Economic voting in Dutch municipal elections: The role
of local and national economic performance and the effect
of political affiliation.
University of Groningen Faculty of Economics and Business
MSc Economics Master Thesis Economics
(EBM877A20)
Arjen Coenradij (S2198436)
Supervisor: Prof. dr. M.A. Allers
24-06-2016
Abstract
This Master Thesis analyzes multilevel economic voting in the context of Dutch municipal elections in the period 1990-2014. The goal of the paper is to determine whether local and national economic conditions affect voting behavior in Dutch municipal elections and to determine whether national incumbent parties are evaluated locally. In order to investigate this, a vote function has been introduced, which has been estimated with a fixed effects and GMM approach. Evidence is found for the existence of second-order economic voting for municipalities that have a level of political affiliation with national office higher than 50%.
JEL classification D72
2
1. Introduction
Economic voting is a concept in the fields of political and economic science that grasps the relationship between the state of the economy and the vote choice of the electorate. The main prediction of the theory of economic voting is that the electorate takes the state of the economy into account in deciding which political party to vote for in an election. The literature on economic voting has originated from pioneering work of Downs (1957) and Key (1966). Since their work the research has expanded and a voluminous body of literature has been created about theories of economic voting. Central to this body of literature has been the responsibility hypothesis, which states that the incumbent government should be held responsible for the economic conditions in its jurisdiction. Following a reward-punishment mechanism, the incumbent government is punished in a shrinking economy and rewarded when the economy is thriving. After more than 40 years of research it is now beyond doubt that the state of the economy affects the voting behavior of the electorate, as has been concluded by multiple extensive reviews of the literature (Nannestad & Paldam, 1994; Lewis-Beck & Paldam, 2000; Lewis-Beck & Stegmaier, 2000; Lewis-Beck & Stegmaier, 2013).
3 Another concern that has been identified in studies is the relatively small sample of countries on which the literature has focused. The USA is the most extensively investigated nation in the literature. Moreover, France, the United Kingdom and Denmark have been quite extensively investigated (Lewis-Beck & Stegmaier, 2000). However, research in most of the other democracies tends to be limited to either one paper, one author or one approach (Lewis-Beck & Stegmaier, 2000). This also applies to the Netherlands for which the number of investigations regarding economic voting is rather limited. There appears to be no analysis of any kind regarding economic voting at the local level in the Netherlands.
The local level of governance in the Netherlands presents itself as an interesting research area. There has been a process of decentralization of responsibilities from the national government towards the municipalities during the past few decades. Moreover, the number of municipalities continues to decrease as a result of municipal amalgamations. This dynamic environment interacts with the clarity of responsibility and interferes with the process of political and democratic accountability. Also, the increasing popularity of local political parties in municipal elections is a development that raises questions on how the electorate evaluates the performance of local incumbents. I&O Research, a market research agency, conducted a survey before the 2014 municipal elections and found that there is a large influence of national politics on municipal elections. More than 50% of the respondents indicated that national politics plays a role in their decision to vote for a certain party locally (I&O Research, 2014).
In light of these developments, it is critical to monitor whether the system of political representation is still functioning properly. If local incumbents are evaluated based on national economic conditions beyond their jurisdiction, the system of political accountability is failing. Moreover, if municipal elections are used by the electorate to evaluate national politics then the system of municipal elections does not fulfill its purpose, which is to evaluate the performance of local government based on policies and policy outcomes that are within their influence.
4 other words, it will analyze whether local elections are used as referenda for national incumbent governments’ performance with respect to managing the national economy. The following two research questions will be addressed:
Do local and national economic conditions affect voting behavior in Dutch municipal elections?
And
Does the electorate punish or reward political parties locally for the effects of policies set by their national counterparts, i.e. are national incumbents evaluated locally?
In order to answer these research questions, a vote-function has been estimated. This vote-function has been estimated with the fixed effect panel data estimation method. Moreover, a Generalized Method of Moments (GMM) approach has been applied in order to efficiently estimate a vote-function with a lagged dependent variable. In addition to the GMM estimation, several robustness analyses have been performed in order to validate the finding of the fixed effects estimation. It is expected that the local economy has a statistically significant effect on local election results. Moreover, it is expected that national economic performance is related to local vote choice and that national politics interferes with local elections.
To perform this analysis, a database containing information on the members of the city council and the executive board of mayor and aldermen and their political affiliation per municipality has been used. From this database a political affiliation variable could be derived. Moreover, data with respect to the number of unemployment benefit receivers per municipality has been used to measure local economic performance. Additionally, national economic indicators were applied to measure the state of the national economy, these include: unemployment, inflation and economic growth. The vote function is estimated based on Dutch municipal election results for the period 1990-2014, covering six electoral cycles.
5
2. Dutch political system
This chapter will provide an overview of the Dutch political system and the process of elections. Moreover, it will evaluate the position of municipalities in the Dutch governmental system. This description of the Dutch system of governance will facilitate the understanding of the context in which economic voting effects will be investigated.
2.1 The Dutch political system and process of elections
The Netherlands is a Parliamentary democracy with directly elected representatives. An important characteristic of a Parliamentary democracy is that voters can freely vote for representatives. In the Netherlands, elections are held for: the House of Representatives, the provincial states, the municipal council, and the European Parliament. Elections for the House of Representatives are held every four years. Early elections are held in case a government is forced to resign or resigns voluntarily. Elections for the provincial states and the municipal council are held every four years, whilst elections for European Parliament are held every five years. The members of the Senate are elected indirectly by the electorate through their vote for the provincial states (ProDemos, 2012). All Dutch citizens are entitled to vote and are entitled to apply themselves as a candidate after the age of 18. The elections are based on a system of proportional representation, giving the Netherlands one of the most proportional electoral systems of the world (ProDemos, 2012). After the election for the House of Representatives a cabinet has to be formed. This government is required to be supported by a majority of the representatives. Note that this does not require the parties in the coalition to have a majority of seats in Parliament. A majority in Parliament can be obtained through support by parties in Parliament that do not take a seat in the cabinet but do support and influence government policy. In this situation the coalition is a minority coalition that functions in a similar vein as a majority coalition.1 In the Netherlands this majority can only be
obtained through a collaboration of multiple political parties in a coalition, which is a result of the large number of parties in the Second Chamber due to the system of proportional representation.
1 An example of such a minority coalition in the Netherlands, is the coalition formed after the Parliamentary
6 The coalitions that have been formed during the past two decades are displayed in table 1 below, all of them contain at least two and at most three political parties.
Table 1, Composition of the cabinet from 1989 till now. Source: Parliament & Politics (2016)
2.2 State, province and municipality
After the formation of a coalition, a cabinet is formed. This cabinet consists of the prime-minister, ministers and state secretaries and functions as the Dutch central government (Rijksoverheid, 2016). The Dutch government consists of three levels: central government, provinces and municipalities. Decentralization, referring to the transfer of tasks and responsibilities from central government to the provinces and municipalities, has led to an increased importance and relevance of lower levels of government (ProDemos, 2012). The provincial level will not be taken into account in this Master Thesis.
The lowest level of government in the Netherlands are the municipalities. In 1905 the number of municipalities was 1121. By 2000 this number was reduced sharply to 537 municipalities (Ramkema, 2009). As of January 1, 2016, there are 390 municipalities in the Netherlands. This sharp reduction in the number of municipalities is related to the process of decentralization of government tasks from the central government to provinces and municipalities. Since the end of the twentieth century municipalities have received an increasing degree of freedom in constructing policy and executing national regulation (Ramkema, 2009). Municipal amalgamations have led to
CDA PvdA VVD D66 CU LPF
Date of election Coalition
7 a large number of small municipalities merging into bigger municipalities. Tasks and responsibilities of the municipal government are very diverse and related to the following policy areas: civil affairs, public order and safety, economic affairs, social affairs and employment, public health, education, spatial planning and public housing, traffic and (public) transport, and culture, sport and recreation (Ramkema, 2009).
The administration of a municipality consists of a council and an executive board, which consists of the aldermen and the mayor. The city council is elected directly by the inhabitants of the municipality and has three main responsibilities: to establish the main municipal policy, to monitor the functioning of the executive board, and to represent the inhabitants of the municipality (Ramkema, 2009). The executive board of mayor and aldermen is the administration of the municipality and is responsible for: the proper execution of legislation, preparing the city council for decision making processes, and executing council decisions (Ramkema, 2009). The aldermen are chosen by the city council, whereas the mayor is appointed by the central government. The number of aldermen and the number of councilors in a municipality depend positively on the number of inhabitants of that municipality.
This Master Thesis will focus on regular elections2 for the city council and will seek to determine whether economic voting occurs in these elections. Elections for the city council in the Netherlands take place in March every four years.3 Moreover, this Master Thesis will analyze the role of the national economy and national politics in local elections. There are differences between local politics and national politics and their corresponding elections with respect to the political parties that are participating. Most of the political parties in national politics have local branches that participate in local elections. However, not all national parties are represented locally or some parties might not be represented in some of the municipalities. An example of this is the PVV (Freedom Party), which has 15 seats in national Parliament but is only represented in two municipalities. Moreover, an important factor in local politics in the Netherlands are local political parties, which are only active at the local level and are not connected to a national party.
2Regular elections for the municipal Council refer to those elections organized in March once every four
years. Thus, this Master Thesis will not consider elections organized in relation with the municipal amalgamations.
3 In the past three decades, municipal elections have been held on: 21-03-1990, 02-03-1994, 04-03-1998,
8 Approximately 25% of municipal council members is member of a local political party rather than a national one.4
3. Literature review
After having sketched the political context, this section will start with an exploration of the economic voting literature and will provide an overview of its main findings. Thereafter, it will consider the importance of the political context. This will be followed by a review of the economic voting literature in the Netherlands. Finally, this section will conclude with an analysis of economic voting at the local level in which several hypotheses will be proposed.
3.1 Economic voting theory
The economic voting theory represents the point where the fields of economics and political science meet. This mix of politics and economics is generally represented by the Vote- and Popularity-function (VP-function) in which the vote for or the popularity of the incumbent government is explained by economic conditions. It suggests economic explanations for political results. The rationale for relating the vote to economic conditions stems from pioneering work of Downs (1957) and Key (1966) who laid the foundation of the economic voting theory and provided the inspiration for formulating the responsibility hypothesis. According to this responsibility hypothesis, the incumbent government is held responsible by the electorate for economic performances. This responsibility is imposed on the incumbent government by the electorate through a reward-punishment mechanism, which directly follows from Key’s (1966: 61) description of the electorate: ‘…an appraiser of past events, past performance, and past actions.’. The economic voting theory predicts that the incumbent government will be rewarded by the electorate when the economy flourishes and punished when the economy is in recession (Key, 1966). The idea that incumbent governments are evaluated based on past economic performance
4Some of these local political groups do have ties to national politics. These are political groups that
9 can be seen as retrospective economic voting, which is considered by the majority of the literature (Lewis-Beck & Stegmaier, 2000). The work of Downs (1957: 39) initiated the theory of prospective economic voting, which is apparent in his description of a voter: ‘He makes his
decision by comparing future performances he expects from the competing parties’. Most findings
in the literature indicate a better fit of the economic voting model to retrospective variables (Lewis-Beck & Paldam, 2000). However, logic implies that prospective and retrospective economic voting are not independent. When an economic agent evaluates the incumbent government based on past economic performance, he/she most likely does so in light of an expectation of future performance derived from past performance.
The responsibility hypothesis has generated a large body of research, which can be seen as an indication of the robustness of the economic voting theory and as evidence of the substantive importance of the economic vote in democracies (Lewis-Beck & Stegmaier, 2000; Lewis-Beck & Paldam, 2000; Lewis-Beck and Stegmaier, 2013). The literature makes a distinction between macro- and micro-studies. The micro-studies use individual micro-level data from election surveys to estimate popularity functions, whereas macro-studies use aggregated data on macro-economic variables to explain election results by estimating vote functions. Micro-studies are invaluable because they are not subject to the ecological fallacy from which macro-studies suffer (Lewis-Beck & Stegmaier, 2013). This ecological fallacy refers to the potential spurious nature of the relationship between macro-economic variables and election outcomes. Voters might not base their vote on economic conditions, but might base their decision on some unknown third variable, rendering the estimates in vote-functions unreliable (Lewis-Beck & Stegmaier, 2013). However, extensive reviews of the literature have indicated that the relationship between macro-economic variables and election outcomes are reflective of the micro-process by which individuals’ economic perceptions weigh in the individual vote choice (Nannestad & Paldam, 1994; Lewis-Beck & Stegmaier, 2013). Therefore, from the observation that the macro-results are emerging from micro-patterns, it can be concluded that the macro-studies correctly estimate economic voting and are not reflecting spurious results (Lewis-Beck & Stegmaier, 2013).
Generally, when considering macro-studies and vote functions, the vote is modelled as a function of economic and political variables:
10 The conventional variables for indicating economic performance in the economic voting literature are: unemployment, inflation and economic growth. Recent literature has found that unemployment and economic growth are the main predictors of vote choice. However, literature in the more distant past has found that unemployment and inflation are the most important variables (Lewis-Beck & Stegmaier, 2013). The vote-function (i.e. equation 1) indicates the importance of including variables representing the political context in order to account for the environment in which the voting takes place. The role of the political context will be discussed in the following section.
3.2 Instability and the importance of political context
An extensive literature review by Lewis-Beck and Stegmaier (2000) has indicated that the USA is the most studied nation in the field of economic voting. Also, countries such as France, the United Kingdom and Denmark have been quite heavily researched. In addition to these four countries, some form of research into economic voting has been performed in a majority of democratic countries around the world. It now seems to be universally accepted that the economy matters in the vote choice. Three tendencies related to economic voting can be identified in literature (Kayser & Wlezien, 2011): performance matters more than the policy itself, national level economic conditions are more important than personal economic conditions, and short run changes in the economy are more important than long run changes. The effect of the economy on elections is not universal. There is strong support for this effect in the USA, France, the United Kingdom and Canada, whereas evidence in other countries is less conclusive (Kayser & Wlezien, 2011). Despite the voluminous body of research, the interest in economic voting theory is persistent due to apparent instabilities of the VP-function over time and across countries (Nannestad & Paldam, 1994; Lewis-Beck & Paldam, 2000). However, the instability is mainly caused by measurement problems and inadequately constructed theory, hence, the observed instability is only apparent and caused by specification errors (Lewis-Beck & Paldam, 2000). This has been confirmed by Bellucci and Lewis-Beck (2011) who find that the empirical instability is apparent rather than inherent, with the apparent instability being caused by faulty measurement of economic and political factors and problems with small individual country samples.
multi-11 nation analysis is Paldam (1991) who analyzed seventeen countries and did not find strong evidence for economic voting effects. An explanation for this lack of conclusive results in multi-nation studies was provided by Powell and Whitten (1993) who found that the political context matters for the strength of the economic vote. Powell and Whitten (1993) conclude that factors of the political context affect the clarity of responsibility of the incumbent governments. A political context that supports high clarity of responsibility will face a strong connection between economic conditions and voting behavior, whereas a political context that blurs clarity of responsibility will face a weak connection between economic conditions and voting behavior (Powell and Whitten, 1993).
Additional evidence for the importance of political context has been provided by Anderson (2000). A key finding of Anderson (2000) is that mechanisms of accountability affect the ability of the voter to evaluate economic performance. Political accountability relates to the extent to which an incumbent government can be held responsible for its actions. The literature has pointed to an increased effect of the economy on government support when responsibilities of the government with respect to policymaking and management of the economy are clear (Anderson, 2000). Building on this literature, Anderson (2000) finds that there is a strong effect of voters’ perceptions of the economy on the vote choice if there is a high institutional clarity of responsibility, if governing party target size is large enough5, and if there are a limited number of clear available alternatives. Anderson (2000) analyzed thirteen European democracies and found that the political context influences the relationship between economic perceptions and vote intention. His work indicates the importance of controlling for political context in order to determine the effect of the economy on the vote.
An explanation for the USA being the breeding ground for studies on economic voting is that a two-party system perfectly suits the responsibility hypothesis in that it is relatively easy to assign responsibility in such a system (Lewis-Beck & Paldam, 2000). European multi-party and coalition systems of government represent a greater burden to the clarity of responsibility than does the two-party system in the USA. Contrary to a two-two-party system where the economic vote is against the
5 It is expected that it is easier for the electorate to target parties and assign credit or blame in a country
12 incumbent government party, in a multi-party system the economic vote may be either against the entire coalition, against one party within the coalition, or be dependent on particular policies being related to particular parties (Lewis-Beck & Paldam, 2000). Since there are no problems with accountability in two-party majority systems it is expected that voters are more likely to punish incumbents in such a system than in a multi-party coalition system (Narud, 1996). However, according to Bellucci and Lewis-Beck (2011), the economic voting effect is both substantial and statistically significant across Western-European countries. The electorate is found to withdraw their support for the incumbent government when the economy is perceived to be in poor shape (Bellucci & Lewis-Beck, 2011). This implies that the economic voting model is applicable to European multi-party systems.
3.3 Research regarding economic voting in the Netherlands
This section will analyze the evidence for economic voting in the Netherlands, which is one of the European multi-party systems the previous section referred to.
13 vote being determined by religion and social class effects. The researchers express the expectation that governments might become more dependent on economic fluctuations in the future due to a reduction of the importance of religion and social class in voting behavior.
Before the 1990’s the Dutch political landscape was characterized by continuity rather than change. The political institutions were resistant to change and party choice was mainly based upon religion and class (Irwin & Van Holsteyn, 1997). This changed dramatically in the 1994 Parliamentary elections, the results of which provided more electoral change than any other election in the 20th century (Irwin & Van Holsteyn, 1997). Since the introduction of universal
suffrage the cabinet formed after the 1994 elections was the first coalition that did not include the Christian Democrats. Irwin and Van Holsteyn (1997) apply a cohort analysis for the 1989 and 1994 elections and found that the structured vote model has lost much of its explanatory power regarding party choice of voters of 40 years and younger. Irwin and Van Holsteyn (1997) conclude that the 1994 election results were not guided by factors related to religion, class or ideology but rather by short term factors such as the state of the economy and the popularity of party leaders. Electoral change in the Netherlands, defined as the number of seats in Parliament changing hands, has been among the highest in Europe since 1994 (Irwin & Van Holsteyn, 2008). The importance of religion and social class for voting behavior has continued to decrease and the government’s ability to control the economy has proven to be important (Irwin and Van Holsteyn, 2008). Swank and Eisinga (1999) have performed an empirical analysis regarding the effect of economic conditions on party choice in the Netherlands for the period 1978-1995. They find evidence for the partisan voter model in the Netherlands, in which right-wing political parties are expected to benefit from poor economic growth prospects, whereas left-wing political parties are expected to suffer (Swank & Eisinga, 1999). The intuition herein is that voters are less willing to sacrifice growth prospects for redistributive policies when the economy is in poor shape. Moreover, by estimating VP-functions, Swank and Eisinga (1999) find evidence for the responsibility hypothesis in the Netherlands, since voters are found to reward (punish) the incumbent government for favorable (unfavorable) economic conditions.
14 literature on local economic voting in other countries and will serve as the transition towards the aim of this paper: investigating the role of the economy in Dutch municipal elections.
3.4 Local economic voting
Anderson (2006) advances the literature by analyzing multilevel governance and finds that it weakens economic voting at the national level. According to Anderson (2006), the wave of decentralization of governance around the world blurs the responsibility for economic conditions and, therefore, mutes incumbent voting and undermines democratic accountability. Implicit in the responsibility hypothesis and the theory of political accountability is the notion that levels of government should only be held responsible for economic conditions that are within their jurisdiction. This section will analyze the literature on economic voting at the local level and will determine whether local governments are held responsible for local economic conditions or whether they are held politically accountable for economic conditions beyond their jurisdiction. 3.4.1 The local economy and local elections
Dassonneville et al. (2016) find that research on economic voting in the local context can be divided into two main approaches. The first approach considers the local election as a referendum for national-level politics and shows that voters credit or blame parties in national office through local elections based on national economic conditions. The second approach considers the effect of local economic conditions on national election results. Dassonneville et al. (2016) identify a gap in the literature since they find that little attention has been devoted towards the effect of the local economy on the local vote choice. The traditional economic voting theory is argued to hold at the local level as well, with voters holding local incumbents accountable for the state of the economy (Berry & Howell, 2007; Boyne, James, John & Petrovsky, 2009; Oliver & Ha, 2007). Dassonneville et al. (2016) contribute to the gap in the literature by analyzing the 2012 Belgian municipal elections and estimating the link between local economic conditions and local election results. They find evidence of a relationship between the local economy and the local vote, with an increase in unemployment reducing the probability that the vote goes to the incumbent. 3.4.2 Hypothesis 1
15 with local incumbents being held accountable for local economic conditions by the electorate. However, in the Netherlands the influence of the municipal government on the economy is limited. Therefore, it might have few tools available in order to fight local unemployment or support local economic growth. In that sense, the existence of a local economic vote effect would mean that the local incumbent is held responsible for economic conditions that are beyond their influence. However, taking into consideration the findings by Dassonneville et al. (2016), the incumbent vote share is expected to depend negatively on local unemployment. This leads to the first hypothesis:
Hypothesis 1: The state of the local economy affects the vote share received by the local incumbent in municipal elections.
3.4.3 The local economy and national elections
In addition to an effect of the local economy on local election results, as discussed in the previous section, the literature has also analyzed the relation between the local economy and national election results. According to Cutler (2002), it is undisputed that national election results and the national economy are linked. However, local economies might not move in line with the national economy as a result of local policies. Cutler (2002) analyzes voting behavior in Canada and finds that local economic conditions and local impacts of national policies affect the popularity of the incumbent central government. The willingness to vote for the incumbent depends positively on the performance of the local economy (Cutler, 2002).
Elinder (2010) aims to extend the literature on the subject by analyzing six Swedish general elections and measuring the response to regional and municipal unemployment and economic growth. Evidence is found of a substantial response in general elections to local economic conditions. If these conditions are omitted from voting equations, estimates related to the effect of national economic conditions may be biased (Elinder, 2010). In a similar vein, Auberger and Dubois (2005) analyze four legislative elections in France and find that national and local economic conditions have an important role in the French legislative election results.
16 manner. Namely the effect of the national economy on the local vote will be investigated. The following section will describe this.
3.4.4 Multiple levels of economic voting
Macro-economic conditions are beyond the influence and jurisdiction of local governments. However, despite the absence of direct control it appears that local elections are used as referenda for the performance of parties in national office.
Traditional economic voting theory has focused on the relationship between the vote for the national incumbent and the national economy and can be labelled as national economic voting (NEV) (Bosch, 2016). More recent literature has hypothesized a relationship between the local economy and the vote for the local incumbent, which is labeled as local economic voting (LEV) (Bosch, 2016). It has already been argued that LEV theory is not as universally accepted as NEV theory due to a higher difficulty to assign responsibility at the local level than at the national level. Moreover, the literature has focused on the restrictive assumption of jurisdiction specific responsibility, according to which local elections are used to assess local incumbents and national elections are used to assess national incumbents (Bosch, 2016). However, this assumption limits the extraction of other potential sources of economic voting in which national and local incumbents may be assessed in both national and local elections (Bosch, 2016). Bosch (2016) developed a framework to summarize the different forms of economic voting in local elections, including economic voting patterns that reflect contamination between national and local level jurisdiction (see table 2 below).
Table 2, Multilevel economic voting in local elections (taken from Bosch (2016))
In this Master Thesis, sections 3.1 – 3.3 have introduced the Genuine NEV theory in which national incumbents are assessed based on national economic conditions. Keep in mind that the Genuine NEV theory refers to national elections, whereas table 2 refers to local elections. Subsection 3.4.1 has introduced the link between local economic conditions and the local incumbents which fits in the Genuine LEV stream of table 2. However, coattail economic voting, second-order economic
Incumbent to be assessed
Local National
Economy to be assessed Local Genuine LEV Coattail EV
17 voting, and C-type economic voting have not been discussed thus far. Coattail economic voting and second-order economic voting refer to the assessment of the performance of the national incumbents by the electorate in local elections. This is an area of research that has received little attention in the literature thus far (Bosch, 2016). Coattail economic voting refers to the assessment by the electorate in local elections of the performance of the national incumbent based on the state of the local economy (Norpoth, 2001). Second-order economic voting refers to the assessment by the electorate in local elections of the performance of the national incumbent based on the state of the national economy (Rodden & Wibbels, 2010; Boehmke, Lewis-Beck & Rydberg, 2010; Fauvelle-Aymar & Lewis-Beck, 2011). According to Bosch (2016), the occurrence of coattail and second-order economic voting depends on the nationalization of the party system. The degree of nationalization of the party system refers to the similarity in the strength with which a political party competes in different levels of government (Rodden & Wibbels, 2010). Ceteris paribus, higher degrees of nationalization of the party system will lead to a higher probability of the occurrence of contaminated forms of economic voting (Bosch, 2016). A final type of economic voting displayed in table 2 is C-type economic voting, according to which local incumbents are evaluated in local elections based on national economic conditions. However, C-type economic voting will not be analyzed, rather the focus will be on coattail and second-order economic voting. 3.4.5 The national economy and local elections
Analyzing the 2012 Catalan election, Bosch (2016) finds that local elections reflect an opportunity for the electorate to assess national incumbents based on national economic conditions. Bosch (2016) expects that second-order economic voting is more common than previously thought and suggests that future research tests for the importance of second-order economic voting in local elections.
18 & Veiga, 2013). This is in line with Carsey and Wright (1998) who state that the electorate may want to punish or reward the national government in second-order elections.
The importance of clarity of responsibility for the strength of the economic vote effect has been discussed above. An important take-away from that discussion is that factors that undermine the clarity of responsibility will misdirect the effect of the economy on election results (Powell & Whitten, 1993). At the local level the clarity of responsibility is undermined by the fact that both national and local authorities can be held responsible for economic conditions (Martins & Veiga, 2013). According to Martins and Veiga (2013), the electorate is better able to assess responsibility when the same political party heads the national as well as the local government. This implies the expectation that local parties are more likely to be held responsible for national economic conditions if they are affiliated with the national government. Martins and Veiga (2013) find evidence for political accountability of the local government for national economic conditions. They find evidence of an effect of national unemployment on local elections results, with local parties affiliated with national office being punished in case of an increase in national unemployment. Moreover, they find that affiliation with national office leads to a stronger effect of inflation on local election outcomes.
Gélineau and Bélanger (2005) analyze Canadian election results and macro-economic data for the period 1953-2001 and find that provincial incumbents are held responsible for poor national economic conditions if the provincial incumbent is of the same partisan family as the incumbent federal party.
3.4.6 Hypothesis 2 and 3
These findings with respect to the interaction between the national economy on the one hand and the local vote on the other hand lead to the formulation of the following hypotheses:
Hypothesis 2a: The vote share obtained by the national incumbent parties in local elections is affected by local economic conditions, following a reward-punishment mechanism (coattail economic voting).
19
Hypothesis 3a: The vote share obtained by the national incumbent parties in local elections is affected by national economic conditions, following a reward-punishment mechanism (second-order economic voting).
Hypothesis 3b: The second-order economic vote effect is higher when the local government is affiliated with national government.
4. Methodology
This chapter will discuss the methodology that has been applied in order to formally test for the hypotheses posed in the previous chapter. It will explain the econometric model specification that has been applied and will discuss the estimation techniques that have been used.
4.1 Econometric model specification
In order to estimate the impact of the economy on the vote and in order to test the hypotheses stated in chapter 3, a vote function will be estimated. As has been discussed above, the vote function estimates the vote as a function of the economy and of the political context. The advantage of studying municipal elections rather than national elections is that the municipalities within a country face, for a large part, the same political context. If the analysis is built upon a panel of countries, then elections take place in different political contexts, which have to be properly accounted for. Municipal elections in the Netherlands to a large extent face similar circumstances. Nevertheless, some variables related to the local political context will be taken into account here. This vote function will have the following empirical specification:
𝑉𝑆𝑖,𝑦 = 𝛽1+ 𝛽2𝐸𝐶𝑂𝐿𝑖,𝑦−1+ 𝛽3𝐴𝐹𝐹𝑖,𝑦+ 𝛽4𝐸𝐶𝑂𝐿𝑖,𝑦−1∗ 𝐴𝐹𝐹𝑖 + 𝛽5𝐸𝐶𝑂𝑁𝑦−1 + 𝛽6𝐸𝐶𝑂𝑁𝑦−1∗ 𝐴𝐹𝐹𝑖,𝑦+ 𝛽7𝑃𝑂𝑃𝑖,𝑦+ 𝛽8𝐿𝑇𝑖,𝑦+ 𝜇𝑖 + 𝜀𝑖,𝑦 𝑖 = 𝑚𝑢𝑛𝑖𝑐𝑖𝑝𝑎𝑙𝑖𝑡𝑦 𝑖𝑛𝑑𝑒𝑥 = 1, 2, … , 403
𝑦 = 𝑚𝑢𝑛𝑖𝑐𝑖𝑝𝑎𝑙 𝑒𝑙𝑒𝑐𝑡𝑖𝑜𝑛 𝑦𝑒𝑎𝑟 = 1994, 1998, 2002, 2006, 2010, 2014
Where VSi,y is the vote share obtained by the incumbent government in municipality i in election
year y. ECOLi,y-1 is a set of variables measuring the local economy in municipality i in election
year y, and ECONy-1 is a set of variables measuring the national economy in election year y. AFF
20 is an indicator variable which indicates the degree of affiliation of the local government with the national government. A local government is affiliated with the national government if the same political parties are incumbent in both levels of government. In addition to including affiliation as part of the interaction terms, it has also been included in the model separately. This is required since failing to do so introduces the danger of inferential mistakes (Brambor, Clark & Golder, 2005). POPi,y is the size of the population in municipality i in year y. LTi,y is the Laakso-Taagepera
index for the effective number of political parties in municipality i in year y. This measure for the effective number of political parties serves as a control for the degree of political fragmentation of the party system (Laakso & Taagepera, 1979). The Laakso-Taagepera index is defined by equation 3 below, where n is the number of parties with at least one seat in the city council and pj is the
share of seats of party j in the council. 𝐿𝑇𝑖,𝑦= 1
∑𝑛𝑗=1𝑝𝑗2
εi,y is the error term and μi is an individual fixed effect for municipality i, which is included to
control for time invariant variables. Because of historical or demographical reasons, voters from certain municipalities have a tendency to vote for certain political parties. This tendency is not likely to be substantially influenced by economic conditions. In addition to individual fixed effects, year fixed effects (i.e. year dummies) will be included in the model specification when estimation is performed on local economic variables only.
This vote function will be estimated with a panel data set, which has a large number of cross-sections relative to the number of time periods (403 versus 7). As discussed in section 3.3, the economy has not been a substantially important factor in Dutch elections until the 1990’s. Therefore, this empirical investigation will start its analysis with the municipal elections of 1990. The dependent variable in equation 2 is the vote share of the incumbent parties. This represents the limitation of a low number of time periods available, since no information is available on the incumbency at the municipal level before 1990. Therefore, it is not possible to construct the vote share of the incumbent for the year 1990. The number of time periods, therefore, is limited to six. 4.2 Hypothesis testing
In order to test whether local economic performance affects the vote share of the local incumbent, i.e. hypothesis 1, a test for the significance of 𝛽2 will be performed. If local economic performance
21 is measured by local unemployment, the coefficient is expected to be negative. That is, hypothesis 1 is operationalized as follows:
Hypothesis 1: 𝐻0: 𝛽2 = 0 | 𝐻1: 𝛽2 < 0
In order to test whether local economic performance affects the vote share of the national incumbent in local elections, i.e. hypothesis 2a, a test for the significance of 𝛽2 will be performed. This is equivalent to the test for hypothesis 1, with the single difference being that the dependent variable now is the vote share of the national incumbent in local elections, rather than the vote share of the local incumbent in local elections. The test for hypothesis 2b, i.e. the positive effect of affiliation on the strength of the economic vote effect, is operationalized as follows:
Hypothesis 2b: 𝐻0: 𝛽4 = 0 | 𝐻1: 𝛽4 ≠ 0
In order to test whether national economic performance affects the vote share of the national incumbent in local elections, i.e. hypothesis 3a, a test for the significance of 𝛽5 will be performed. This coefficient is expected to be positive when national economic performance is measured by economic growth, and negative when national economic performance is measured by unemployment and inflation. For hypothesis 3b, a test for the significance of 𝛽6 will serve as the
tool to determine whether affiliation increases the economic vote effect. This leads to the following tests of significance of coefficients
Hypothesis 3a: 𝐻0: 𝛽5 = 0 | 𝐻1(𝑒𝑐𝑜𝑛𝑜𝑚𝑖𝑐 𝑔𝑟𝑜𝑤𝑡ℎ): 𝛽5 > 0
| 𝐻1(𝑢𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡, 𝑖𝑛𝑓𝑙𝑎𝑡𝑖𝑜𝑛): 𝛽5 < 0
Hypothesis 3b: 𝐻0: 𝛽6 = 0 | 𝐻1: 𝛽6 ≠ 0
4.3 Testing
22 Hausman test has been performed in order to determine whether fixed effects or random effects are preferred. The results of these tests for the different specifications are displayed in table D.1 in appendix D. The results indicate that estimation with random effects yields inconsistent estimates. Therefore, estimation will be performed with municipality fixed effects. Year fixed effects will only be included in the model when estimation is performed without national economic variables, since these do not differ across municipalities and can therefore not be included in the model simultaneously with year dummies.
A test for serial correlation has been performed.6 This test has indicated evidence for the existence
of serial correlation between the economic indicator variables.7 Moreover, a test for heteroskedasticity has been performed.8 No evidence for heteroskedasticity has been found, i.e. a failure to reject the null hypothesis of homoscedasticity. However, the test is known to have low power for fixed effects models with large N and small T. Therefore, a failure to find evidence does not enable the rejection of the existence of heteroskedasticity in this situation. Thus, in order to adjust for serial correlation and to take into account the potential for heteroskedasticity, cluster robust standard errors have been applied.9
The possibility of endogeneity, the existence of correlation between the explanatory variable and the error term, has to be considered as well. There might be a potential source of endogeneity in the vote-function under consideration. Specifically, one might argue that there is reverse causality between the explanatory variables related to economic performance and the dependent variable, vote share of the incumbent. This line of reasoning goes as follows. The votes received by the incumbent influence the type of government that will be formed, which in turn influences the type of economic policies that will be introduced, which in turn influences economic performance. However, it is expected that this source of endogeneity does not have a substantial impact. The dependent variable in the different model specifications is the vote share of the local or national incumbent in municipal elections. The office at stake is thus the local office. The influence of municipalities on (macro-) economic performance in the Netherlands is limited, if not zero.
23 Moreover, the specification includes lagged values of economic performance. This further reduced the likelihood of endogeneity.
Nevertheless, the next section will discuss and apply a GMM estimation of an alternative model specification. The GMM is an alternative estimator that can deal with potential endogeneity problems without the need for obtaining external instruments. However, the main reason why the GMM is applied is in order to be able to estimate the alternative model specification that will be proposed in the following section. This specification is popular in the literature and includes a lagged dependent variable in the list of regressors as a measure for party identification or vote stability.
4.4 Alternative econometric model specification
An alternative to the model specification of equation 2 would be to specify a model that includes a lagged value of the dependent variable as an additional explanatory variable. This has been observed in the literature quite frequently. In this alternative specification (i.e. equation 4) the lagged dependent variable is included in the vote function as a proxy for party identification or as a control for the aggregate stability of the vote over the electoral cycle. The lagged dependent variable is defined as the vote share obtained by the current incumbent parties in the previous election (i.e. in year: y – 4). However, including a lagged dependent variable in a panel data model has the potential to introduce a dynamic panel bias. This bias is also known as the Nickell bias and is most likely to occur in a setting in which the number of cross sections is large relative to the number of time periods. This applies to the panel data set used here, since the number of cross sections is 403, whereas the number of time periods is only six. A possible solution to this problem is to apply dynamic panel data modelling.
𝑉𝑆𝑖,𝑦 = 𝛽1+ 𝛽2𝑉𝑆𝑖,𝑦−4+ 𝛽3𝐸𝐶𝑂𝐿𝑖,𝑦−1+ 𝛽4𝐸𝐶𝑂𝐿𝑖,𝑦−1𝐴𝐹𝐹𝑖,𝑦+ 𝛽5𝐸𝐶𝑂𝑁𝑦−1
+ 𝛽6𝐸𝐶𝑂𝑁𝑦−1𝐴𝐹𝐹𝑖,𝑦 + 𝛽7𝑃𝑂𝑃𝑖,𝑦 + 𝛽8𝐿𝑇𝑖,𝑦+ 𝜇𝑖 + 𝜀𝑖,𝑦
The error term (𝜀𝑖,𝑦) is included in all of the lagged terms of the dependent variable as well. Therefore, the lagged dependent variable is not independent of the error process. Due to the endogeneity of one of the regressors, estimation by fixed effects and random effects will not be consistent. A solution to the bias in this model specification is to estimate the model using the GMM approach. Instead of the instrumental variable approach, this dynamic panel data approach
24 will provide more efficient estimates. The Arellano-Bond estimator and Arellano-Bover/Blundell-Bond are the most popular dynamic panel estimators (Roodman, 2009). Roodman (2009) devised a command that implements these estimators.10 This estimator is designed for datasets with many cross sections and few time periods and is therefore highly suited to this model.
5. Data
This section will provide a description of the data. Moreover, it will explain which variables are used in the analysis and what the sources of the data are.
5.1 Dependent variable
The dependent variable of the model is the vote share (VSi,y) received by the incumbent political
parties at the different municipal elections. In order to measure the vote share received by the incumbent, data on the election results for Dutch municipal elections are used. Data is obtained from the Election Board (Dutch: Kiesraad). The data set contains information regarding the number of votes cast at each election and on the share of votes received by the different political parties. For the local political parties, the election results are grouped together rather than published separately for every individual local political party. An extensive dataset regarding the composition of the city councils and executive boards of mayor and aldermen of all Dutch municipalities has been applied to determine which political party is the incumbent in which municipality. This dataset has been provided by COELO11 for the time period 1990-2014. Hence, according to the above description, the vote share of the incumbent for the election of 1994 is the share of votes received by those political parties that had one or more seats in the executive board of the municipality. A comprehensive overview of all the data used in this Master Thesis and their sources is published in Appendix A. A complete list of political parties that are available in the dataset of this thesis, together with a description of the level of government they are operating on, is published in Appendix B. The descriptive statistics with respect to the data regarding municipal election results are displayed in table 3 below. What should be noted from this table is the lower number of observations for the local votes variable (number of votes received by local parties in
10Xtabond2 in Stata
11 Center for Research of the Economy of Local Authorities (Dutch: Centrum voor Onderzoek van de
25 municipal elections). This is because of the missing data on local political party election results for election year 1990. Moreover, the table does not contain information on the election results for the Freedom Part (PVV), which is a big player in national politics and supported the minority coalition by CDA and VVD from 2010-2012. The reason for leaving PVV out of the sample for this Master Thesis is the absence of participation of the party in municipal elections in most of the municipalities. Moreover, the CU is included in the sample. The CU was formed in 2000 as a result of a merger between the RPF12 and the GPV13. Therefore, data from the elections in 1990, 1994
and 1998 with respect to votes for the RPF and the GPV have been summed to construct the CU-variable.
Table 3, Descriptive statistics vote data
A limitation of the dataset regarding municipal elections results is that there is no differentiation between the different types of local political parties in the data. Instead, the votes for the different local political parties are all grouped together. This implies that the vote share obtained by local incumbent parties cannot be constructed from this dataset. Therefore, the COELO dataset has also been applied for constructing the vote share of the incumbent variable. This variable takes the vote share of the incumbent to be the share of seats in the city council received by the incumbent political parties after the municipal elections. The descriptive statistics with respect to the municipal council and executive board data can be found in Appendix C in tables C.1 and C.2 respectively. For some political parties the number of observations is lower than for others. For some political parties data was available only for the most recent election(s).
12Reformed Political Federation (Dutch: Reformatorische Politieke Federatie)
13 Reformed Political Alliance (Dutch: Gereformeerd Politiek Verbond)
Variable Observations Mean Std. Dev. Min. Max.
26 One of the predictions stated in chapter 3 was that the effect of the economy on vote choice is higher when the local government is affiliated with the national government. Implicit in this prediction is that political parties are held responsible locally for performance of political parties nationally. Looking at the data it appears that vote shares decline during periods in which a political party is taking responsibility by governing nationally (see figure 1 below). Not only does it appear that the national incumbent parties vote share declines in periods of governing, but also it seems like the local incumbent parties vote shares decline when a political party takes responsibility nationally. If this holds true in the empirical estimation results also, then this signals that the process of democratic accountability is not functioning properly, since local incumbents should only be held responsible for policies within their jurisdiction. Testing of the hypotheses might give further insights. This will be discussed in chapter 6.
5.2 Explanatory variables: state of the economy and political affiliation
In order to measure local economic performance (ECOLi,y) a proxy for unemployment has been
used. Data availability on local economic performance is limited, with no (reliable) data being available for local economic growth and local inflation. Therefore, locally the economy is measured by a proxy for unemployment: the number of unemployment benefits receivers. This variable is defined as follows: the sum of the number of persons per municipality receiving WW-benefits14 and the number of persons per municipality receiving ABW-benefits.15 This data is available on a per municipality basis and ranges from 1991 to 2014 for the WW-benefits, and ranges from 1992 to 2014 for the ABW-benefits. The local unemployment variable is constructed by dividing the total number of unemployment benefits receivers by the working-age population (population aged between 15 and 65).16 Even though this is an imperfect measure of unemployment, it is an appropriate proxy, given that changes in the unemployment benefits law over time apply to all municipalities. Data is obtained from Statistics Netherlands. Hence, only one variable related to local economic performance is available. This is a limitation of this study. However, including unreliable local economic indicators would not be desirable.
14 Unemployment Law (Dutch: Werkloosheidswet): social insurance program that offers unemployment
benefits to the unemployed in the labor force.
15General Assistance Act (Dutch: Wet werk en bijstand/Algemene bijstandswet (ABW)): social security
program that offers benefits to people without sufficient income to fulfill their basic means of subsistence.
16Local unemployment = (ABW-receivers + WW-receivers)
27
Panel A, VVD Panel B, PvdA
Panel C, CDA Panel D, D66
Panel E, CU Panel F, Local Parties
28 In order to measure national economic performance (ECONy), data on unemployment, inflation
and economic growth is used. This is in line with common practice in literature, in which these three economic indicators are used to perform estimations regarding the economic vote effect. Data on these variables is obtained from Statistics Netherlands, ranging from 1990 to 2014, and is measured nationally. Figure 2 traces the development of these indicators over the past two decennia, the patterns display quite some fluctuations in the movement of these variables which can be exploited when measuring the economic vote effect.
Figure 2, National Economic Indicators
29 that takes a value of one if all parties that are in national office also hold office in a municipality. If one or more of the parties in national office do not hold office locally, then the variable takes a value of zero. In national election years, the incumbent parties are taken to be those that were governing for the longest period of time in that specific year. Hence, for the year 2012, when national elections were held at September 12, the incumbent parties are those that were part of the coalition for the first 8 months of the year.
The dataset provided by COELO can be applied to determine the Laakso-Taagepera index at the municipal level. For the computation of the index the members of the city councils have been used as a starting point. The share of seats of every political party with one or more seats in the city council has been calculated, after which this information was used to construct an index for political fragmentation at the municipal level.
Data on the population per municipality has also been obtained from Statistics Netherlands. This variable is included in the model in order to control for the effect of municipality size on the difficulty to obtain a large vote share. This follows from the expectation that more political parties participate in the elections in large municipalities than in the elections in small municipalities. Moreover, population data has been applied in order to construct the local unemployment variable. The descriptive statistics of the explanatory variables are displayed in table 4 below. The descriptive statistics for the municipal council and executive board data can be found in table C.1 and C.2 in Appendix C. The number of observations for the national economic indicators is low relative to the other variables. As discussed before, the low number of time periods is one of the limitations to this study. What should be noted from table C.1 and C.2 is the lower number of observations for some of the local political parties because for them data is only available for the most recent election(s).
30 amalgamations (Dutch: gemeentelijke herindelingen) published by Statistics Netherlands have been used (Statistics Netherlands, 2016).
Table 4, Descriptive statistics explanatory variables
6. Results
This chapter will publish the results of the estimations. Moreover, it will interpret the results and discuss their implications. Firstly, the results of the empirical specification of equation 2 will be discussed. Secondly, this section will turn to the results of the estimation of the empirical specification of equation 4 and will discuss some additional robustness analyses. Finally, the overall results will be analyzed and reviewed.
6.1 Estimation results hypothesis 1
The first empirical results to be discussed are those for the estimations with the vote share of the local incumbent as the dependent variable. These estimations are performed in order to test the first hypothesis. Results of these fixed effects estimations are published in table 5 below. The first column contains the results for the estimation that does not include national economic variables as explanatory variables, instead this estimation includes year dummies. The second and third columns contain the results for the estimations that only include national economic variables and do not include local unemployment and year dummies. The fourth column presents the results for the estimation that includes variables with respect to both local and national economic performance. In all four model specifications the coefficient for the control variable measuring political fragmentation is insignificant, whereas the coefficient related to population size is significantly negative.
Variable Observations Mean Std. Dev. Min. Max.
Number of WW-receivers 9666 734 1363 10 23365
Number of ABW-receivers 9269 961 3629 0 63935
Population (total) 10075 39640 60341 921 810937
Population (age 15-65) 10075 26831 42276 554 588743
National economic growth (%) 26 2.16 2.00 -3.8 5.1
31 In the model specification with the vote share of the local incumbent as the dependent variable and local unemployment as the main explanatory variable, i.e. column 1, no statistically significant evidence for the occurrence of local economic voting appears to exist. The marginal effect of local unemployment on the vote share of the local incumbent is defined as follows:
𝜕 𝑉𝑆𝑖,𝑦
𝜕 𝐸𝐶𝑂𝐿𝑖,𝑦−1= 𝛽2+ 𝐴𝐹𝐹 ∗ 𝛽4
Table 5, Estimation result of equation 2. Dependent variable is the vote share obtained by the local incumbent (1) (2) (3) (4) Local unemployment 0.636 0.087 (1.63) (0.31) Affiliation (%) 0.041 0.129** 0.106*** 0.120*** (1.58) (2.14) (2.70) (3.07)
Local unemployment * Aff (%) -0.080 -0.983**
(-0.19) (-2.17)
National unemployment 0.005 -0.002
(1.29) (-0.63)
National inflation 0.017*** 0.012**
(2.88) (2.09)
National economic growth 0.006** 0.004
(2.31) (1.48)
National unemployment * Aff (%) -0.012* -0.012**
(-1.84) (-2.15)
National inflation * Aff (%) -0.012 -0.029***
(-1.11) (-2.70)
National economic growth * Aff (%) 0.016*** 0.022***
(4.05) (4.68)
Population -1.84E-06** -3.87E-06*** -1.85E-06*** -2.23E-06***
(-2.56) (-5.26) (-2.74) (-2.91)
Political Fragmentation 2.43E-04 -0.005 -8.24E-04 -0.001
(0.05) (-1.03) (-0.18) (-0.25)
Number of observations 2075 2075 2075 2075
Number of municipalities 402 402 402 402
F-statistic 20.03 [0.000] 9.17 [0.000] 30.00 [0.000] 23.11 [0.000]
- Dependent variable is the vote share obtained by the local incumbent (based on the number of council seats) - T-statistics in parentheses, P-values in square brackets
- *** 1% level of significance, ** 5% level of significance, * 10% level of significance
- Municipality and year fixed effects for column 1, municipality fixed effects only for columns 2, 3, and 4 - Heteroskedasticity and serial correlation robust standard errors (clustered by municipality)
32 Which has a standard error of:
𝜎̂ 𝜕𝑉𝑆𝑖,𝑦
𝜕𝐸𝐶𝑂𝐿𝑖,𝑦−1
= √𝑣𝑎𝑟(𝛽̂2) + 𝐴𝐹𝐹2𝑣𝑎𝑟(𝛽̂
4) + 2𝐴𝐹𝐹𝑐𝑜𝑣(𝛽̂2𝛽̂4)
In order to give an exact estimate of the effect of local unemployment, the marginal effect and standard error should be calculated for every possible value of the affiliation variable. This can be done by applying equation 5 and 6 specified above, and by using information from the variance-covariance matrix17. Table 6 below indicates the effect of local unemployment on the local vote share for the different values of the affiliation variable. The marginal effects to be discussed in the remainder of this section have been derived in the same manner. The marginal effects displayed in table 6 are all positive, and only significant at the 10% level for values of the affiliation variable equal to 33.33%, 50% and 66.67%. Therefore, no evidence was found in favor of the local economic voting hypothesis, which predicts a negative relationship between local unemployment and incumbent vote share. If anything, the results indicate a positive relationship between local unemployment and votes received.
Table 6, Marginal effect of local unemployment on vote share of the local incumbent for different levels of affiliation between local and national government (Table 5, column 1)
Affiliation (%) Marginal effect Standard error t-statistic p-value
0 0.636 0.390 1.63 0.103
33.33 0.609* 0.343 1.78 0.075
50 0.596* 0.340 1.75 0.080
66.67 0.582* 0.352 1.66 0.098
100 0.556 0.414 1.34 0.180
*** 1% level of significance, ** 5% level of significance, * 10% level of significance
However, the estimation results from column 4 of table 5 differ in some respects. The 𝛽4 coefficient is significantly negative in this estimation. The total marginal effect of local unemployment on the vote share of the local incumbent is found to be negative and statistically significant at the 5% level for values of affiliation of 66.67% and 100% (see table D.2 in Appendix D). These significant marginal effects for high levels of political affiliation signal the potential
17 Brambor (2005: 8). The standard error of the marginal effect can be computed with the use of the
variance-covariance matrix, which is available in Stata.
33 existence of a coattail economic voting effect hypothesis. However, no definitive claims with respect to coattail economic voting can be made as of yet, since this requires to consider the vote share of the national incumbent in municipalities which are not politically affiliated with national office. This will be done in the next section.
With respect to the national economic indicators, some findings should be highlighted as well. Specifically, the marginal effect for high levels of affiliation are interesting as they can be interpreted in relation to the second-order economic voting hypothesis. From the second column of table 5 it follows that national unemployment has a significantly negative effect on the local incumbent vote share for municipalities that are 100% affiliated with national office. Moreover, the results from column 3 indicate a significantly negative effect of national unemployment for levels of affiliation that are greater than zero (see table D.3 and D.4 in Appendix D).
Interestingly, according to the second column of table 5, the effect of inflation is significantly positive for levels of affiliation smaller than 100% (see table D.5 in Appendix D). Moreover, the results from column 4 indicate that national inflation has a negative effect on the vote share of the local incumbent for affiliation levels equal to 66.67% and 100% (see table D.6 in Appendix D). With respect to national economic growth, the marginal effects from the estimation of column 3 indicate a statistically significant (at the 5% level) positive effect for the full range of possible values of the affiliation variable (see table D.7 in Appendix D). Moreover, for column 4 the marginal effect of national economic growth is positive and statistically significant (at the 1% level) for positive values of affiliation (see table D.8 in Appendix D).
These results appear to be indicative of the existence of a second-order economic vote effect. According to the second-order economic voting hypothesis, parties in national office are punished or rewarded locally for the performance of the national economy. The results discussed above signal an effect of the national economy on the local incumbent vote share primarily in cases for which there is a positive degree of affiliation between local and national office. This holds for national unemployment as well as national economic growth, and to a limited extent for national inflation.
