60 votes in municipalities with more than 150,000 inhabitants.

Different from most other studies on local elections, we will use micro and macro data to analyse deviant voting behaviour. The advantage of official aggregate data is that they are not subject to distortions due to false self-reporting by respondents (Ranney 1962), whereas individual data provide information on the influence of demographic and socio-economic characteristics (e.g. income, education, age) as well as subjective attitudes and perceptions on individual voting behaviour. Although the data does not allow for multi-level regression modelling, our models will give first insights into whether and how local context factors shape deviant voting in German local elections. The aggregate data were collected by the states’ statistical offices for the descriptive part of the paper regarding deviant voting in second-order elections. The dataset covers 6,501 municipalities across nine states and includes information on population size, electoral systems, and election results in the 2013 federal elections and the 2014 local elections.

Individual data on local voting behaviour in Germany is hardly existent. The local elections in 2014 were an exception. On behalf of the Konrad-Adenauer-Foundation, Infratest Dimap conducted exit polls for these elections providing important insights on voting behaviour at local elections including information about the individual voters’ socio-structural backgrounds (e.g. income, education, age) as well as their political attitudes (Neu and Konrad-Adenauer-Stiftung 2018). The Computer Assisted Telephone Interviews (CATI) were conducted between May 26 and June 10, 2014. The representative sample covers 2,514 respondents. Due to our selection of states, 1,354 respondents remain for the analyses.

Dependent Variables: Deviant Voting Behaviour at the Micro and Macro Level

Deviant Voting at the Macro Level (Volatility)

When operationalising and measuring deviant voting behaviour at the macro level, we use the Pedersen volatility index V (Pedersen 1979): For each municipality differences in the vote shares for the single parties between the respective local election in 2014 and the latest federal election in 2013 are calculated by $\mathrm{DVmacr}{o}_i={\displaystyle \frac{1}{2}\sum _{p=1}^n|V_{p,l}-V_{p,f}|}$where $\mathrm{DVmacro}$ is the degree of deviant voting, measured in municipality $i$; $V_{p,l}$ is the share of votes for party $p$ in the 2014 local election $l$; $V_{p,f}$ is the share of votes for party $p$ at the latest federal election $f$.

The index is then the sum of the absolute differences in vote shares for all parties divided by two. As data on vote shares in federal and local elections are only available for those parties represented in the 16th Bundestag (2009-2013: CDU, SPD, Left Party, FDP, Greens), vote shares for all other parties are summed up as vote shares for ‘Other Parties or Lists’.

Deviant Voting at the Micro Level (Deviant Voting)

When measuring deviant voting at the individual level, we follow Marien, Dassonneville, and Hooghe (2015) by comparing local voting behaviour and the current national voting intention. If a voter's local voting behaviour deviates from his or her national vote intention, this is considered as deviant voting behaviour.

For Germany, however, the operationalisation and measurement of deviant voting behaviour in local elections with individual level data is more complex than in the Belgian case. Because of the different electoral systems in our local candidate and party-oriented democracies, different measurement approaches are needed. This, in turn, has consequences for the comparability of our results. We will therefore present alternative measurement models for the comparison of deviant voting and interpret our results cautiously.

In North Rhine-Westphalia (NRW) and Saarland (SL), where voters cast their votes in local elections via closed lists with only one vote for one list available, we measure deviant voting when respondents deviate regarding the party, they had voted for in the 2014 local election (‘Which party or list did you vote for in the city council elections?’) and their vote intention regarding the next federal election (‘Which party would you vote for if there were general elections next Sunday?’)

For all other states, we develop two ways of measuring deviant voting in local elections, as until now deviant voting has never been studied in local candidate-oriented democracies with free list systems. In states with free list systems, where voters have more than one vote and where free party lists are more common, it is possible to allocate all votes to one list, or to distribute the votes among candidates from different lists (Van der Kolk 2007). Respondents in these states were asked whether they had chosen a complete list (‘Did you vote for candidates from different parties and lists, or did you give all votes to one party or list?’), which list got the most of their votes (‘Which party or list did you give the most votes?’), and which individual lists they had chosen, e.g. CDU, SPD. Additionally, the respondents were asked about their vote intention if there were federal elections. We operationalise deviant voting behaviour in both of the following measurement models as a dummy-variable with deviant voting being coded ( = 1) and non-deviant voting ( = 0):

1. Measurement model looking only at list voters: To enable comparability of deviant voting between free and closed list systems, we focus on the 48.8 per cent of voters who have voted for a complete list in the local election. Deviant voting behaviour ( = 1) is here measured when the voting decision deviates from the national voting intention.

2. Measurement model including all local voters: An alternative second measurement model is calculated, including all voters: Deviant voting behaviour ( = 1) here occurs when the allocation of the majority of votes (including the selection of a single list) deviates from the national voting intention.³

The first measurement model, on the one hand, allows comparing deviant voters in party- as well as in candidate-oriented local democracies. At the same time, however, more than half of the voters are excluded from the analysis, especially those who have allocated their votes to individual candidates and who might also have chosen candidates from different lists. It can therefore be assumed that these voters in particular voted on local political factors. Hence, deviant voting in candidate-oriented local democracies might be underestimated using this measurement model. On the other hand, when using the second measurement model, comparability of the deviance measures in both types of local democracy is challenged. But here we also include those voters, where we expect deviant voting to be more obvious than among only list voters. To cope with this theoretical and methodological challenge, we will present our analyses using both measurement models to test the robustness of our results.

Independent Variables

Our central independent variable is the type of local democracy. The dummy variable ‘Type of Local Democracy’ for the macro data used indicates whether a municipality is located in a state where local candidate-oriented democracies ( = 1) or a local party-oriented democracies ( = 0) dominate. At the individual level, the variable indicates whether a respondent lives in a local candidate-oriented ( = 1) or a local party-oriented democracy ( = 0).

To measure the local factors that might affect local voting decisions respondents were asked whether they know a member of the local council personally ( = 1), or not ( = 0). Following Marien, Dassonneville, and Hooghe (2015), we expect a correlation between personal acquaintance with local politicians and deviant voting behaviour. Furthermore, we test the influence of the perceived local problem-solving competence. The two dummy variables Local Problem-Solving Competence: Local Lists and Local Problem-Solving Competence: National Parties implies whether a voter ascribe problem solving competence in local politics to a local list⁴/national party ( = 1) or not ( = 0). To test the impact of the local type of democracy on the nationalisation of local elections we use interaction effects of the variable ‘Type of Local Democracy’ and these independent variables. Last, the dummy variable Party Identification indicates whether a respondent has a party identification, or not.

Due to the significant differences between candidate- and party-oriented democracies, it should also be examined whether the two types differ from each other in terms of the manifestations of the independent variables. For this reason, we analysed the correlation between the independent variables and the two types of democracies (see Table A2 in appendix). The correlations are generally weak. However, it is noticeable that municipalities in candidate-oriented democracies tend to be smaller than in party-oriented democracies.

Control Variables

There are some indications in the literature for a relationship regarding municipality size and local voting behaviour. First, in smaller municipalities, elections are less professionalised. The contacts between voters and the candidates are closer (e.g. Czarnecki 1992, 29; Kjær and Steyvers 2019). Residents are also more embedded in the local communities and therefore more exposed to local issues (Denters et al. 2014). So, voters should also be ‘more interested and knowledgeable about local politics, and base their decisions on parochial issues, whereas in larger democracies, voters are more likely to utilise heuristics to make decisions’ (Oliver, Ha, and Callen 2012, 21). Some authors argue accordingly that, especially in smaller municipalities, the influence of party identification is weaker (e.g. Holtmann, Rademacher, and Reiser 2017, 87). If it is easier for voters to develop a personal relationship with local candidates as well as being informed about local parties and issues in smaller municipalities, size of the municipality should also affect the impact of local factors like local candidates, local parties, and issues on local voting behaviour. The larger the municipality, the more likely local voting behaviour should correspond to voting behaviour in national elections with the impact of party identification increasing, and the impact of local factors decreasing. Different degrees in deviance might also be due to the parties and lists offered in the respective local party systems. Voters might change their voting behaviour in local elections because the preferred national party does not run in these elections. As local party systems in smaller municipalities are accordingly less nationalised, deviance in smaller municipalities should be higher than in larger municipalities (Kjær and Elklit 2010; Vetter and Kuhn 2013). Municipal size is measured via four dummy variables: 1. Rural Community, 2. Small Town, 3. Medium-Sized Town, and 4. Large City. The variables correspond to the number of inhabitants with rural communities having less than 5,000 inhabitants, small towns more than 5,000 but less than 20,000 inhabitants, medium-sized towns more than 20,000 but less than 100,000 inhabitants, and large cities more than 100,000 inhabitants. The categorisation is based on the definition provided by the Federal Institute for Research on Building, Urban Affairs, and Spatial Development (Bundesinstitut für Bau-, Stadt- und Raumforschung 2021). Other control variables are Political Interest, Age, Gender, and Education.⁵

Analysis

To test our hypotheses, we start with some descriptive macro level statistics followed by logistic regression models based on our micro level data to explain deviant voting behaviour in German local elections in different institutional contexts.

Descriptive Statistics

The analysis of the gains and losses of the national parties in the 2014 local elections compared to the 2013 federal elections seem to confirm the second-order-election thesis from Reif and Schmitt 1980 (see Tab. 2). The success of other parties and the failure of the national established parties – especially those parties governing in Berlin (at the time CDU and SPD) – varies from state to state: Official data at the macro level shows that the national parties’ vote shares are much lower in local elections than in federal elections, while the smaller parties’ vote shares are higher (see Tab. 2). In most states the incumbent national government parties – the Christian Democratic Party (CDU) and the Social Democratic Party (SPD) - got a smaller share of votes in local elections than in the 2013 federal election (Table 2).

Table 2. Differences in vote shares between local and national elections.

	CDU	SPD	Greens	FDP	Left	Other Parties	Volatility
Party-Oriented Democracies
North Rhine-Westphalia	−2.8	.4	2.3	.1	−3.3	3.4	12.8
Saarland	1.4	6.3	-.4	−1.2	−2.8	−3.3	12.7
Candidate- Oriented Democracies
Baden-Württemberg	−24.6	−6.7	−6.3	−4,6	−3.9	46.1	50.3
Brandenburg	−7.1	−25.3	−2.0	.1	−12.5	46.9	54.1
Mecklenburg-West Pomerania	−22.6	−10.6	−2.8	-.9	−14.4	51.3	58.5
Rhineland-Palatinate	−34.9	−17.9	−5.4	−4.7	−5.3	68.3	71.6
Saxony	−10.7	−6.9	−1.9	.1	−8.6	28.1	35.1
Saxony-Anhalt	−14.0	−6.0	−1.5	.1	−12.7	34.1	40.3
Thuringia	−21.8	−10.4	−3.4	−1.1	−17.8	54.5	67.7

Source: The numbers are based on official electoral data from the states’ statistical offices and the Federal Returning Office (2021).

But the results also provide evidence for different degrees of deviant voting in the German States. In Saarland and North-Rhine Westphalia – where the party-oriented model of local democracy prevails - deviant voting behaviour is substantially lower than in the states with a more candidate-oriented type of local democracy.

Now turning to our micro level analyses: We first look at our central dependent and independent variables in Table 3 descriptively: In party-oriented local democracies, nearly 40 per cent of all voters deviate in local elections from their national vote intention. Looking at deviant voting in local candidate-oriented democracies, the results are less clear than theoretically expected, depending on the two different ways of measuring deviant voting. If only the single list voters (measurement 1) are considered, the share of deviant voters in candidate-oriented democracies decreases to 30.5 per cent, which runs even counter to our expectations. However, when measuring deviant voting looking at all voters (measurement 2) the extent of deviant voting behaviour in candidate-oriented democracies strongly exceeds the share of deviant voters in party-oriented democracies with 53.7 per cent compared to 39.4 per cent of all voters. Obviously, the two different ways of measuring deviant voting behaviour in candidate-oriented democracies significantly affect the results concerning the amount of deviant voting and underline the importance of an analysis based on both measurements regarding the reliability of results. Our first hypothesis therefore seems only partially confirmed.

Table 3. Share of deviant voters.

Party-Oriented Democracies	Candidate-Oriented Democracies
	Measurement 1: Only List Voters	Measurement 2: All Voters
39.4	30.5	53.7

Source: Own presentation.

In sum, there is only some evidence for different deviant voting behaviour in the two types of local democracies. Looking at the aggregate data, in local candidate-oriented democracies deviant voting seems definitely more common. Looking at the individual level data, however, our expectations are only partly met, depending on the way deviant voting is measured in candidate-oriented local democracies. The question of measurement turns out to be of importance when looking at local electoral behaviour in free list systems with possibilities to share votes between lists. We will keep this in mind when taking the next step in explaining deviant voting in local elections, where we expected local factors playing a more prominent role when casting the local ballot in candidate-oriented local democracies.

Explaining Deviant Voting in Candidate and Party-oriented Democracies

To test the impact of local and national factors on local deviant voting, we run logistic regression models for voters in party- and candidate-oriented democracies.⁶

The results of model 1 and 5 in Table 4 confirm the previous results. The degree of deviant voting behaviour again depends on the respective measurement of deviant voting in candidate-oriented local democracies. According to our first measurement in model 1, the probability for deviant voting behaviour is 40.1 per cent in party-oriented democracies, and only 25.5 per cent in candidate-oriented democracies. If all voters are taken into account (measurement 2), the proportion of deviant voters in candidate-oriented democracies increases to 49.8 per cent, which surmounts deviant voting in party-oriented democracies by 9.4 percentage points.

Table 4. causes of deviant voting.

	Measurement 1: Only List Voters				Measurement 2: All Voters
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Type of Local Democracy	.510** (.192)	.501* (.283)	.448** (.198)	.446* (.337)	1.457** (.132)	1.457 (.196)	1.364* (.135)	.925 (.285)
Political Attitudes
Local Council Members	.832 (.169)	.826 (.198)	.829 (.170)	.835 (.170)	.788 (.133)	.788 (.191)	.788 (.133)	.787 (.133)
Local Problem-Solving Competence: Local Lists	2.498* (.376)	2.496* (.376)	.718 (.521)	2.565* (.381)	2.686** (.273)	2.686** (.273)	.874 (.503)	2.964** (.278)
Local Problem-Solving Competence: National Parties	.394** (.211)	.394** (.211)	.384** (.213)	.394** (.262)	.538** (.169)	.538** (.169)	.536** (.169)	.383** (.255)
PID	.299** (.171)	.300** (.172)	.298** (.173)	.298** (.171)	.338** (.132)	.338** (.132)	.339** (.132)	.336** (.132)
Interaction Effects
Type of local Democracy* Local Council Members		1.029 (.351)				1.054 (.251)
Type of local Democracy* Local Problem-Solving Competence: Local Lists			9.639** (.692)				4.277** (.553)
Type of local Democracy* Local Problem-Solving Competence: National Parties				1.205 (.384)				1.758 (.311)
Constant	24.692** (.576)	24.728** (.576)	25.170** (.576)	26.059** (.588)	21.616** (.463)	21.616** (.465)	21.956** (.464)	27.677** (.486)
Observations	947	947	947	947	1354	1354	1354	1354
Nagelkerke R²	0.23	0.23	0.23	0.23	0.20	0.20	0.21	0.20

Source: Neu/Konrad-Adenauer-Foundation (2018). For complete regression models see Table A3 in appendix. Type of Local Democracy: Dummy-Variable (0 = Party-Oriented Democracy, 1 = Candidate-Oriented Democracy).

Note: Logistic Regression with Odds Ratios. Standard errors are clustered by states.

**p < 0.01.

*p < 0.05.

Apart from the extent of deviant voting behaviour, the influences of the independent variables are similar in both measurements. Contrary to our assumptions and the paper by Marien, Dassonneville, and Hooghe (2015), voters do not deviate from their current national preference because they know candidates personally. Another indication of localised voting behaviour is the influence of the attribution of local problem-solving competences. According to our expectations voters should deviate from their national voting intention depending on the problem solving competences they attribute to national or local parties/lists. The results of both measurements do meet these assumptions. Trust in the problem-solving competence of local lists is associated with deviant voting behaviour. These voters have a 55.4 per cent probability for deviant voting behaviour, which is an increase of 22.2 percentage points. Based on measurement 2 the probability increases from 43.2–67.1 per cent.

At the same time, the attribution of local problem-solving competences to national parties reduces the probability of deviant voting behaviour from 52.2–30.1 (measurement 1) and 57.1–41.8 precent (measurement 2).

Across all German states, party identification is an important predictor of deviant voting behaviour. According to measurement 1, the probability for deviant voting decrease for voters with a party identification from 56.0–27.6 per cent, based on measurement 2 from 64.4–38.0 per cent. Consistent with previous studies (Marien, Dassonneville, and Hooghe 2015), party identification reduces the.

likelihood of deviant voting behaviour in local elections – independent of the respective measurement of deviant voting.

The interaction effects provide some evidence of a stronger nationalisation of local voting behaviour in party-oriented democracies. According to our expectation there is significant interaction between the type of local democracy and the ascription of local problem-solving competences to local lists. The odds ratio is 9.639 and 4.277, and the effect is therefore stronger in candidate-oriented than in party-oriented democracies. There is no difference between the two types of democracy regarding the influence of local problem-solving competences on national parties and the familiarity with local candidates.

Discussion

In complex multi-level political systems representatives from different levels of government should be held accountable according to level-specific factors affecting the respective voting behaviour. Having this in mind, we compared deviant local electoral behaviour (local vs. national electoral behaviour) in local candidate-oriented and local party-oriented democracies, asking whether there are institutional and political context factors that account for such a deviance. Our comparative analyses are based on aggregate and individual level data from local council elections in nine German federal states in 2014. We assumed (a) that in local candidate-oriented democracies deviant voting is more common than in local party-oriented democracies, that (b) local deviant voting is more affected by local than national factors in candidate-oriented local democracies, and (c) voters with a party identification are less willing to deviate from their current national vote intention.

Besides this theoretical interest we also contributed to the methodologically challenging question about how to measure deviant local voting behaviour in contexts with different local electoral systems: closed vs. free list systems in a comparable way.

Using aggregate data, we showed substantial degrees of deviant local voting in local elections compared to the latest federal election in 2013: About half of all voters deviated from their national preferences, with substantial variance in deviant voting across the different types of local democracy: Deviance at the macro level (volatility) is higher in local candidate-oriented democracies than in local party-oriented democracies.

Using individual level data, however, the results strongly depend on the measurement of deviant local voting behaviour: Using a restrictive measurement model of deviant voters (looking only at list voters) in local candidate-oriented democracies for better comparison with deviant voters in local party-oriented democracies, our hypotheses had to be rejected. Contrary, however, using a more comprehensive (but less comparable) measure including all voters, our results confirm the assumed relationship that different types of local democracy are associate with differences in deviant local voting behaviour: In candidate-oriented local democracies deviant voting is then more common than in party-oriented local democracies, showing that the methodological challenge of measurement affects the empirical results substantially: In our first measurement model for deviant voting more than half of the voters are excluded from the analysis: those voters who have allocated their votes to individual candidates/different lists. While this way of measuring deviant voting in free list systems (candidate-oriented democracies) is easily comparable to deviant voting in closed list systems (party-oriented democracies), the degree of deviant voting seems to underestimated. It therefore still remains a free theoretical and methodological question, of how to measure deviant electoral behaviour in different electoral systems in a comparable way. We suggest that further studies should be aware of this problem by applying alternative measures.

Additionally, there is some evidence for local political factors causing deviant voting behaviour in both types of democracy. Some voters deviate from their national preferences, because they ascribe local problem-solving competences to local lists. In addition, voters do not deviate because of the perceived competences of national parties. Party identification also reduces the probability for deviant voting behaviour. In candidate-oriented democracies the effect of ascribing local problem solving-competences to local lists is more significant and matches our expectations of a stronger localisation of deviant voting behaviour.

What do our findings as a whole mean for the second-order character of local elections? On the one hand, and in line with Kjær and Steyvers (2019), local political institutions and factors seem to influence the degree of nationalisation of local elections. In local party-oriented democracies, where the electoral system is concentrated on parties, local election campaigns are dominated by parties, local party systems are characterised by national parties, local voting behaviour seems less localised. At least the impact of local problem-solving competences is smaller in party-oriented democracies.

On the other hand, in candidate-oriented democracies, where the electoral system and the local election campaign are concentrated on candidates, and local lists play an important role within the local party systems. Overall, local elections in local party-oriented democracies appear to be more similar to second-order elections than elections in local candidate-oriented democracies, because deviant voting is more based on local politics.

Finally, further research might also have a closer look at the representational quality of local elections, as lower turnout and nationalised local voting behaviour might finally also result in a misrepresentation of local political preferences, thereby lowering the quality of local democracy. Extending the analyses to other countries could provide further insights into the extent and causes of deviant voting behaviour. In the context of the debate on the classification of elections as first- and second-order, a respective comparison of deviant voting in local and European elections could constitute an indication of different degrees of second-order.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Frederic Graeb

Frederic Graeb is research associate at the Institute of Social Sciences at the University of Stuttgart, Germany. His main fields of research are local elections and electoral systems.

Angelika Vetter

Angelika Vetter is Professor at the Institute of Social Sciences at the University of Stuttgart, Germany.

Notes

1 We explicitly do not study split-ticket voting, which usually refers to different voting behaviour in simultaneous elections.

2 Although local elections in 2014 also took place in the city state of Hamburg, these elections will not be included as city states (Berlin, Bremen, and Hamburg) have the responsibilities of cities and states at the same time, but local council responsibilities differ between city states and all the other states in the way that some municipal council functions in the city states are taken over by the state parliament. For this reason, the local elections in Hamburg are not included in our analyses.

3 For the results of our robustness checks see table A.3 in the appendix

4 Local lists run in local elections, but not in state, national or European elections (Otjes 2018).

5 For a descriptive overview of all independent variables see tab. A.1 in the appendix.

6 For the complete regressions models see table A.3 in the appendix.

Appendices

Table A1. Summary statistics for the dependent and independent variables.

Variables	Description	Min.	Max.	Mean	N
Dependent Variables
Deviant Voting	Voters who do not vote according to their national preference in the 2014 local elections	0	1	.49	1354
	Party-Oriented Democracies	0	1	.43	622
	Candidate-Oriented Democracies Measurement 1: Only List Voters Measurement 2: All Voters	0 0	1 1	.31 .54	325 732
Independent Variables
Candidate-Oriented Democracies	Respondents in a municipality with a candidate-oriented democracy	0	1	.55	1354
Party-Oriented Democracies	Respondents in a municipality with a party-oriented democracy	0	1	.45	1354
Local Council Members	‘Do you know one or more members of your city council personally?’ (Dummy-Variable)	0	1	.63	1354
Local Problem-Solving Competence: Local Lists	‘All in all, which party, which local lists or which voters’ association do you think is most likely to solve the most important political problems in your municipality? (Dummy-Variable; 1 = Local Lists, 0 = Other Parties, No Parties)	0	1	.10	1354
Local Problem-Solving Competence: National Parties	All in all, which party, which local lists or which voters’ association do you think is most likely to solve the most important political problems in your municipality?’ (Dummy-Variable; 1 = National Parties, 0 = Other Parties, No Parties)	0	1	.75	1354
National Problem-Solving Competence: National Parties	All in all, which party, which local lists or which voters’ association do you think is most likely to solve the most important political problems in Germany?’ (Dummy-Variable; 1 = National Parties)	0	1	.81	1354
Population Size
Rural Community	Respondents: Population < 5.000 (Dummy-Variable)	0	1	.12	1354
Small Town	Respondents: 5.000 ≤ Population < 10.000 (Dummy-Variable)	0	1	.25	1354
Medium Sized Town	Respondents: 10.000 ≤ Population < 100.000 (Dummy-Variable)	0	1	.34	1354
Large City	Respondents: Population ≥ 100.000 (Dummy-Variable)	0	1	.28	1354
Political Attitudes
Party Identification	Dummy-Variable (1 = Yes, 0 = No)	0	1	.69	1354
Political Interest	Five-point Likert-scale (4 = Very strong, 3 = Strong, 2 = Medium, 1 = Less strong, 0 = Not at all)	0	4	2.5	1354
Social Demographics
Age	Age in years	18	95	52.31	1354
Gender	Dummy-Variable (0 = male, 1 = female)	0	1	.51	1354
Education	Five-point scale	0	4	2.16	1354

Table A2. Corellation (PEARSON).

	Type of Local Democracy
Independent Variables
Knowledge of Local Council Members	.156**
Local Problem-Solving Competence: Local Lists	.164**
Local Problem-Solving Competence: National Parties	-.150**
National Problem-Solving Competence: National Parties	−0.040
Population Size
Rural Community	.308**
Small Town	.208**
Medium Sized Town	-.139**
Large City	-.256**
Political Attitudes
Party Identification	-.025
Political Interest	-.061*
Socio Demographics
Age	.022
Gender	.049
Education	.006

Source: Neu/Konrad-Adenauer-Foundation (2018).

Note: Type of Local Democracy (1 = candidate-oriented democracy; 0 = party-oriented democracy).

**p < 0.01.

*p < 0.05.

Table A3. Causes of deviant voting.

	Measurement 1: Only List Voters				Measurement 2: All Voters
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Type of Local Democracy	.510** (.192)	.501* (.283)	.448** (.198)	.446* (.337)	1.457** (.132)	1.457 (.196)	1.364* (.135)	.925 (.285)
Political Attitudes
Local Council Members	.832 (.169)	.826 (.198)	.829 (.170)	.835 (.170)	.788 (.133)	.788 (.191)	.788 (.133)	.787 (.133)
Local Problem-Solving Competence: Local Lists	2.498* (.376)	2.496* (.376)	.718 (.521)	2.565* (.381)	2.686** (.273)	2.686** (.273)	.874 (.503)	2.964** (.278)
Local Problem-Solving Competence: National Parties	.394** (.211)	.394** (.211)	.384** (.213)	.394** (.262)	.538** (.169)	.538** (.169)	.536** (.169)	.383** (.255)
PID	.299** (.171)	.300** (.172)	.298** (.173)	.298** (.171)	.338** (.132)	.338** (.132)	.339** (.132)	.336** (.132)
Interaction Effects
Type of local Democracy* Local Council Members		1.029 (.351)				1.054 (.251)
Type of local Democracy* Local Problem-Solving Competence: Local Lists			9.639** (.692)				4.277** (.553)
Type of local Democracy* Local Problem-Solving Competence: National Parties				1.205 (.384)				1.758 (.311)
Population Size
(Rural Community	.854 (.355)	.852 (.356)	.769 (.355)	.847 (.354)	.770 (.229)	.770 (.229)	.752 (.231)	.768 (.228)
Small Town	.458** (.236)	.458** (.236)	.458** (.236)	.457** (.236)	.682* (.178)	.682* (.178)	.687* (.178)	.680* (.178)
Medium Sized Town	.502** (.186)	.503** (.187)	.522** (.186)	.502** (.186)	.778 (.153)	.778 (.153)	.799 (.153)	.776 (.153)
Large City	Reference Category
Political Attitudes
Political Interest	.818** (.087)	.818** (.087)	.830** (.087)	.819* (.087)	.808** (.067)	.808** (.068)	.813* (.068)	.811* (.068)
Socio Demographics
Age	.993 (.005)	.993 (.005)	.993 (.005)	.992 (.005)	.987** (.004)	.987** (.004)	.987** (.004)	.987** (.004)
Gender	1.214 (.158)	1.216 (.158)	1.194 (.158)	1.221 (.159)	1.170 (.121)	1.170 (.121)	1.161 (.121)	1.180 (.121)
Education	.832* (.081)	.831* (.081)	.825* (.081)	.832* (.081)	.819** (.063)	.819** (.063)	.818** (.063)	.823** (.063)
Constant	24.692** (.576)	24.728** (.576)	25.170** (.576)	26.059** (.588)	21.616** (.463)	21.616** (.465)	21.956** (.464)	27.677** (.486)
Observations	947	947	947	947	1354	1354	1354	1354
Nagelkerke R²	0.23	0.23	0.23	0.23	0.20	0.20	0.21	0.20

Source: Neu/Konrad-Adenauer-Foundation (2018). For complete regression models see Tables A3 in Appendix

Note: Logistic Regression with Odds Ratios. Standard errors are clustered by states.

**p < 0.01.

*p < 0.05.