Contesting the Status Quo

Transcription

1 STVM01 Tutor: Jan Teorell Department of Political Science Contesting the Status Quo A comparative study of democratic electoral outcomes in electoral authoritarian regimes Michael Wahman

2 Abstract There has recently been an increased interest among academic scholars on the importance of elections as a promoter for democratic change in semi-democratic contexts. This thesis positions itself somewhat on the side of this debate asking; when do elections in electoral authoritarian regimes result in democratic consequences? Recognising that elections can be as well a tool for democratic empowerment as authoritarian legitimisation, the study tries to avoid an obviously democratic bias and seeks the causes for as well democratic improvements as erosion in connection with elections. Using a quantitative method with a global population of 283 elections in the period , the sample of the study is significantly larger than previous investigations on this subject. The main findings of the analysis is that while most of the theories laid out in the theoretical literature do not stand the test of a large scale quantitative test, a high number of demonstrations preceding the election and a larger historical experience of previous elections, is important for the democratic electoral outcome. Key words: elections, democracy, demonstrations, authoritarianism, comparative, multiple regression analysis

3 Table of contents 1 Introduction Theory The foggy zone Electoral authoritarianism is not a stepping stone Elections in electoral authoritarian regimes The cumulative effect of elections in authoritarian regimes Hypothesis based on previous research The opposition The incumbent regime Institutional factors Economic and social factors External factors Method Defining the population The dependent variables Democratic electoral outcome (DEO) Electoral upturns and downturns Inspired but different Quantitative Analysis of electoral outcome Democratic Electoral outcomes The opposition The incumbent regime Institutional factors Economic and social factors External factors Electoral upturns and downturns Demonstrations The number of previous elections Conclusion References Appendix... 31

4 8.1 Elections in the population Codebook Dependent variables Independent variables Additional models... 42

5 1 Introduction During the spring of 2008, when this thesis is written, the world s attention has been focused on Zimbabwe. After a history of several highly contested elections within this country, the current election has turned into an absolute mockery of the democratic process. Following the March elections, the results are still unclear. Even though the opposition party is confident that their leader, Mr. Morgan Tsvangirai, collected enough votes for a clear victory against the incumbent president Robert Mugabe, the official authorities still at this moment insists that the election result was indecisive. Zimbabwe is only one of many recent and historical examples of semiauthoritarian regimes where the election has been an actual setback for the democratic process or where the cynic could argue that the democratic institutions has proven to do little for the country but to provide a democratic legitimising façade for the authoritarian regime (Carothers 2002: 9 ff). Other elections around the globe have however proven to be useful tools for democratic breakthroughs where empowered electorates and opposition forces have managed to use the electoral institution to enforce real political change. The recognition of the diversity of democratic electoral outcomes (DEO) in these electoral authoritarian regimes is the starting point of this thesis. Within the field of democratisation some essential work has been written on the causal effects and mechanisms that allow for different democratic outcomes in connection with elections. Most of this work has however been of a more theoretical kind (e.g. Schedler 2002), case studies or small number comparative studies (e.g. Bunce and Wolchik 2006) or quantitative studies with a relatively small sample (e.g. Howard and Roessler 2006). The aim with this study is to use a large number analysis, with a significantly higher number of observations than used in previous research and answer the following question; Q1: What factors affect democratic electoral outcomes, within electoral authoritarian regimes? 1

6 2 Theory The aim with this theory section is to relate this thesis to the theoretical discussions on electoral authoritarianism generally and on elections within this context specifically. 2.1 The foggy zone Regime classifications have been the object of many scholars attention during recent years. Long gone is the bipolar typology, only recognising authoritarian and democratic regimes (Munck 2006:28 ff). Not only has the study of democratisation become more sophisticated, but the empirical reality has also changed in a way which makes it difficult to neglect the shades of gray located between the two endpoints of democracy and authoritarianism. Especially after the end of the cold-war, the world has seen a rapid growth of regimes that fits in to neither of the two above mentioned categories. In diverse countries like Russia, Bangladesh and Tunisia, regimes have been established that combines democratic rules with authoritarian governance (Letvitsky and Way 2002:51). This middle position between democracy and complete autocracy is the sphere which Andreas Schedler (2002:37 ff) labels the foggy zone. This sphere has by different scholars been given numerous names, such as semi-democratic, semiauthoritarian, illiberal democracies, hybrid regimes or partly free regimes. Schedler does however remark that the distance between the categories in this three-point scale is too wide. He suggests a typology consisting of four categories; Liberal democracies (e.g. the United States), are countries that combine civil liberties and rule of law with free and fair elections. Electoral democracies (e.g. Brazil), are countries that hold free and fair elections, but do not guarantee an adequate level of civil liberties to be placed in the liberal democratic category. Electoral authoritarian regimes (e.g. Venezuela) are countries that hold recurrent, at least partially competitive but flawed elections. In contrast to the electoral democratic category the elections within the electoral authoritarian category do not meet the minimum democratic requisites to be labelled as democratic. The elections are however not pointless. Even though the game is not played on equal terms the elections are real events of political competition. Closed authoritarian regimes (e.g. China) are regimes that either do not have any elections or that have elections that are merely very transparent façades. Elections within the closed authoritarian category could for instance be of an 2

7 intra-party kind (e.g. Cuba) or events where candidates run for relatively powerless positions (e.g. Iran). This thesis will use the classifications outlined above and concentrate on the electoral authoritarian category. I believe in accordance with Diamond (2002:33 f) that it is important to acknowledge the different shades of gray in this foggy zone. It is possible to imagine that institutional mechanisms operate differently in these different regime types, and by recognising the diversities it is easier to reach conclusions with a higher degree of validity. A possible point of critique is that one could be even more specific in regime type classifications. Following Letvitsky and Way (2002:52), one could for instance make a distinction between competetive authoritarian (regimes with a higher degree of competetiveness) and hegemonic authoritarian regimes (regimes with one very dominant party), within the electoral authoritarian category. It is possible that this would have increased the validity further, but I have chosen not to make this distinction of purely practical reasons Electoral authoritarianism is not a stepping stone It is very important to acknowledge that electoral authoritarianism is neither a stepping stone between authoritarianism and democracy, nor is it a temporal phase for a state sliding back to authoritarianism. Even though it is true that most closed authoritarian regimes path to democracy goes through the intermediate democratic phases, it is essential to keep in mind that electoral authoritarianism is a regime type that can become long lived and consolidated (Hadenius and Teorell 2007:153). Hence, these regimes have a logic of their own and are not merely authoritarianism light or incomplete democracies. In the same way as one could assume that there are certain elements of society that benefit from the status quo in authoritarian or democratic systems, the same could be said about the electoral authoritarian system. My ambition is to not fall into this common pitfall of studying authoritarian electoral regimes with an obvious democratic bias, which one does when looking at the system as a transitional phase (Letvitsky and Way 2002:51 ff). This theoretical argument implicates that even though there are possibilities for a country to develop their democratic institutions within the realm of the electoral authoritarian system and develop into a liberal or electoral democracy, this development is in no way predestined. 1 Practically, competitive authoritarian regimes are often identified in quantitative studies by setting a threshold on the number of seats in parliament or the number of votes for the majority party. Since I do not have access to electoral data for the whole time period, and due to the problem of comparability when one uses legislature seats as operationalisation (The number of seats for the majority party is to a high degree depending on institutional rules, where majority election systems have a significantly higher possibility to reach the hegemonic threshold than proportional systems has), I have chosen not to divide the electoral authoritarian group into one hegemonic and one competitive group. 3

8 2.2 Elections in electoral authoritarian regimes It is undoubtedly true that elections are a prerequisite for democratic progress and it is unthinkable to imagine a democracy without them. Despite this recognition of the essentiality of elections, a central theme of this thesis is that elections do neither have to have a liberalising effect, nor should the election necessarily be perceived as a sign of weakness from the authoritarian regime. An essential question that should be asked is why authoritarian regimes bother to hold elections. The answer to this question is of course not easy to provide and are probably rather contextual. In many contexts the authoritarian regimes are practically forced to hold elections due to as well internal as external pressure, but in other cases the elections themselves can be perceived as an important tool to remain in power. By arranging these recurrent elections the authoritarian regimes get the legitimacy they need to avoid the most forceful internal and external reactions on their management, and if they succeed to keep control over the electoral process the chance of losing power is minimised (Schedler 2002:36 f). In an investigation by Gandhi and Przeworski (2007) the authors come to the perhaps puzzling conclusion that authoritarian rulers who allow multipartism and incorporate rebel parties into the parliament generally enjoy longer tenures than those who reject rival parties representation in the parliament. Finally, one possible point of critique against looking closer at the causes of different democratic electoral outcomes is that the elections themselves are not an isolated phenomena but a reflection of the state of a society where information about the relative strengths of the regime and the opposition becomes evident (Brownlee 2007: 11). This is obviously a relevant objection. It is important to remark that the contest over democratisation is not only fought during election periods. Specific structural developments are ongoing in a society independently from the electoral arena that might increase or decrease the probability for democratisation. It is however this thesis theoretical standpoint that the struggle between different political forces within society becomes especially apparent during elections, and that some specific institutions and structures could affect the probability of certain democratic electoral outcomes. These factors can make an impact on the democratic electoral outcome in several ways, for instance by increasing/decreasing the probability of electoral misconduct or by affecting the probability of oppositional success in connection with elections The cumulative effect of elections in authoritarian regimes A closely related issue to the main theme of this thesis is whether elections as such have a generally positive effect on democratisation on an aggregated level. This question has been the subject of some interest from students of international democratisation and has been investigated as well theoretically as empirically. In the empirical work the general opinion tends to be that elections seem to have a positive effect on democratisation, even though different authors do not agree on 4

9 what kind of elections that have a liberalising effect and whether it is an instant or a cumulative long-term effect (e.g. Bratton and van de Walle 1997 and Lindberg 2004). In a recent contribution to this debate, Teorell and Hadenius (2008) make a summary of the state of the research on democracy and elections. They acknowledge two central theoretical streams within this field. The first is labelled mouse nibbling. The metaphor suggests that democratisation by elections are achieved through small instant democratic gains during election periods. The other theoretical thought is named the pressure chamber. This theory regards the liberalising effect of elections as the consequence of a constantly higher pressure on the authoritarian system that builds up in every election. In their empirical investigation the authors conclude that elections have a positive effect on the degree of democracy both in the short run in connection with the election, and in the long run (i.e. a high number of elections in a country s history tend to give higher levels of democracy). On a more theoretical note, other authors have however had a more sceptical view of the relationship between elections in authoritarian contexts and democatisation. Carothers (2002: 9ff) does for instance argue that the democratic gains made from elections could be expected to be very limited, since they have not been able to create any participation beyond voting. So what have we learnt from this previous research that can be useful in answering the central question of this thesis? I believe that the previous research has shown that elections might perfectly well contribute to a countries democratisation. The institution of elections is not meaningless even though there are good reasons to be careful of ones expectations when it comes to the effect of elections in specific cases. Elections can be an opportunity for opposition empowerment and civil uprising (Lindberg 2004), but it might as well be a tool for autocratic survival (Gandhi and Przeworski 2007). Through acknowledging the complexity of elections in authoritarian settings, it is easy to motivate why it is essential to more closely investigate under which circumstances elections have a liberalising effect and when they do not manage to change the status quo, or even has a negative effect on the level of democracy. This recognition of the duality of elections makes the theoretical standpoint of this study mostly independent from the mouse nibbling and pressure chamber theories (Teorell and Hadenius 2008). It does however share the assumption put forward by the mouse nibbling theory that the small democratic gains matter. Even though the nibbling of the mouse does not necessarily lead to a state of democratic consolidation, it can provide at least a chance for further democratic improvements (McFaul 2002: 113 f). A troubling aspect of the metaphor of the nibbling mouse is however that it suggests that democratisation is just a matter of time in electoral authoritarian states. Every time the mouse gets the chance to take yet another bite of the cheese, it is one chew closer to getting entirely eaten. This metaphor is, according to me, close to the assumption of regarding electoral authoritarianism as a stepping stone to democracy. 5

10 3 Hypothesis based on previous research The purpose with this section is to briefly account for the independent variables that will be used in this thesis. One of the perks with the quantitative research design is that it allows the researcher to test a high number of different theories. It is however important to only test variables that have solid theoretical support since the introduction of theoretically hard motivated variables may soak up some of the cross-country variance (Lieberman 2005:438 f). Hence, my strategy has been to identify the most prominent explanations to different democratic electoral outcomes in electoral authoritarian regimes within the theoretical literature, and then test those in a large number analysis (LNA). Some important variables have been excluded from the LNA, due to problems of making an adequate quantative operationalisation or lack of data. Below, I have divided the most important independent variables into five different thematic sections. In each section the theory and hypothesis about the specific variables will be presented. The operationalisation will also be discussed in varying detail, depending on its complexity The opposition Many scholars have stressed the importance of a strong and cohesive opposition movement in order to achieve democratic progress in electoral authoritarian contexts (Howard and Roessler 2006:371 f). Bratton and van de Walle (1997) have made an important contribution to the understanding of opposition dynamics in electoral authoritarian regimes, in their comparative study of regime transitions in Africa. They conclude that democratisation in this context was more likely to be a consequence of pressure from below, than the result of a bargaining game between regime hard-liners and soft-liners in accordance with O Donnell and Schmitter s (1986) classical argument. According to the authors two specific opposition qualities and actions become crucial for a successful democratic outcome; Political protests seem to have been an important trigger for political change in Africa. Protest movements often started of as a consequence of dissatisfaction 2 For more detailed information on operationlisation, number of observations and sources for the statistics please consult the codebook in the appendix. 6

11 with economic and social conditions, but turned into more comprehensive movements that connected their discontent with more concretely defined goals of political and democratic change. These movements became platforms for mass mobilisation and important articulators of political change (Bratton and van de Walle 1997: 97 ff). Protests can occur in different forms. The form which most frequently is affiliated with democratic improvements is demonstrations, but it is also conceivable that more drastic forms of protest such as riots and strikes could have an effect on the DEO of an election (ibid. 159 ff). Opposition cohesion is the other important oppositional factor that affects democratisation according to Bratton and van de Walle (ibid. 198 ff). A prominent problem within many electoral authoritarian regimes is that the social movements aiming for change often remains divided into different local or ethnical groups with small resources and very limited possibilities to challenge the incumbent regime in the arranged election. It is inevitably hard to capture this aspect using quantitative data. Howard and Roessler (2006: 371 f) solve this problem by using a qualitative assessment and codes opposition coercion on the ground of information gathered in secondary sources. Due to the considerably larger reach of this study, in terms of the time and spatial dimensions, this has however not been possible. Hence, a rough proxy will be utilised, namely an index accounting for the fractionalisation among opposition parties within the parliament. There are obviously several flaws with this operationalisation. For instance, the results from the previous election might not reflect the current conditions, and even though there is a high fractionalisation among opposition parties the cooperation between them might be substantial. Despite these objections, I believe that this operationalisation should give at least a partial picture of the cohesion within the opposition. This operationalisation is also similar to the one uses by van de Walle (2006: 81), when the author accounts for oppositional cohesion by regarding the share of the votes held by the main oppositional contender in relation to the presidential election. H1: A higher number of demonstrations, strikes and riots should increase the probability of a positive DEO 3. H2: A higher degree of oppositional cohesion should increase the probability of a positive DEO. 3 For the definition of DEO, see

12 3.2 The incumbent regime The DEO is of course not only a result of the actions by the opposition. It is important to stress that the incumbent regime should not be regarded as a passive actor. The authoritarian government can and does try to manipulate the election in different ways and if this is done right the probability of authoritarian survival increases (Case 2006: 95 ff). In a case study of Malaysia, Thomas Pepinsky (2007) finds that the authoritarian government uses as well sticks as carrots to ensure their continuating rule. He finds that elections is a time where the authoritarian regime in some cases chose to strengthen governmental control and more frequently use repressive measures, but also a time when the governments choose to increase public spending. The sticks are hard to investigate within the realm of this study 4, but it would however be interesting to see if the governmental tactic to increase public spendings is successful in preserving the authoritarian system. Besides the tactics of the incumbent regime it is also interesting to regard the characteristics of it. In previous research it has been hypothesised that strong charismatic leaders with longer tenures are less likely to lose control of an election and open up for political change (Bratton and van de Walle 1997: 61 ff) Therefore, the length of the tenure for the authoritarian chief executive, measured as years of consecutive rule, will be included in the model. H3: A higher increase of public spending increase the probability of a negative DEO. H4: A longer tenure for the chief executive should increase the probability of a negative DEO. 3.3 Institutional factors Institutional design is one of the more disputed aspects within the democratisation literature. It has been said to have an especially important role in the mitigation of conflicts in ethnically divided societies (Belmont et al: 2002), but it has also been related to other consequences in the literature concerned with elections in authoritarian electoral regimes. Sarah Birch (2007) argues the electoral system is important, and that an election using a proportional (PR) electoral system is more likely to be free and 4 If the government has engaged in a higher degree of repressive measures it is probable that this will affect the civil liberties scores, and since this measure is a part of the dependent variable it would obviously be tautological to include it into the regression model. 8

13 fair, while single member district (SMD) elections exhibits a higher degree of electoral misconduct, since a fewer number of votes have to be changed in order to swing election results around. The SMD system does also require a highly coordinated opposition in order to contest the incumbent regime. According to Pripstein Posuney (2002), these aspects of the institutional design are often well known among authoritarian regimes, and at least in the Arab world the adoption of SMD systems has been a conscious strategy to secure continuous rule for the incumbent regime. It is however possible to imagine that the SMD system would encourage the opposition to a higher degree of cooperation, since this could be their only chance to win the contested seats (van de Walle 2006: 88) By the same token, Presidential systems have been connected with a lesser chance to achieve electoral liberalising outcomes. As in the SMD system, presidential systems are more prone to winner-takes-all politics. The fall for an authoritarian leader is much harder in a presidential system than it is in a parliamentary system, where the leader at least can continue to enjoy a seat in the parliament if his party is defeated (Howard and Roessler 2006: 373 f) 5. This study will only account for presidential and parliamentary systems and not the mixed system since this category is very wide in definition and very small in number of observations within the population. As already suggested in the theory chapter it has been argued that a longer previous history of elections could increase the likelihood of a positive DEO 6, since this gives the electorate experience of the voting institution (Lindberg 2004). Therefore following Teorell and Hadenius (2008:10), there will be a variable included to the regressions accounting for the cumulative number of elections held since 1919 in the country in question 7. H5: A proportional electoral system should increase the probability of a positive DEO, while a majoritarian system should increase the probability of a negative DEO. H6: Parliamentarianism should increase the probability of a positive DEO, while presidentialism should increase the probability of a negative DEO. H7: A higher number of cumulative historic elections should increase the probability of a positive DEO. 5 An additional difference between presidential and parliamentary systems is obviously the fact that presidential states arrange presidential elections. If this variable would prove to be significant, it would therefore be necessary to investigate if this significance is a consequence of a difference in DEOs in presidential and parliamentary elections. 6 For a lengthier discussion on this subject see chapter This measure is of course somewhat arbitrary. It is hard to set boundaries for when to start measuring the history of elections and one could also argue that a more recent history of elections should have a larger impact on the DEO than an older one. Teorell and Hadenius (2008) elaborate with different weights on elections, but use the cumulative stock of elections from 1919 and onwards as their main variable. Since this variable is more straight forward than the other more complicated measures used by the authors, I have chosen to focus on this more simple operationalisation. 9

14 3.4 Economic and social factors The modernisation theory is without doubt the most prominent and well spread theory of democratisation. Even though economic and social development are perceived to have beneficial consequences for democratisation in the long-term, short-term economic growth could increase the support for the authoritarian regime and decrease the popular demands for political change. By contrast, the support for the current political system gets severely damaged in instances of economic crisis (Geddes 1999). This variable will simply be accounted for by regarding the annual growth in real GDP. A relatively high initial socio-economic development could however increase the chances of political liberalisation through elections. Bunce and Wolchik (2006) argue that a large part of the explanation for why a number of postcommunist countries in Europe have experienced electoral revolutions recently, while very few African countries have had the same electoral development, is that the economic structures are much more beneficial in the former context. The relatively high socio-economic development together with a higher level of education increases the possibilities for citizens to engage in civil society and to critically evaluate information. The socio-economic development will be measured by Bank s (1997) modernisation index that accounts for several important socio-economic aspects. H8: A higher degree of economic growth should increase the probability of a negative DEO. H9: A higher degree of a socio-economic development should increase the probability of a positive DEO. 3.5 External factors Traditionally, external factors have been perceived as secondary to the internal conditions in the democratisation literature (Bratton and van de Walle 1997: 181 ff). In an increasingly globalised world, some scholars have however questioned if not the pressure from the economically superior west could have an effect on other countries domestic political systems. Aid and trade treaties are often amended with clauses of human rights and democratic governance. If the country does not meet limited democratic standards or if they arrange obviously flawed elections this might have a negative economic effect. It is accordingly possible to imagine that countries that are more dependent on trade with or aid from the west are less prone to electoral misconduct (Wolchik and Bunce 2006). In order to 10

15 account for these aspects, trade (trade as share of GDP) and aid (aid as share of GDP) dependence will be included in the regression models. The recent development in post-communist Europe has questioned the independence of different political systems. Some scholars have regarded it unlikely that dramatic events, like for instance the rose-revolution in Georgia and the orange revolution in Ukraine, are not related. They have argued that democratic progress within a country s own cultural sphere can inspire the opposition movement and alienate remaining authoritarian regimes in relation to their liberalised neighbours. This process has been labelled democratic diffusion (Ambrosio 2007:234). It is important to remark that to operationalisise diffusion is not an easy task. As described above, diffusion is here related to actual triggering events in other countries that could effect domestic democratisation. Sometimes diffusion is rather related to the tendency for countries to adapt to the general level of democracy within their cultural sphere (Brinks and Coppedge 2005). This notion of diffusion could be perceived as a more long-term effect, while the instant inspiration of a democratic breakthrough in a culturally closely related country could be perceived as more important in the actual electoral situation. There is no flawless way to capture this phenomenon. Questions like; How should one set boundaries for a cultural sphere? How long time after the democratic progresses in country X could one expect that this event could inspire a transition in country Y? Is it only dramatic democratic progresses that has this effect or is it possible that even smaller democratic progresses could inspire movement in other neighbouring countries? I have however decided to design a variable that calculates the total number of elections within five years of the current year that has had a liberalising effect within a certain region, 8 divided with the total number of elections held within this region during the same time-period. H8: A high dependency on aid and trade should increase the probability of a positive DEO. H9: A high number of recent positive DEOs in the countries region should increase the probability of a positive DEO 8 The world is divided into ten regions; Eastern Europe and Central Asia, Latin America (including Cuba and Haiti),Middle East and North Africa, Sub-Saharan Africa, Western Europe and North America (including Australia and New Zeeland), East Asia, South-East Asia,South Asia,Pacific Islands (excluding Australia and New Zeeland) and the Caribbean (including Belize, Guyana and Suriname). 11

16 4 Method As already mentioned in the introduction chapter this study will use a large number analysis (LNA) method to answer its guiding question. The raison d être for this study is to contribute to the knowledge of the mechanisms of elections in electoral authoritarian regimes, by using a sample significantly larger than the ones presented in earlier research on this area. Since this is the aim and due to the fact that the question is situated relatively high on the by Lundquist (1993: 64) presented ladder of abstraction, it is natural to utilise a quantitative method to answer the question. This method will enable a thorough and widely generalisable test of some of the most prominent theoretical arguments laid out in more qualitative and/or theoretical orientated work performed within this field. Of course the quantitative method is not without its limitations. Compared to small number analysis it is harder to evaluate causal relations between the variables tested and it is also harder to go into detail about more specific mechanisms on how X affects Y (King et al 1994:9 ff). I do acknowledge this weakness of the applied methodology. It is however important to add that a detailed description of causal mechanisms is not the principal aim of this study, since this should be better recognised in a study with a different research design. A conceivable design that could integrate the strengths of the qualitative as well as the quantitative method is a nested method, which combines these two scientific approaches (Lieberman 2005: 435). However, because of the limited realm of this thesis I have decided to focus on the quantitative study, even though some smaller illustrations on actual cases will be presented at some instances of the analysis. The cases that have been selected as examples are what Seawright and Gerring (2008:303 f) would refer to as influential cases 9. The purpose with these is not to prove that there exists a relationship between X and Y (this is already done by the LNA). They should rather be perceived as examples on possible causal relations and can also investigate if some of the most influential cases are consistent with the underlying theory. If this is not the case there are reasons to expect that the found connection between X and Y is spurious. 9 Influential cases are those where X have stronger correlation with Y than normally within the population. They are not representative to the entire population but are useful in order to test and illustrate causal patterns (Seawright and Gerring 2008:303 f). 12

17 4.1 Defining the population This section will account for the selection criteria guiding the design of the population. Questions regarding the basic level of analysis, the time dimension, and which countries that will be regarded as electoral authoritarian will be addressed below. The population or the units of analysis will not be countries or country years but elections. This implicates that only years when an election is held are accounted for in the regressions, and that one country can have had many elections during the time period. This approach is the same as the one used by Howard and Roessler (2006: 368). Since the aim of this study is to keep the number of observations on a level as high as possible, I see no reason for restricting the population in time as long as there is accessible data. Since my dependent variable is dependent on Freedom House scores (see section 4.2), the study can not go back further in time than , and since observations on many of my independent variables start to decrease in the most recent years, the last observed year will be Only electoral authoritarian elections will be included to the population. The operationalisation of this category will be equivalent with Hadenius and Teorell s (2007: 147 f) category limited multiparty systems (LMS). LMS codes all countries that in a given year have a government that has been elected in a popular and recurrent election, where at least one real oppositional candidate was allowed to participate, but where the democracy score 11 is less than 7.5 on a 10-point (combined Freedom House and Polity) scale as a LMS state, or in my case as electoral authoritarian. I believe that this category is well coherent with the definition of electoral authoritarianism used by Schedler (2002:37 ff). 12 The aim is to measure the effect of the election, therefore it is the state of the country prior to the election that is important, and not what state the election transferred the country into 13. Since only countries that was categorised as electoral authoritarian in the year prior to the election will be regarded, founding elections 14 will be excluded. This is as Howard and Roessler (2006: 369) points out desirable since the outcomes of these are seldom an effect of what happens in connection with the election, but rather a consequence of power struggles that occurred in the process of setting up the electoral institutions. 10 The first coded year for Freedom House ratings is 1972, but since the design of the dependent variable is depending on a comparison from the prior year, 1973 is the first possible year to include into the analysis. 11 Fore more information see the section below. 12 The threshold of 7.5 on the combined Freedom House and Polity scale is obviously not God given, but it is chosen by Hadenius and Teorell (2007) to reflect the thresholds of some of the most central works within comparative democratisation studies. 13 More explicitly, if country X experience a positive DEO in year t, which transfers it into the democratic category. This election will still be counted since country X was electoral authoritarian at year t Do not confuse this term with the first free and fair election. Many countries hold remarkably flawed elections for a long time before true competitive elections take place. As long as the prior elections include a small degree of competition they are counted as electoral authoritarian. 13

18 In this study both parliamentary and presidential elections will be accounted fore. There are accountable data for both sources and there is no theoretical reason why one of the election types should have a larger impact on a countries level of democracy than the other. Plebiscites or municipal elections are however not a unit of analysis. The definition of the population outlined above leaves the study with 283 elections ranging from This is a significantly larger number than previous studies on this matter The dependent variables This chapter will account for the dependent variables and motivate the design of these. The main dependent variable will be democratic electoral outcome (DEO) and hence most effort will be put into explaining this variable. Finally a smaller explanation of the additional upturn and downturn test will be presented Democratic electoral outcome (DEO) There are good reasons for devoting some effort to the design of the studies main dependent variable, since there are some essential decisions to motivate on this issue. The first question to address is of course what exactly one mean when using the term democratic outcome of an election. An election can be said to affect a country s level of democracy in multiple ways; prior to the election (for example when an authoritarian leadership adapts there behaviour and tactics in advance of an election, to mitigate the challenge from the opposition) in connection (as a consequence of for instance oppositional victories or electoral misconduct) or after the election (through for instance the empowerment of the electorate, that has been given a chance to participate in an election) (Lindberg 2004: 158). Consequently, the practical question is whether one should measure the DEO as a process occurring prior, in connection with or after the election or possibly during all these stages. There is no obvious solution to this question. Howard and Roessler (2006:369) choose to measure the effect of an election as the difference in the level of democracy between the year of the election and the year prior to the election, without elaborating too much on the reasons for this decision. Lindberg (2006: 141) on the other hand measures the effect of an election as changes in the level of democracy occurring both in the year preceding the election and in the 15 The number of observations could be compared to Howard and Roessler (2006), who only accounts for 50 elections ranging from For a more thorough technical description of the dependent variables including sources, see the attached codebook in the appendix. 14

19 given election year. He argues that election related activities preceding the election such as campaigning is also a type of electoral effect. Even though the argument put forward by Lindberg is reasonable, I have chosen to use the same method as Howard and Roessler on this issue. The problem with applying a method that also accounts for the democratic development in the year prior to the election is that this measure increases the risk of measuring the consequence of other events, not related with the election itself. Elections are of course central events that will determine the democracy score applied to a country for a certain year, in the year preceding the election this is however not as certain 17. By the same token, the year subsequent to the election will not be accounted for. With this said it is important to remark that it is not only the few days during the election that is important in the measuring of the DEO. It is conceivable that the change in a countries level of democracy is not a direct consequence of the judgement of the democratic qualities of the election but on events related to it. One example is this studies most negative DEO, namely the Haitian election of At this instance the military removed the elected president, Jean-Bertrand Aristide, only nine months after the election. I do however not perceive this as a problem since the DEO should ideally not only be a consequence of the few days during the election, but also a reflection on other related events in the nearest history and future. The other important question is how to measure the level of democracy. There are two absolutely dominant indices of democracy, the Freedom House and the Polity measure, both having their particular strengths and weaknesses (Munck and Verkuilen: 2002). Following Teorell and Hadenius (2004), I will therefore use a combined inverted Polity and Freedom House civil liberty and political rights score 18. Other authors have decided to do this in a slightly different way. Howard and Roessler (2006:369) only measures the electoral outcome by accounting for change in Polity and Freedom House s political rights score 19, for no given reason. Lindberg (2006:141) on the other hand does only use the Freedom House civil liberty score. Since his ambition is to investigate the effect of elections on the level of democracy per se 20, he wants to assure that the change in the dependent variable is not an effect of the assessment of the democratic qualities of the election itself but a consequence of it. This problem of tautology is however not relevant when investigating what circumstances that explains different DEOs. In this study it is interesting to observe the change in political as well as civil rights. 17 It is also important to keep in mind that democracy scores are applied on an annual basis; hence it is possible for events occurring almost two years prior to the election to affect the democracy score for the year prior to the election. 18 Ranging from 1-10 where 1 equals the lowest level of democracy and 10 the highest. 19 Hence, leaving Freedom House s civil liberties score out of the variable. 20 Not on under which circumstances an election gets a certain effect on the level of democracy. 15

20 Below the design of the DEO variable is illustrated: dem τ - dem τ-1 As already mentioned in the introduction chapter this investigation will account for as well elections that increase as decrease the level of democracy, and as understood by the illustrated equation above the DEO variable is a continuous variable that can adopt as well positive as negative values Electoral upturns and downturns A supplementary upturn and downturn test will be done. This approach is inspired by Teorell and Hadenius (2007: 79). Even though I favour the use of a continuous dependent variables such as the DEO, it is useful to complement this main test with the upturn and downturn dummy variable test. By analysing the results of this test it is possible to assess if a certain variable gains significance because of its importance for the situations where the level of democracy increases as a result of an election, or because of its association with the situations resulting in lower levels of democracy or if the relation is significant in both directions 22. The upturn variable is coded by keeping the value from the DEO variable, for all cases with positive DEOs but coding all downturns as zero. The downturn variable is consequently constructed in the totally contrary way. 4.3 Inspired but different For the reader who is familiar with the work of Howard and Roessler (2006) it is obvious that the general theme and basic methodology of this thesis is similar to the study performed by the mentioned authors. The question about similarities and dissimilarities between this study and the one written by Howard and Roessler has been addressed at some instances earlier in this text. I do however find it 21 For a discussion on the advantages with this design see section This should not be confused with the assessment about if a variable is negative or positive for the level of democracy, this result is of course read out by the coefficient in the regressions. A good use of this method could for instance be if the variable for aid dependence gained statistical significance with a positive coefficient. By regarding this test, it is possible to conclude if it became significant due to the fact that is uncommon for countries with a large aid dependence to have elections that decrease the level of democracy, or if this result is due to the fact that many countries experiencing liberalising electoral outcomes have a high dependency on aid. 16

21 appropriate to in a more comprehensive manner provide information about how this study distinguishes itself from its source of inspiration. The most important differences are methodological (thus the placement within the method chapter), but some are also theoretical. Moreover, some of the methodological differences are motivated by different theoretical assumptions. Below the most important differences are outlined. Firstly, this study accounts for the whole strata of possible electoral outcomes, in contrast to Howard and Roessler, who use a dummy variable for liberalising electoral outcomes that only accounts for positive electoral outcomes. As noted in the theory chapter, elections in authoritarian contexts are not only instances for possible democratic breakthroughs, but might just as well be an opportunity for authoritarian consolidation of power (Schedler 2002: 36). Hence, it is equally important to account for the elections that lead to democratic deterioration as it is to account for those that improve the democratic level of a country. Secondly, this study uses a continuous dependent variable. This is, according to me, preferable due to multiple reasons. Foremost this measure makes it possible to account for different degrees of electoral outcome and hence give a more detailed description of the relationship between the dependent and independent variables. Just as one could assume that a higher degree of X makes it more probable to achieve Y, one could also imagine that a higher degree of X should lead to a higher degree of Y. The fact that more significant movements on the democracy scale has a larger impact on the results than smaller ones are also desirable. Smaller movements are more likely to be the effect of measurement errors by the sources (Polity and Freedom House) and are hence less reliable than larger movements (Collier and Adkock 1999:545 ff.). Finally, an additional technical advantage with the use of a continuous dependent variable is that this makes it possible to use a standard OLS-regression. This is not possible with a dichotomous dependent variable that calls for a probit regression, which delivers results considerable more complicated to interpret (Teorell and Westholm: 1999: 142 f). Thirdly, the population of this study is significantly larger than in Howard and Roessler s investigation. The population of their investigation does only consist of 50 elections, where of only 15 are counted as liberalising, compared to the sample of 283 elections in this study. Moreover, the time span of this study is longer and reaches over the period , in comparison with Howard and Roessler s The considerably larger N and longer time span of this study makes the results more reliable and increases the possibilities of generalisation. Finally, this study tests some additional hypotheses not used by Howard and Roessler. Variables not tested in the mentioned study include number of previous elections, diffusion, increase in public spending, modernisation and electoral systems. 17