All Bark and no Bite? Coalition/Opposition decision-making in European Municipal Councils

(1)

All Bark and no Bite?

Coalition/Opposition decision-making

in European Municipal Councils.

Rafael Diogo Carrilho (s2498758)

Leiden University

Faculty of Governance and Global Affairs

A thesis submitted for the degree of:

(2)

i

ii

5.1: Reasons for the experiment ... 39

5.2: Results ... 39 5.3: Academic implications ... 40 5.4: Limitations ... 41 5.5: Further Research ... 42 Bibliography ... 44 Appendixes ... i

Appendix 1: Data clean-up R-Script ... i

Appendix 2: Data treatment R-Script ... vi

(4)

1 Chapter 1: Introduction

1.1: General introduction

The COVID-19 virus reached Europe, and (as of writing this) it served as a cataclysmic scare to many countries to implement multiple measures to not only combat the spread of the virus, but also to address the economic consequences of the virus’ impact worldwide (OECD, 2020). Because of the virtually unprecedented nature of a wide-scale pandemic in modern times, there is a lot of debate and uncertainty revolving around it. Political decision-makers find themselves in a place where they must face a lot of uncertainty surrounding the crisis, while balancing its many aspects in terms of consequences, and exert measures to aptly address the situation. This is perhaps a great opportunity to understand how decision-makers approach such a complicated task.

Interestingly, many measures are applied at a local, municipal level, especially those primarily focused on dealing with the virus’ future aftermath (Euronews, 2020). Due to the plethora of municipalities across Europe, this should be taken as an opportunity to further our understanding of political choices in such crisis-ridden situations. Municipal councils work as good example of representative democracy across the entire continent, and often display many interesting, noteworthy behaviours while reacting to major world events, even if they exist at a very localized and low level, administratively.

There are three main components in this experiment. The first is the Preferences variable shows the policy preferences of participants between more abstract, and more specific policy proposals. Afterwards, there is the Treatment variable, which presents participants with negative economic prospects. Lastly, there is a moderator variable between the two others: The Coalition/Opposition variable. This variable determines which participants are a part of a party in the Coalition, and which ones are not.

1.2: Research question

Studying the interaction between a pre-established focus of one given crisis and policy preferences regarding specificity could give insights on how decision-makers make their choices. However, it is also important to take into account that there are more factors that can affect the relationship

(5)

2

between the initial focus and the policy outcome, and perhaps, one of these factors may be the division between coalition and opposition. By looking at academic literature on the topic, one could believe that the dichotomy between Coalition council members and Opposition council members would generate different circumstances

The focus of this thesis is to be able to answer a simple, yet central question: Does a focus on negative economic prospects lead to a preference for more/fewer specific policies, and does being in an opposition/coalition party influence this effect?

1.3: Relevance

Firstly, literature in the academic field surrounding decision-making is already very popular. The bulk of the research, however, often focuses on voter patterns rather than on political decision-makers (Bauer, 1990; von Mettenheim, 1990) which although interesting, often fails to include the dynamics between political decision-makers themselves. This experiment seeks to fill out this lacking part of academia, by abstaining from studying voters’ tendencies, and instead focusing exclusively on the inner workings of political decision-makers’ rationales and choices.

Secondly, there are a limited number of studies that use experiments on a pan-national scale. (Levitt, 1996; Bauer, 1990; Wald, 1983). This is where the ongoing pandemic becomes relevant: the COVID-19 virus’ spread an impact has been very similar among European states, creating a common problem that impacts various communities at the same time. Because of this simultaneous homogenous world issue, it is possible to compare multiple instances of decision-making from multiple countries; this experiment does not fully focus on a country in particular, but the general factors and effects visible within the accumulated data. It does, however, also consider alternate factors (like language or political wing) to set the differences between Coalition and Opposition municipal council members into relative context. Municipal councils are also an interesting basis for an experiment as they are plentiful, usually not very hard to contact, and can serve as a microcosm of elected decision-making.

Thirdly, in practical terms, the results could also serve as a basis for a better understanding of real-world events. Understanding the impact of such a factor (being in the opposition, or the coalition) in political decision-making is a worthwhile endeavour, as it would better let, among others,

(6)

3

journalists and voters better evaluate decisions made by elected individual decision-makers. The rationale behind different decisions may be something that those interested in the topic may think differently while approaching their own analysis. Decision-makers in comparable positions may also use this study to reflect on how their own decisions are affected (knowingly or otherwise), and how their peers’ decisions are taken (which may be used for strategic purposes).

1.4: Thesis outline

To provide a valid answer to this central question, a questionnaire-based experiment was conducted in 5 countries in Europe: France, the Netherlands, Spain, Switzerland, and the United Kingdom; this questionnaire was aimed at local-level politicians. Each country was provided with a questionnaire (except Switzerland, who was provided with 3, in French, German, and Italian respectively), including a text-based treatment. In total, there were 3312 registered answers from all 5 countries.

The experiment, stemming from the survey, relies on three variables. Firstly, each participant was either given a text with general information about the pandemic (a placebo text), or instead was given a text focusing on negative economic prospects (the treatment text), thus separating participants between two groups, the control- and treatment-group; and formulating the Treatment variable. The survey also recorded their policy preferences, regarding how abstract/specific they are, in both economic and social affairs. This thus generates the Treatment variable. By combining the two, it is possible to estimate the effect that a focus on negative economic prospects has on decision-makers’ preferences for abstract/specific policies. Additionally, participants were asked if they belonged to a Coalition or an Opposition party, which creates the last necessary variable of the experiment: The Coalition/Opposition variable. With this third variable, it is possible to estimate the moderating effect this variable may have on the interaction between the two earlier variables.

This thesis participates in and contributes to ongoing academic discussions . The experiment is quite unique, mainly due to the unprecedented situation generated by the spread of the COVID-19 virus, and as such it should provide useful insights on the inner workings of decision-making on a greater scale.

(7)

4

This thesis is divided into six chapters. The first chapter is the one you currently are reading and serves as an introduction to the rest of the thesis. The second chapter combines pre-existing academic knowledge to generate a theoretical framework, upon which the research design is based upon. The third chapter discusses the research design and conceptualizes the variables’ materialization, the setting of the experiment’s hypotheses, and finally, topics such as data clean-up, and data treatment. Afterwards, the fourth chapter deals with the empirical findings from the experiment and how to best interpret them. The fifth chapter analyses the results of the experiment and evaluates how well the hypothesis hold up when faced with the results. Lastly, the sixth chapter concludes the thesis by studying the implications of the experiment’s final verdicts in academic and practical terms, as well as what direction further research efforts could go on the matter.

(8)

5 Chapter 2: Theoretical Framework

2.1: Chapter introduction

The purpose of this chapter is to present relevant academic theories and combine them to form a solid framework within which the experiment can take place. In order to fulfil this goal, this chapter is composed of 6 parts, the first of which serves as the introduction to the chapter.

The second part discusses the theoretical modelling of the decision-making process. It also discusses how the introduction of negative economic prospects may affect the process, and lead to different outcomes.

The third part discusses how these outcomes are theoretically analysed. It focuses primarily on discussing how responses to one same problem serve as a measurable, and reliable outcome variable in the case of this experiment.

Afterwards, the fourth part further develops on the conceptualization of outcomes, focusing on how preferences within individuals’ responses may reflect different decision-making rationales. This part also further contextualizes the experiment in a political setting, so as to demonstrate how preferences work in this setting in particular.

As for the fifth part; this chapter introduces the first two hypotheses of the experiment. These hypotheses pertain to whether or not the introduction of negative economic effects has any impact on the preferences of an individual (and, by extension, their preferences). Their main focus o The sixth part then discusses the last important factor of the experiment: The Opposition/Coalition variable. Stemming from the theoretical framework revolving around decision-making and rationale, this part explains how the division between opposition party members and coalition party members may have a moderating effect in the relationship between the Treatment and the Preferences variables.

Lastly, the seventh part introduces the last two hypotheses of the experiment, which, just like the first two hypotheses, are also based on pre-existing academic literature. These hypotheses

(9)

6

presuppose the existence of an effect between introducing negative economic prospects and showing different preference.

2.2: Negative economic prospects in the decision-making process

The importance of negative economic prospects, and its impact on decision-making, can only be understood via a framework of the decision-making process individuals go through. This process begins with the simple question: What is the problem? (Stone, 2011). The answer to this question, depends on the first step of the process: problem framing (Stone, 2011).

Problem-framing involves firstly a gathering of all different elements related to one issue; a hunt for all available knowledge (Stone, 2011). Once all information has been gathered, individuals organize all the different elements in two groups: relevant, and irrelevant (Stone, 2011). This division among all different elements creates the concrete problem itself: Which outcomes are being considered, which collateral damages are up for discussion, what level of change would be a success, etc… (Stone, 2011). A different focus (meaning, a change in what is considered relevant, and what is considered irrelevant) can therefore greatly impact the shape of the problem. The shape of the problem determines not just what options are considered viable, but also what a successful solution is; virtually everything is (at least partially) dependent on a problem’s framing.

Framing alone does not serve to fully explain the decision-making process, and the role that a focus on negative economic prospects would have. Framing of an issue also involves the concept of feedback responses (Stone, 2011; Leonard, 2018). Feedback-based responses suppose already a framing (used to provide the feedback), and mainly deal with the second step of the decision-making process: assessing available options. Via the framing, it should be clear what the ultimate goal an individual has is, and how it is measured (Leonard, 2018). In the context of the ongoing pandemic, this can be, for instance, how many countries (like the Netherlands) pay close attention to COVID-19’s Basic Reproduction Number (or R0).

Decisions tend to take two routes: if the feedback is positive (as in, the goals are properly attained), then decision-makers will attempt to find a way to make the process more efficient (for example, should the R0 be satisfactorily low, policy-makers may consider relaxing some preventive measures to make the approach more efficient) (Leonard, 2018). However, should the feedback be negative,

(10)

7

a decision-maker tends to look for alternative measures to be taken, to improve the result (Leonard, 2018). In this case, feedback already serves as the original framing of the problem at hand: it determines what parts are important, and what parts are negligible by default; further cementing the importance of the problem conceptualization.

The conceptualization of a problem is highly dependent on the aggregation of information, to afterwards decide which elements are relevant, and which are not, and the more of it there is, the easier and clearer the problem becomes (Stone, 2011). Similarly, should feedback be very extensive and precise, then the path towards resolution is also made easier (Leonard, 2018). In the case of this pandemic, the opposite is true: there is very little certainties regarding the multiple issues pertaining to it, thus providing little to no clarity. Bounded rationality thus becomes a considerable factor in the decision-making process: The way rational individuals take decisions is often reduced to simple input-output models, where information about an issue goes in, and decisions come out (Simon, 1972). However, much like in computer science or statistics, the concept of “garbage in – garbage out” also applies; if information is ambiguous and/or uncertain or even unclear, then decisions will ultimately just as ill-aimed as the information (Simon, 1972). This is particularly important for the conceptualization of problems, because the selection of relevant and irrelevant elements is thus far less certain, and far more open for other factors (Stone, 2011). Effectively it creates blurrier predetermined lines as to “where” the problem is, and thus it gives “carte blanche” to decision-makers to decide it for themselves (Stone, 2011).

This is where the introduction of negative economic prospects plays a role. Arguably, the reasoning for retrospective feedback should also apply for forward-looking feedback (or, rather, prospects), and as such the conceptualization of a problem is already prefabricated and not let up to individuals to decide (Leonard, 2018; Stone, 2011; Simon, 1972). Providing negative economic prospects automatically sets the problem as being primordially an economic one, before anything else. Negative economic prospects thus make up the Treatment variable: Experiment-wise, this creates a control group that has a blank slate to approach the pandemic as they wish, and a treatment group that was provided with negative economic prospects as a focus, removing thus the option to frame the problem any different way.

(11)

8

2.3: Different responses for one same problem

As discussed in the previous part, one same problem can be analysed in different ways, resulting in different conclusions among individuals (Stone, 2011). The part of decision-making that this thesis takes a particular interest in is the problem conceptualization stage: By introducing negative economic prospects as an initial framing, rather than giving individuals free reign on the matter. Individuals are not perfectly alike one another, as they may differ in aspiration level. An individual’s aspiration level is the element that determines what the most rational and efficient course of action will be to achieve a goal (Gigerenzer, 2001). The process that determines the aspiration level is, in essence, the determination of personal goals, which can change overtime and change based on performance feedback (Gigerenzer, 2001). Different aspiration levels can serve to explain why individuals may diverge in their approach to decision-making: depending on their goals, decision-makers may follow different choice-making patterns (Diecidue and van de Ven, 2008). Aspiration levels also take into account how risk-averse (or risk-tolerant) an individual is, and how much power an individual exerts within the given circumstances; both of which also changes how they may analyse all different options, after the problem conceptualization stage (Otjes and Louwerse, 2012; Diecidue and van de Ven, 2008).

An example of how one same problem may be framed the same, yet conclusions may be different, is the “trilemma of the services economy” (Iversen and Wren, 1998). Politicians must choose two out of three goals that are achievable simultaneously: wage equality, plentiful employment, and budgetary restraint (Iversen and Wren, 1998). Virtually all politicians would agree that these are indeed important, however, depending on their aspiration levels (in this model, determined by their own political views), they will choose two goals over the remaining one (Iversen and Wren, 1998). Neoliberal politicians prefer fiscal discipline and plentiful employment, at the detriment of wage equality; Christian-Democrats would prefer fiscal discipline as well but would prioritize wage equality over plentiful employment, and Social-Democrats would opt for plentiful employment and wage equality, even if that means neglecting fiscal discipline (Iversen and Wren, 1998). The aspiration goals, in this model, can be traced to political preferences that are already present even before a problem is conceptualized. When faced with the framework of a trilemma, the aspiration level of each individual plays a determining role in choosing which one of the three goals to

(12)

9

abandon, and which two to pursue; thus providing a good reference as to how important differences in responses are, when analysing the impact of different factors in decision-making.

Based on the combination of these complementary decision-making models, it becomes clear that there is an analysis individuals do even before conceptualizing a process: they evaluate their own goals, the risks they are willing to take to achieve them and the power they withhold (Stone, 2011; Gigerenzer, 2001; Diecidue and van de Ven, 2008). All of these analyses still are subjective to change, based on later stages of the decision-making process (Gigerenzer, 2001). This makes the difference in responses the main measurable outcome, when looking into how a pre-set problem framing (in the case of this experiment) may change the entire decision-making process.

2.4: Different rationales and types of responses

The rationale used for approaching problems can have a great deal of influence in the way that individuals take decisions. The theoretical framework thus far assumes that all individuals solely aim to solve the problem, but research shows that ulterior motives also play a role. There are two main rationales that an individual may choose to approach a problem: a problem-solving rationale, or a self-enhancing rationale (Jordan and Audia, 2012).

A problem-solving rationale, often times assumed to be the default one, gives the individual the main motivation to improve performance (Jordan and Audia, 2012). This is valid for when performance is above aspiration levels, but also for when performance is below aspiration levels (Jordan and Audia, 2012). It is important to note that the standards of evaluation are consistent all throughout the process, and the priority of performance goals is equally fixed (Jordan and Audia, 2012). The primary temporal orientation is also important: the reasoning behind each decision is the main motor guiding to conclusions (Jordan and Audia, 2012); in other words, each choice is analysed for its potential in responding to the problem, and then the process concludes by choosing the best one.

The self-enhancing rationale is very different. Under this rationale, an individual’s main motivation is not to improve performance, but instead present oneself in a positive light by assessing performance as satisfactory (Jordan and Audia, 2012). Both the standards and the priorities of performance goals are fluid and can shift with time (depending on what is more favourable for the

(13)

10

individual in question) (Jordan and Audia, 2012). But the most crucial contrasting point would be the primary temporal orientation: individuals that act under a self-enhancing rationale retrospectively analyse their choices and motivate their reasoning (Jordan and Audia, 2012); or in other words, decide now, explain later. This last element is best summed up by the concept of intended rationality (Jones, 2001).

At this point it becomes important to talk about the object of study: municipal council members. As politicians, they engage in campaigning in order to justify their political choices to their lectorate (Boin and t’Hart, 2001). This process goes hand-in-hand with the differences in the primary temporal orientation; the way they justify a choice is entirely different if they have a problem-solving or a self-enhancing rationale. Explaining their changes to policies (especially when it comes to big reforms like those needed due to the COVID-19 pandemic) is a campaigning process that especially takes voters into account (to secure enough votes next elections (Boin and t’Hart). This process is heavily influenced by a major element: language (Hart et al., 2004). At first sight, it would be easy to argue that political culture and different contexts would make this a moot point as the usage of language would vary greatly; however, research shows that the core elements of the usage of language as a political tool remain the same throughout countries quite solidly (Wodak, 1989).

The usage of language for political campaigning can serve as an indicator of an individual’s rationale, for the purposes of this experiment. The choice in specificity in policy and its presentation is something that is vital for political image-shaping (Hart et al., 2004; Stone, 2011). Generally speaking, more abstract policies are usually geared to image-shaping (campaigning) purposes, rather than problem-solving ones, and the opposite applies to more specific policies (Hart et al., 2004; Stone, 2011). Coupling these notions with the framework of the two rationales, the type of language used for policies can indicate which type of rationale an individual is using: abstract language is coupled with a self-enhancing rationale, and specific language is coupled with a problem-solving rationale. Abstract language is more flexible and allows an individual the flexibility required to pursue self-enhancing strategies (such as having a way to flexibly switch the evaluation criteria, and justification method) (Jordan and Audia, 2012; Hart et al., 2004). On the other hand, more specific language matches a problem-solving rationale as it stems from the

(14)

pre-11

set evaluation criteria and is an exact answer to the problem at hand, even at the cost of less flexible campaign potential (Hart et al., 2004; Jordan and Audia, 2012).

This thus further refines the Preferences variable of the experiment. Previously it was established that the individuals’ answers to their policy preferences would serve as an outcome variable, however now this is further refined that the unit of measure is their preferences for either abstract or specific policies.

2.5: First two hypotheses

Having established both the Treatment and Preferences variables, the first two hypotheses can be made about the main effect between both variables.

The first hypothesis builds upon pre-existing models on decision-making rationales. By introducing a pre-set problem framing, via negative feedback, the main theoretical expectation would be that this would push individuals towards a problem-solving rationale (Jordan and Audia, 2012). The cause for this would be that the introduction of a pre-existing well-defined goals incentivizes individuals to work towards them, even when receiving the same level of performance feedback (Jordan and Audia, 2012). Thus, the first hypothesis is: Being presented with negative economic prospects causes individuals to show a preference for specific policies over abstract ones. The second hypothesis, on the other hand, is the direct antithesis of the first. In this case, by introducing a pre-set problem framing, via negative feedback, the expectation (opposite to that of current theory) would be that this would push individuals towards a self-enhancing rationale. In this case, the causation would be that, because of great uncertainty and the importance of political campaign for municipal council members, individuals may prefer more abstract policies as they are far easier to justify to the public; whereas if there is no provided framing, individuals may choose more specific policies as they more easily control the rest of the elements surrounding it. The second hypothesis is thus: Being presented with negative economic prospects causes individuals to show a preference for abstract policies over specific ones.

These two hypotheses cover the existence of an effect that the Treatment variable may have on the Preferences variable, by being antithetical and mutually exclusive. However, they are both

(15)

12

falsifiable, and even simultaneously so: should there be no difference in the preferences individuals show, both hypotheses are false and decidedly there is no correlation between the variables. 2.6: The opposition/coalition variable

This experiment aims further than just researching the effect of negative economic prospects on individuals’ policy preferences. A new, moderator variable is needed, to further understand the relationship between the Treatment and Preferences variables.

The structure in which individuals find themselves is an important factor in decision-making (Ringe, 2005). The power an individual holds, compared to others within the structure, leads an individual to evaluate their situation differently, thus influencing their aspiration level and the entire of the subsequent decision-making process (Ringe, 2005; Gigerenzer, 2001). In the case of municipal council members, the power division is very clear cut: there is an opposition and a coalition. The coalition holds the most power, as they are usually the individuals who (due to their size in numbers) can pass policies, or stop proposals from the opposition. The way parties act is partially also a performance act for their voters, as a part of campaigning (Farrel et al., 2002; Boin and t’Hart, 2004).

At often times, self-enhancing individuals will choose for performative change, rather than real chance, to boost their image (rather than solve the problem) (Jordan and Audia, 2012). Specifically, a belief that an individual’s ability to solve an issue is fixed, may push that same individual more towards a self-enhancing rationale more than an individual without such a belief (Jordan and Audia, 2012). Opposition members, by the logic of an opposition, hold less power than a Coalition due to being the minority of the representative organ they belong to; therefore it’d be easy to claim that they indeed may face both an image and an internal belief that they do not have the ability to fix the issue. The same, however, can be argued for Coalition members, who, because of mass uncertainties surrounding the pandemic and the importance of elections, may also feel like they do not have the ability to adequately handle the situation, and thus opt for a self-enhancing rationale. Because of this importance of campaigning, and power divisions, it is plausible that individuals in the opposition and in the coalition may not only show different preferences in general, but more importantly, this can impact the effect that the introduction of negative economic prospects has. It

(16)

13

is, however, important to note that there is, at least to the extent of current academic decision-making models, no clear way to predict the actions of either group. This third and final variable, will be the Coalition/Opposition variable.

2.6: Remaining hypotheses

The first two hypotheses focused mainly on the existence of an effect in policy preferences that negative economic prospects may have. The last two hypotheses, on the other hand, are focused on the effect that the opposition/coalition dichotomy among individuals may have on this effect. Because the last two hypotheses are targeted on the effect of a third variable on the effect, they presuppose the existence of an effect to begin with. Should both first hypotheses be false (thus there being no effect), these hypotheses would directly be invalid and unusable.

There are three elements that can make the dichotomy between Opposition and Coalition members have an impact on the effect of the Treatment variable over the Preferences variable. The first of these is the power difference between Opposition and Coalition members. As explained earlier, the amount of power an individual holds plays a role in whether they opt for a problem-solving or a self-enhancing rationale (Jordan and Audia, 2012). Opposition and Coalition members directly have a different power; lower power (or rather, ability) to address the problem pushes individuals to use a self-enhancing rationale (Jordan and Audia, 2012). Thus, there are already some initial circumstantial divisions related to power.

Interestingly, there are also differences in risks for individuals, which serves as the second element. Coalition members, as they hold more power, are also expected to provide apt responses to the problem at hand, but this expectation also means there is a higher risk for the Coalition members. The importance of securing votes next elections is a concern for Coalition members: underperformance results in disappointed voters reconsidering voting for them again, thus there is a greater urgency to solve any negative feedback if possible. Greater risks, however, are a factor that pushes individuals towards a self-enhancing rationale as well (Jordan and Audia, 2012). There are therefore factors for both sides that pushes them towards a self-enhancement rationale, but it is not clear which one is stronger, and how these factors affect decision-making with the introduction of negative economic prospects.

(17)

14

The third and last element concerns political campaigning strategies, and how either side may act, depending on if they are a part of the Coalition or the Opposition. Starting with the Coalition, individuals in the category may be guided towards a repetition of their own policy preferences as a justification of them (rather than reconsider them); a campaign tactic that’s been seen in the past (Hart et al., 2004; Boin and t’Hart, 2001; Jones, 2001). What this means is that there may be a lower effect from Coalition members, based on campaigning-motivated justification of pre-existing policy preferences. For the Opposition, on the other hand, the same applies, but instead of justifying a party programme being currently applied, they may be seeking to portray themselves as a better alternative than the Coalition (Boin and t’Hart, 2001). In this case, they make use of language in terms overgeneralizations and oversimplifications (coinciding thus with more abstract policies) to attract voters, as easy arguments that are not easily addressed by opponents due to the complexity of the pandemic (Hart et al., 2004; Boin and t’Hart, 2001). Thus, it seems that, at least theoretically, the weight of campaigning can heavily influence the effect of the Treatment variable.

To understand how these different elements work in relation with the introduction of negative economic prospects, two hypotheses were made:

1. Being a part of the Coalition causes the effect of the Treatment variable to be smaller 2. Being a part of the Opposition causes the effect of the Treatment variable to be smaller Similarly to the first two hypotheses, they are opposite and therefore mutually exclusive, they cannot both be true. But should the Opposition/Coalition variable have no impact on the relationship between the Treatment and Preferences variable, both hypotheses are false. By determining which one (assuming one is bound to be true), it then becomes possible to understand how power-differences, risk-differences, and campaign-differences affect decision-making.

(18)

15 Chapter 3: Research Design

This chapter focuses on the practical operationalization of the thesis’ experiment, it focuses on the operationalization of all-important variables, and the way the data is processed. In order to cover these topics, this chapter is composed of 7 parts, the first of which is, again, the chapter introduction.

The second part discusses the practical conceptualization of the Treatment variable, followed by the third and fourth parts, that discuss the practical conceptualizations of the Coalition/Opposition variable, and the Preferences variable respectively.

The fourth part deals with the set of methods which analyses the questionnaire data and quantifies the statistical relationships between the variables. The results of such methods are presented and discussed in detail in Chapter 4.

Afterwards, the sixth part describes the ways under which the questionnaires were carried out for different countries. This includes data on the responsiveness from different countries, but also the methods used to reach municipal council members (and how many in total were reached).

Lastly, the seventh part discusses the finalized dataset, after being refined and cleaned up. There is also data provided on how the data is distributed regarding key factors, as well as a discussion on how valid the dataset is (and how this may affect the validity of the experiment as a whole). 3.2: The Treatment variable in practice

In its most basic form, an experiment involves three components: a test subject, a treatment, and an outcome. This allows to test how a test subject’s outcome changes because of a treatment. To be able to estimate the effectivity of a treatment, the counterfactual is also needed (as in, what would happen if there were no treatment at all), thus, there’s the need for a Control and a Treatment division, to compare the two. The Treatment variable works in this same basic manner: some participants in the questionnaire are a part of a Control group, and the others a part of the Treatment group.

(19)

16

The questionnaires had a randomizer feature built-in: each participant would randomly be assigned either the Control or Treatment version of the first question. The first question involves a text, and subsequently a question about the participant’s own views regarding the topics discussed in the text. This question does not serve any purpose besides easily distinguishing between Control and Treatment participants (if they answered one version of the question, they cannot have possibly answered the other).

The Control text provided mainly health impacts and general information regarding the COVID-19 pandemic in the world. This information already is fairly well-known, and provided as just a generic text that, by design, should not impact the participants’ initial focus. The Treatment text, on the other hand, was designed with the specific intent of guiding the participants’ focus onto negative economic impacts of the COVID-19 pandemic.

Questionnaire participants are exposed to, in essence, a strong stimulus (the Control/Treatment texts) that serve as a “priming” factor (Krpan, 2017). Priming is providing individuals with a stimulus (be it big or small) to guide them towards different behavioural patterns (Krpan, 2017). In this case, the priming effect puts the focus entirely on one of two sides of the COVID-19 pandemic’s repercussions in the world at large.

In order to distinguish which participants were in the Control group, and which ones were in the Treatment group, the first questions’ answers are taken into account. If a participant answered the Control version of the question, they automatically are in the Control group; the same applies for the Treatment group, composed of those who answered the Treatment version of the question. Some observations did not have any answers to either question, making it impossible to determine which group a participant belongs to; regrettably, because of the importance of this distinction, these observations were removed.

3.3: The Coalition/Opposition variable in practice

One of the questionnaire questions was: “What is your position within the municipal council?” and it included four possible answers:

1. Mayor

(20)

17

3. Member of an Opposition Party or the Minority 4. Not Applicable

Questionnaire participants that answered either option 1 or 2 are a part of the Coalition group, whereas participants who answered option 3 are a part of the Opposition group. All participants that responded with option 4, were left with an “N/A” value instead.

3.4: The Preferences variable in practice

The Preferences variable is based on not one, but two questionnaire questions. Both these questions presented participants with 6 different policy proposals (in no particular order), making a total of 12 policy proposals. One question proposed policies pertaining to socially oriented policies, and the other pertaining to economically oriented ones. This division was primarily due to the survey’s design, and although it has no bearing theory-wise, hypothesis testing is repeated for all three groups. Three of these policy proposals in each question were deliberately abstract, and the other three were deliberately specific. For each policy proposal, participants were asked how far they believe the policy should be implemented in their municipality. The proposed policies were:

Table 1: Proposed Preferences per Type of Question and Specificity

Question type Specificity Policy proposal

Social Specific Establishing a volunteer program to combat loneliness among elderly Social Specific Improving the capacity of health care by strengthening regional

cooperation

Social Specific Preventing evictions by mediating between landlords and tenants Social Abstract Ensuring the health of citizens

Social Abstract Maintaining/Improving the well-being of residents Social Abstract Guaranteeing the safety of vulnerable citizens

Economic Specific A short-term municipal investment programme to support local entrepreneurs

Economic Specific A one-year tax deferral for all municipal taxes

Economic Specific Audit and/or review the budget to ensure the future sustainability of the municipality’s finances.

(21)

18

Economic Abstract Supporting the local economy

Economic Abstract Keeping the municipality's budget balanced.

Note: Each policy proposal can be either specific or abstract; and belong to either a social or economic question type. In total, each possible combination has three policy proposals

For each policy proposal, participants had 6 options to choose from, which afterwards was then granted a value, to reflect their accord/discord numerically, using a 5-point Likert scale (McLeod, 2019).

The Preferences variable, for each observation, consists of three values. Each value is a numeric reflection of a participant’s preferences either regarding socially oriented policies, economically oriented policies, and the general preference among both.

Taking the Likert scale answers, the values of the answers given to Abstract policies in the Social category are summed together. Afterwards, the Specific policies in the Social category undergo the same process. Then, the result of the Abstract policies is subtracted from the result of the Specific policies. This yields a value that can be within the interval [-12;12]. Should the value be positive, the observation shows a preference for Specific policies, when it comes to socially oriented policies. If the value, however, is negative, it would signify a preference for Abstract policies. The same process is then applied for policies in the Economic category, generating thus the second preference value, that works the same way as the socially oriented policy preferences value, but this time regarding economically oriented.

Lastly, the general preferences value is the simple sum of both the socially oriented and the economically oriented policy preference values. By itself, this would cause a value that theoretically could be anywhere within an interval of [-24;24]. To have this value behave identically to the other two values, it is subsequently divided by two, bringing the interval down to the same [-12;12] interval, and its behaviour directly comparable to the other preference values.

Table 2: Likert Scale Answer Values Answer Value Definitely not 1 Probably not 2 Maybe 3 Probably yes 4 Definitely yes 5 Not applicable N/A

(22)

19

3.5: Testing methodology

With the goal of challenging the pre-set hypotheses in mind, there are multiple analysis that were done for this effect, all of which were based on a two-way analysis of variance (two-way ANOVA). Conceptually speaking, the Coalition/Opposition variable is a moderator variable between the independent variable (the Treatment variable) and the dependent variable (the Preferences variable). The hypotheses indicate that there should be an interaction effect between both the Treatment and the Coalition/Opposition variables, therefore they were used as the two independent variables within a two-way analysis of variance, and the dependent variable was the Preferences variable.

It is important to note that the methodology is split into three distinct phases, each with a different goal in mind, but all designed in the greater objective of analysing the data in an appropriate fashion.

The experiment’s first phase tests the independent variables’ effect on the outcome variable directly. For this, some assumptions must be met, and are tested via the means of statistical analysis (for instance, a series of Levene tests of variance equality). This ensures that the data used does not have any inherent issues and can thus be trusted during the analysis process. Afterwards, using the Least Squares Means (LS Means) method, the contrasts between the Coalition and Opposition groups is estimated, regarding the general policy preferences. The same test is then repeated twice, for socially oriented and economically oriented preferences. Lastly, the three LS Means tests are repeated once again, but instead of estimating the contrasts between the Coalition and Opposition groups, the Treatment variable is used instead, to estimate the contrasts between the Control and Treatment groups.

The second phase uses a two-way ANOVA. The two-way ANOVA is ran using the Treatment and Coalition/Opposition variables as independent variables, and as a dependent variable the general policy preferences of survey participants. Subsequently, the two-way ANOVA analysis is repeated two more times, but using specifically the economically oriented and socially oriented policy preferences. The results of all three two-way ANOVAs show the extent of which there is an interaction effect between the Treatment variable and the Coalition/Opposition variable.

(23)

20

Should there indeed be an interaction effect between the two variables, there is a post-hoc test to estimate the effects of different combinations of groups. This test uses the Tukey Honest Significant Differences adjustment, because it allows to easily estimate the statistical significance of the interaction effect. The post-hoc test is repeated a total of three times, for socially oriented policy preferences, economically oriented policy preferences, and general policy preferences respectively. Should there be no interaction effect, this post-hoc test is not used, as it would not serve any purpose.

Lastly, there is a third phase, to contextualize results from both phases. Other additional variables (language of Swiss participants, country, political wing, gender, and age) are used as an addition to the Coalition/Opposition variable, in order to show how strong other variables are. This allows for a more nuanced reading of the results, as it can show the relative strength compared to other factors. To do so, all available variables are placed into a multi-variable linear regression model, that uses a reference group of each variable to estimate effects.

3.6: Conducting the survey

The survey firstly was designed in English (as a Master version), and afterwards 7 copies were made. 6 of the copies were translated into French, German, Italian, and Spanish, and later attributed to their respective country. Switzerland was the only country that, instead of just one survey, had three surveys, for three of its national languages: French, German, and Italian.

In the meantime, municipal council members’ email contacts were gathered for each of the countries. These contacts were not only direct, personal addresses, but also generalized council ones for some countries.

For France, around 20.000~30.000 municipal council members were contacted via indirect email addresses (because disparities on the size of councils among municipalities, the number is not entirely clear). It is important to note, however, that the timing of this survey correlated with the 2020 French municipal elections, thus it was hard to contact municipal council members. Despite so many individuals being contacted, only 87 responses were recorded from France, making it the country with the fewest respondents.

(24)

21

The United Kingdom saw 14.000 of its municipal council members contacted, all via individual email addresses. No municipalities were contacted through general email addresses. At the end, the United Kingdom’s survey yielded 368 responses.

For Switzerland, in total, 1.434 municipalities were contacted all over the country, among the three Swiss surveys. According to their canton’s language, only one version of the survey was sent to each address. The French-language survey received 330 responses, the German-language survey received 203 responses, and finally the Italian-language survey received 31 responses. In total, Switzerland provided 564 respondents.

Regarding Spain, around 300 municipalities were contacted. Spanish council members were either contacted by individual emails or general council-wide emails. There were circa 5,000 individual email addresses found, and around 200 general council-wide ones. Despite the high number of emails sent, only 563 responses were recorded.

In the case of the Netherlands, the country has in total 8.261 council members, 8.216 were reachable via individual email addresses. The country also provided a surprisingly large number of responses, 2.231 in total.

3.7: Data and its validity

The data clean-up process (see Appendixes 1 and 2) was done with R (R Core Team, 2013), RStudio (RStudio, 2020), and packages Readr (Wickham et al, 2018) and CAR (Wickham et al, 2020).

The data includes 9 variables per each observation. These observations are each participant’s data regarding the three experimental variables (the Coalition/Opposition variable, the Treatment variable, and the Preferences variable). The other variables included are IP address, country, language, gender identity, political wing, and age.

IP addresses were collected automatically as participants filled out the survey. The language variable, instead, was added posteriorly to each survey. All surveys except the Swiss ones were attributed the value “Other”; the Swiss ones instead were attributed “French”, “Italian”, or

(25)

22

“German”, based on the language they were in. All other variables were collected via other questions in the survey (see Appendix 3).

It is important to note that some of these variables display NA values, which would signify that the participant did not answer the questionnaire questions regarding this effect. For the age variable, two values (0 and 4) were removed since they were not plausible ages, and one answer was in the participant’s birth year (1950) rather than their age (70), so this was edited and corrected.

The finalized, cleaned up dataset includes 1958 observations in total. This in turn leads to a new distribution among the different variable as can be seen in Tables 3, 4, and 5.

(26)

23

Table 3: Distribution of Respondents per Country

France 1% (20) Spain 11% (212) United Kingdom 10% (201) The Netherlands 67% (1318) Switzerland 11% (207)

Note: percentages have been rounded. The number of respondents per country can be found within parenthesis.

Table 4: Distribution of Languages Spoken by Swiss Participants

Italian 7% (14) German 35% (73) French 58% (120)

Note: percentages have been rounded. The number of respondents per country can be found within parenthesis. The language variable only applies to the 207 Swiss participants.

Table 5: Distribution of Respondents

Treatment group 53%

(1038)

Control group 47%

(920)

(27)

24

There are some relevant concerns that need to be addressed. These concerns involve primarily the data collection process, but also secondarily the responses received.

Firstly, 1958 observations seems to be very little, especially considering that this data is supposed to represent about a fourth of the European continent. The sample seems big enough to provide meaningful results, but it should still be underlined that the number is unexpectedly low, considering the amount of council members contacted (directly and indirectly).

Secondly, municipal council members’ emails (or, alternatively, the Municipality’s general email) were collected (mostly manually), and later contacted to invite them to participate in the questionnaire; but this automatically excludes any council member who was not reachable by email. The contact collection process was very different among all countries, and in some cases even countries’ subdivisions (for example, it was noticeably more difficult to find any contacts in Galicia (Spain), whereas Valencia was far more accessible). This constitutes a network sampling, which means that the sampling was not entirely random, and its selection mechanism has not been corrected for (Elliot and Valliant, 2017).

As a third point, the distribution of participants is also not consistent throughout all countries. It is important to point out that the Netherlands’ municipal council members, by far, participated the most. Others (notably municipal council members from France, or Italian-speaking Switzerland), participated surprisingly little. This creates a noticeable disproportion, that may affect ensuing results. In fact, it may be that different countries are already different to begin with, meaning that there’d be internal inconsistency.

To check the internal consistency of the data, the package Psych (Revelle, 2020) was used. By using Cronbach’s Alpha on the data, it is possible to check the internal consistency; the results were a standard Alpha of 0.77 (the lower and upper Alphas of 0.7 and 0.75), thus being under 0.8. This, in turn, supports that the data is consistent, and internal inconsistency is not a problem. All-in-all, there are some concerns to be raised about the data, and the validity of the experiment as a whole, as a result (albeit, as demonstrated, excluding internal inconsistency). Whereas these concerns are indeed present, the experiment should still yield plausible results, that can serve to portray reality in a sufficiently accurate fashion.

(28)

25 Chapter 4: Empirical Findings and Analysis

This chapter is dedicated to all data analysis experiment empirical findings. Just like all other chapters, it is divided into several parts (in this case 6 parts), starting with an introductory one. The second part of this chapter deals with important assumptions that need to be addressed before the first phase of the analysis can take place. These include not only logical assumptions that are mainly verified by confirming that the data is set up correctly, but also important statistical tests to ensure that the data’s own contents do not contain any anomalies that may skewer results.

Next, as a third part, the chapter starts with the first phase of the data analysis. This phase studies the effects of the Treatment and Coalition/Opposition variables on the Preferences variable. This is done via three sets of LS Means test, each corresponding to one of the values of the Preferences variable.

The fourth part of this chapter follows the second phase of the data analysis. Via a set of two-way ANOVA tests, the interaction effect between the Treatment and Coalition/Opposition variables is estimated; each of the tests corresponding, again, to each of the values of the Preferences variable. Subsequently, the fifth part of the chapter focuses on the third, and last phase of the data analysis. Using an all-inclusive regression model, the alternate variables are tested, to contextualize how the strength of the Treatment and Coalition/Opposition variables stack up with other factors.

Lastly, the sixth part aggregates and analyses all results from the three phases of the data analysis. Afterwards, this part compares and contrasts the results with the hypotheses set in Chapter 2.

4.2: Important assumptions

Before addressing any of the assumptions, it is important to note that all data manipulations were done with R (R Core Team, 2020) and RStudio (RStudio, 2020) as well, with packages LSMeans (Lenth, 2016), ISLR (James et al, 2017), dplyr (Wickham et al, 2020), leaps (Lumley, 2020), and devtools (Wickham et al, 2020).

(29)

26

The first assumption to address relates to the dependent variable (in this case, the Preferences variable). The Preferences variable must be measured on a ratio level. As explained in Chapter 3, the Preferences variable is a set of three values, within the interval [-12,12]; these values form an interval scale variable, thus this condition is met.

The second assumption concerns the independent variables. The independent variables must consist of two or more categorical groups, each. This is also not a problem for the Treatment variable, because it is a dichotomous variable between Treatment and Control groups; nor is it a problem with the Coalition/Opposition variable, which is a trichotomous variable between Coalition, Opposition, and Not Applicable.

The next assumption relates to observations as a whole: all observations must be individual and must not have any relationship between one another. The dataset also included each observation’s IP address, so to verify this assumption, the amount of IP addresses present was compared to the total number of observations. The results were 1.849 IP addresses for 1958 observations, meaning that only 84 observations had repeated IP addresses. This means that, except for the 109 observations whose IP address is repeated at least once, each participant was connected to a different network altogether, suggesting an entirely different location. The 109 observations whose IP address is repeated at least once can be explained by participants using the same internet connection for filling out the survey, which can be explained by the fact that council members have a common workplace. Additionally, this check also suggests that the chances that one participant may have filled out the survey multiple times in order to tamper with results are very small. Fourthly, the data must contain no outliers. For this, summary data on the values of the Preferences variable was used:

(30)

27

Table 6: Summary Data of the Preferences variable

General policy preferences

Socially oriented policy preferences

Economically oriented policy preferences Min. -7.00 -10.00 -9.00 1st_Quadrant _-2.00 _-3.00 _-2.00 Median -1.00 -1.00 -1.00 Mean -1.03 -1.337 -0.724 3rd_Quadrant _0.00 _0.00 _0.00 Max. 5.00 8.00 8.00

Note: Values are whole numbers encompassed between -12 and 12.

All the values seem to demonstrate that there are no outliers in any of the values, as the preferences tend to be quite even overall. In fact, it is interesting to see that, out of a range of [-12; 12], not only were preferences in all categories slightly veering to abstract preferences, but there does not seem to be very strong preferences for either side of the spectre.

Afterwards, the dependent variable is assumed to be normally distributed for all possible combination of the independent variables’ groups. This can be achieved via a Shapiro-Wilk test for normal distribution. The test was repeated three times, one corresponding to each value of the Preferences variable:

All the p-values for each preference are below 0.05 (by a very large extent), thus there is no reason to believe that abnormal distributions may cause an issue in this case.

Lastly, for each possible combination of the groups from the Treatment and Coalition/Opposition variables, it is assumed that there is homogeneity between all the variances. To check for this, Levene tests were used for each value of the Preferences variable:

Table 7: Shapiro Wilk Test Results

General policy preferences Socially oriented policy preferences Economically oriented policy preferences W=0.99348 p-value=<0.001 W=0.99568 p-value=<0.001 W=0.98638 p-value=<0.001

(31)

28

Table 8: Levene Test Results

General policy preferences Socially oriented policy preferences Economically oriented policy preferences df=3 F-value=0.1653 p-value=0.9198 df=3 F-value=0.4006 p-value=0.7526 df=3 F-value=0.2789 p-value=0.8406

Note: The maximum number of observations (1958) was used

All the p-values are above 0.05, which indicates that all three combinations of groups are homogenous when it comes to variance. Thus, the data meets its final expectation.

4.3: Factor effects

With the goal of estimating the effects that the Treatment and Coalition/Opposition variables have on the Preferences variable, a LS Means design was applied to both variables, having each value of the Preferences variable as an outcome variable. The goal is to see the contrast between different groups of one variable. The set of contrasts between groups focus on the general policy preferences:

Table 9: Contrasts Among Groups for General Policy Preferences.

Contrasts

Estimate SE df t-value p-value

Coalition - Opposition -0.31 0.0806 1608 -3.849 0.0001

Control - Treatment 0.0995 0.0794 1608 1.252 0.2108

Interestingly, both dichotomies present an effect (albeit both effects are very small), however only the difference between the Coalition and Opposition is statistically relevant, as its p-value is below 0.05, whereas the dichotomy between the Control and Treatment groups had a p-value above 0.05 and therefore it cannot be considered statistically relevant. Practically speaking, it seems like being in the Opposition makes decision makers prefer more abstract policies, rather than specific ones. The next set of contrasts focuses on the socially oriented policy preferences:

(32)

29

Table 10: Contrasts Among Groups for Socially Oriented Policy Preferences.

Contrasts Estimate SE df t-value p-value Coalition - Opposition -0.566 0.109 1614 -5.187 <0.0001

Control - Treatment -0.181 0.108 1614 -1.678 0.0936

Once more, the Treatment variable did not have a statistically relevant effect as its p-value is too high. However, the Coalition/Opposition still shows an effect (even if, again, it is very small). Once again, being in the Opposition seems to cause a slightly bigger preference for specific policies than abstract policies.

Lastly, the tests were repeated with a focus on the economically oriented policy preferences:

Table 11: Contrasts Among Groups for Economically Oriented Policy Preferences.

Contrasts Estimate SE df t-value p-value Coalition - Opposition -0.0487 0.108 1608 -0.449 0.6536

Control - Treatment 0.378 0.107 1608 3.539 0.0004

Again, the results are surprising. The effect presented by the Coalition/Opposition variable is not only small but also not statistically relevant as its p-value is too high, and even its standard error is bigger than the estimate itself. The Treatment variable, on the other hand, does indeed have an effect, and it is statistically relevant as its p-value is substantially lower than 0.05. When it comes to economically oriented policy preferences, it seems that being in the Control group causes decision-makers to prefer specific policies over abstract ones.

4.4: Looking for an interaction effect

As mentioned previously, the second phase focuses on looking for an interaction effect between the Coalition/Opposition and the Treatment variable. Two-way ANOVAs were used to check the interaction effect between both the Treatment variable and the Opposition/Coalition variable. Firstly, this was done with the general policy preferences:

(33)

30

Table 12: Interaction Effect Test for General Policy Preferences.

Sum Sq DF F-Value Pr(>F) (Intercept) 480.7 1 190.1372 < 2.2e-16 Treatment Variable 1.3 1 0.5145 0.473288 Coalition/Opposition Variable 21.1 1 8.3552 0.003898 Interaction effect 0.4 1 0.1480 0.700503 Residuals 4062.7 1607

There does not seem to be a plausible interaction effect between both variables, when it comes to general policy preferences. The p-value is too high for the effect to be declared statistically relevant. The second two-way ANOVA conducted focuses solely on the interaction effect between both variables, but this time it focuses solely on socially oriented preferences:

Table 13: Interaction Effect Test for Socially Oriented Policy Preferences.

Sum Sq DF F-Value Pr(>F)

Intercept 1209.1 1 259.7134 < 2.2e-16

Treatment Variable 18.5 1 3.9712 0.04645

Coalition/Opposition Variable 87.1 1 18.7056 1.619e-05

Interaction effect 5.6 1 1.2067 0.27214

Residuals 7509.6 1613

Once again, it seems like there is no plausible interaction effect at all. The p-value is still far too high to properly claim the existence of any statistical relevance.

Lastly, a third two-way ANOVA focuses on the interaction effect once more, with regards to economically oriented preferences:

Once more, the p-value is too high to declare any statistically relevant interaction between the two variables.

(34)

31

Decisively, the data seems to be very firm on the implausibility of an interaction effect between both variables, even regardless of what type of policies were considered. In practical terms, this means that being either in the Coalition or the Opposition of a municipal council does not have any influence on the way the Treatment variable impacts decision-makers’ policy preferences.

4.5: Considering other variables

To better understand the results for the Treatment and the Coalition/Opposition variables, a third phase focuses on other variables also present in the dataset. For this, one common model with all 7 variables (excluding thus the IP Address variable). By having all variables in one common model, it becomes possible to directly compare the size of effects. Three models were made, each corresponding to one of the values of the Preferences variable.

The first model regards the general preferences from decision-makers:

Table 15: All-Inclusive Regression Model for General Policy Preferences.

Estimate Std. Error t-value Pr(>|t|) Intercept -2.167704 0.518527 -4.181 3.07e-05 Country Variable: Netherlands 1.239890 0.471570 2.629 0.00864 Spain 0.672733 0.486447 1.383 0.16688 United Kingdom 0.634103 0.485066 1.307 0.19133 Switzerland 0.093570 0.503153 0.186 0.85250

Table 14: Interaction Effect Test for Economically Oriented Policy Preferences.

Sum Sq DF F-Value Pr(>F) Intercept 80.3 1 17.5367 2.71e-05 Treatment Variable 44.9 1 9.8079 0.001769 Coalition/Opposition Variable 0.1 1 0.0280 0.867054 Interaction effect 1.9 1 0.4112 0.521437 Residuals 7355.8 1607

(35)

32

Language Variable (Switzerland only) Italian -0.918315 0.614588 -1.494 0.13533 German -0.233903 0.317878 -0.736 0.46195 Other NA NA NA NA Gender Variable: Male 0.015691 0.090727 0.173 0.86271 Other -0.461619 0.898184 -0.514 0.60736 Coalition/Opposition Variable (Opposition only) 0.214476 0.082358 2.610 0.00913

Political Wing Variable:

Left-wing 0.132110 0.126163 1.047 0.29520 Centre-Left-wing 0.190645 0.112070 1.701 0.08912 Centre-Right-wing -0.017593 0.121718 -0.145 0.88509 Right-wing -0.342695 0.161266 -2.125 0.03375 Age Variable 0.002280 0.003329 0.685 0.49356 Treatment Variable (Treatment only) -0.102347 0.079700 -1.284 0.19929

Note: The maximum number of observations (1958) was used. For reference groups, refer to Table 16. The Language Variable results only apply to the 207 Swiss participants; other participants fall under the category “Other” for this Variable.

Reference groups were used, as a result, these groups are not represented on the table above. When it comes to Italian, English, and Spanish, no values were given because of linear dependencies with the Country variable (all participants with “Other” as a language are from countries other than Switzerland).

It is further confirmed that, in the case of general policy preferences, the Treatment variable has no effect (as its p-value is higher than 0.05); furthermore it would seem like Age, Language, and

Table 16: Reference Group per Variable Variable Reference Group

Country France

Language French

Gender Female

Coalition/Opposition Coalition Political Wing Centre-wing Treatment Variable Control

All Bark and no Bite? Coalition/Opposition decision-making in European Municipal Councils