1. Introduction
This paper explores the relationship between blatant electoral fraud and malapportionment by focusing on when political leaders utilize each of these electioneering strategies to win parliamentary elections.Footnote 1 Political leaders often resort to illiberal electoral strategies. For instance, some may use electoral cheating methods by tampering with the ballot box, strengthening illegal vote-buying, and packing election management bodies (Kelley, Reference Kelley2012). Others may exercise violence during elections to repress opposition figures and supporters (Hafner-Burton et al., Reference Hafner-Burton, Hyde and Jablonski2014). These techniques of blatant electoral manipulation contribute to boosting the votes of governing parties to levels that the parties could not otherwise attain (Simpser, Reference Simpser2013). Blatant electoral fraud are frequently observed in authoritarian regimes and emerging democracies. Even in some old democracies, the levels of electoral integrity are dissimilar in an era of democratic backsliding (Norris, Reference Norris2017).
Besides blatant fraud, political leaders and their parties also engage in manipulating electoral rules, i.e., by deliberate institutional manipulation. Generally, the specific electoral systems chosen by governments significantly impact upon its electoral performance through seats–votes disproportionately (Boix, Reference Boix1999; Gandhi and Heller, Reference Gandhi, Heller, Herron, Pekkanen and Shugart2018). Gerrymandering, namely redistricting in favor of ruling parties, may be also employed as a further form of institutional manipulation (Wong, Reference Wong2019). Similarly, malapportionment, ‘the discrepancy between the shares of legislative seats and the shares of population held by geographical units’ (Samuels and Snyder, Reference Samuels and Snyder2001: 652), helps incumbents win elections by increasing the value of a vote primarily within the ruling party's strongholds (Ong et al., Reference Ong, Kasuya and Mori2017). For instance, Malaysia's United Malay National Organization increased the overrepresentation of rural districts where the party's major support base resides, which contributed to its election victory (Washida, Reference Washida2018). In sub-Saharan Africa, unequal values of a vote between rural and urban areas helped incumbents to win elections (Boone and Wahman, Reference Boone and Wahman2015: 341–344). Even in advanced democracies such as Japan, the ruling Liberal Democratic Party's (LDP) main support base had been overrepresented until the early 1990s, which helped the party to stay in power for long (Ong et al., Reference Ong, Kasuya and Mori2017: 119).
We know for a fact that political leaders use these methods of electoral manipulation to win elections. We know less, however, about the conditions under which incumbents prefer to choose one specific electoral strategy. In particular, little research has been conducted thus far on the relationship between blatant electoral manipulation and a specific form of institutional manipulation, or malapportionment in legislative elections. This is unfortunate for policymakers and political scientists, given that predicting when and what electioneering strategy leaders are more likely to use is meaningful for improving electoral integrity.
This paper suggests that both malapportionment and blatant electoral fraud have pros and cons as electioneering strategies. Although both help the incumbent win parliamentary elections, blatant electoral manipulation is risky because it often undermines the leader's legitimacy and can subsequently lead to popular protests. By contrast, malapportionment is less risky, because it is an indirect, less overt form of electoral manipulation. However, it is also difficult for political leaders to flexibly manipulate the value of a vote in accordance with their electoral demands, because even pro-regime redistricting and reapportionment often face strong opposition from ruling legislators, who often have different, irreconcilable preferences over the design of electoral districts and legislative apportionment.
With these factors in mind, I test observable implications on a cross-national dataset of the value of a vote and overt election fraud. Replicating Hafner-Burton et al.'s (Reference Hafner-Burton, Hyde and Jablonski2014) statistical model, I use the starter to test my hypotheses. Cross-national statistical analysis covering 98 countries (from 1993 to 2012) finds that, although the levels of malapportionment are not associated with election cheating and violence individually, political leaders become less likely to engage in the simultaneous use of these two electoral manipulation techniques as the high levels of malapportionment are historically endowed. The overall results suggest that political leaders may become less inclined to utilize every possible fraudulent measure when a high level of malapportionment has already guaranteed a significant seat premium to the ruling party. Furthermore, the inability to find clear evidence of correlations between malapportionment and each electioneering strategy, along with sluggish within-country changes in the extents of malapportionment, suggests that leaders may not be able to manipulate the value of a vote for their political needs. These results are robust to a battery of robustness checks, including alternative measures of election cheating and electoral violence, different estimators, outlier analysis, and different model specifications.
This paper's contribution to the literature is twofold. First, it sheds new light on a political consequence of malapportionment, arguing that malapportionment could affect the leader's incentives to employ overt electoral fraud even after controlling for other confounding factors affecting blatant electoral fraud, such as levels of democracy. Second, this paper explores commonalities and differences between blatant electoral fraud and malapportionment. Thus, I situate the study of malapportionment into an already broad literature of electoral manipulation.
2. The incumbent's election toolkit
Multi-party elections are when incumbents may lose their office. Even in elections where leadership turnover is not directly staked (e.g., parliamentary elections in presidential systems and semi-competitive elections in electoral autocracies), failure to win (big) weakens leaders' power bases and even provokes instability through protests and coups (Simpser, Reference Simpser2013; Higashijima, Reference Higashijima, Norris, Frank and Martinez i Coma2015; Wig and Rod, Reference Wig and Rod2016; Hafner-Burton et al., Reference Hafner-Burton, Hyde and Jablonski2018). In this respect, election periods are a critical moment for political leaders.
Researchers have studied various methods in which political leaders win elections. Blatant electoral manipulation is a typical electioneering strategy most frequently observed in authoritarian regimes but also in more than a few democracies. Electoral cheating is a series of non-repressive, yet undemocratic measures that bias election results (Kelley, Reference Kelley2012; Simpser, Reference Simpser2013: 34). During election campaigning, a government may undermine the level playing field by placing strong restrictions on the opposition's freedom to conduct campaigns, institute a pro-government media bias, and use other non-violent intimidation techniques (Frye et al., Reference Frye, Reuter and Szakonyi2018). On election day, the incumbent may pack the central election committee in order to tamper with ballot boxes, and/or induce their supporters and party brokers to participate in illegal actions (Stokes, Reference Stokes2005; Sjoberg, Reference Sjoberg2016).
Political leaders may also resort to pre-electoral violence, a very direct form of blatant electoral manipulation (Dunning, Reference Dunning2011; Bekoe, Reference Bekoe2012; Hafner-Burton et al., Reference Hafner-Burton, Hyde and Jablonski2014, Reference Hafner-Burton, Hyde and Jablonski2018). Election violence prevents opposition figures from carrying out effective election campaigns and often coerces them into boycotting elections. Election violence also plays an important role of depressing voter turnout among opposition supporters. Together, these techniques contribute to the incumbents' election victory (Hafner-Burton et al., Reference Hafner-Burton, Hyde and Jablonski2018). When institutional constraints on the ruler are weak and a close result is expected, the incumbent is more likely to resort to using election violence (Hafner-Burton et al., Reference Hafner-Burton, Hyde and Jablonski2014).
Governments also engage in institutional manipulation, namely, the manipulation of electoral institutions and electoral districts (Birch, Reference Birch2011). Besides electoral system change and gerrymandering, it is also well-known that malapportionment induces pro-incumbent, conservative, and rural biases (Samuels and Snyder, Reference Samuels and Snyder2001; Snyder and Samuels, Reference Snyder, Samuels and Gibson2004; Boone, Reference Boone2014; Boone and Wahman, Reference Boone and Wahman2015; Daxecker, Reference Daxecker2019). Numerous studies have demonstrated that malapportionment significantly helps ruling parties boost their parliamentary seats in both democracies and autocracies. According to Ong et al. (Reference Ong, Kasuya and Mori2017), although some levels of malapportionment exist across virtually all types of political regimes, competitive autocracy and new democracies are most likely to be malapportioned for ruling parties. Using constituency-level data from eight sub-Saharan African countries, Boone and Wahman (Reference Boone and Wahman2015) offer empirical evidence demonstrating that high levels of malapportionment lead to overrepresentation of ruling parties. Similarly, by analyzing cross-country panel data from Latin America, Bruhn et al. (Reference Bruhn, Gallego and Onorato2010) show that malapportionment helps ruling groups preserve their power by insulating their political support from electoral competition.
It should be noted that leaders do not use these electioneering strategies at random. With regard to the relationship between election cheating and election violence, Simpser (Reference Simpser2013) claims that election cheating and election violence go hand in hand, because coercing regime supporters to undertake these techniques enables rulers to signal their strength to potential opponents. Similarly, Hafner-Burton et al.'s (Reference Hafner-Burton, Hyde and Jablonski2014) cross-national analysis suggests that electoral cheating is positively correlated with pre-electoral violence.
Despite the fact that varying combinations of electioneering strategies have been of interest, we know little about how malapportionment is related to blatant electoral manipulation. Previous research on the value of a vote has primarily focused on the economic consequences of malapportionment (Horiuchi and Saito, Reference Horiuchi and Saito2003) or determinants of malapportionment (Samuels and Snyder, Reference Samuels and Snyder2001; Horiuchi, Reference Horiuchi2004; Kamahara and Kasuya, Reference Kamahara and Kasuya2014; Ong et al., Reference Ong, Kasuya and Mori2017). One important exception is Daxecker (Reference Daxecker2019), who uses the constituency-level data of six parliamentary elections in India and finds that highly malapportioned districts tend to experience less electoral violence. The current study builds upon her research to extend the theoretical focus to election cheating. I explore the relationship between three major electioneering strategies (cheating, violence, and malapportionment). Empirically, this paper utilizes cross-national data from 98 countries to test hypotheses about the relationship between these electioneering strategies.
3. Blatant electoral fraud and the value of a vote
3.1 Effect of malapportionment on individual use of cheating and violence
In regimes with multi-party elections, political leaders must successfully pass through two points of the election cycle (Hafner-Burton et al., Reference Hafner-Burton, Hyde and Jablonski2018). First, leaders need to win the election itself. The aforementioned electioneering strategies increase the likelihood of winning elections. Second, the post-election period can be also uncertain. This phase may include protest movements that force the incumbents to resign, hold new elections, or make large concessions to the opposition.
Employing blatant electoral fraud, political leaders can increase the likelihood of winning elections.Footnote 2 However, blatant electoral manipulation is also a risky strategy: it is overtly illegal and undemocratic, which undermines the incumbent's political legitimacy. Both electoral violence and election cheating damage popular perceptions about the fairness of elections and thus can invoke popular protests which can often be violent and destructive (Ong, Reference Ong2018). Much research has shown that blatant electoral manipulation backfires on political leaders. Norris (Reference Norris2014) demonstrates that electoral malpractice undermines people's confidence in governments and legal compliance. Excessive electoral cheating and electoral violence are often followed by popular protests (on electoral cheating, see Tucker Reference Tucker2007 and Bunce and Wolchik Reference Bunce and Wolchik2010; regarding electoral violence, see Hafner-Burton et al., Reference Hafner-Burton, Hyde and Jablonski2014, Reference Hafner-Burton, Hyde and Jablonski2018).Footnote 3
Compared to election violence and cheating, malapportionment is less costly and risky (Birch, Reference Birch2011). First, political leaders need not delegate brokers and supporters to manipulate elections in their regions through malapportionment. Political leaders within parliaments can revise their election laws to implement electoral redistricting and legislative reapportionment. Second, compared to electoral cheating and electoral violence, malapportionment is an indirect, and mostly, ‘invisible’ form of electoral manipulation. Redistricting and reapportionment can be implemented before election campaigning, enabling political leaders to distract citizens' and international organizations' attention away from the manipulation. Even if rulers manipulate electoral boundaries and district magnitudes during electoral periods, malapportionment is not blatant in the sense that it attempts to bias election results not through relentlessly thwarting opposition's election campaigns (as well as lowering their vote shares by fraud) but through reapportionment and redistricting. Along this line, Ong (Reference Ong2018: 162) asserts that malapportionment is a highly obscure electoral manipulation technique that requires a large amount of pre-existing knowledge to comprehend it and thus is less salient to voters. Therefore, high levels of malapportionment are less risky and less likely to be followed by protest movements. Indeed, my cross-national analysis shows that the levels of malapportionment are not correlated with the likelihood of post-electoral protests in a statistically significant way (Appendix Table C-1).
Importantly, high levels of malapportionment enable incumbents to win elections as ‘cleanly’ as possible without resorting to outright election cheating and electoral violence. Given the high costs of blatant electoral manipulation, political leaders may be more likely to refrain from using blatant electoral manipulation by increasing the level of malapportionment prior to elections. Such strategic manipulation of malapportionment levels leads to reducing the necessity of blatant electoral fraud techniques such as election cheating and election violence. Therefore, I derive the first hypotheses, which are about the effects of malapportionment on the individual use of cheating and violence:
Hypothesis 1-a: Increasing malapportionment is likely to reduce the probability of electoral cheating.
Hypothesis 1-b: Increasing malapportionment is likely to reduce the probability of electoral violence.
3.2 Effect of malapportionment on simultaneous use of cheating and violence
The discussion thus far assumed that political leaders can manipulate the levels of malapportionment flexibly enough to satisfy their political needs. However, reality suggests that different regimes are ‘historically endowed’ with the various levels of malapportionment, which reduces leeway for leaders to manipulate the value of a vote at their disposal. For instance, by investigating malapportionment in Latin America, Bruhn et al. (Reference Bruhn, Gallego and Onorato2010) document that malapportionment was so historically path-dependent in the region that it was not greatly adjusted even after democratic transitions. Ostwald and Courtin (Reference Ostwald and Courtin2020) also find that in Myanmar the usage of colonial-era administrative boundaries to delineate electoral constituencies contributed to a very high level of malapportionment that over-represented non-Bamar and rural votes. Similarly, Boone and Wahman (Reference Boone and Wahman2015) demonstrate that the levels of malapportionment in sub-Saharan African countries exhibited striking stability even in the era of democratic transitions in the region (1990–2010). Indeed, my cross-national comparison suggests that the levels of malapportionment are less liable to change over time, compared to extents of election violence and electoral cheating (Appendix Table C2).
Broadly, there are two factors which make the levels of malapportionment unlikely to be manipulated by politicians over time. First, malapportionment may have an important drawback as electioneering strategies – an intractable coordination problem among ruling politicians. When political leaders manipulate electoral institutions, they have to pass the revised electoral law through their respective legislative bodies. However, ruling politicians who may hold veto power and thus engage in decision-making processes in adopting the new electoral laws are often likely to have diverse, often mutually conflicting interests over the designs of redistricting and reapportionment. As Tsebelis (Reference Tsebelis1990) argues, extant electoral rules may shape the interests of legislators within each party, which makes it difficult to change electoral systems, even if an alternative electoral system is rational for parties as a whole. Indeed, Boone and Wahman (Reference Boone and Wahman2015: 340–341) report that in sub-Saharan Africa legislative proposals for reapportioning were rejected by ruling legislators in many countries including Kenya, Zambia, Malawi, and Ghana.
Second, a high level of malapportionment is also endowed via demographic changes across electoral districts. For instance, when a large number of people move from rural to urban areas (and thus the value of a vote becomes higher in the former), the ruling party with its main political support base in rural areas has no incentive to deal with the generated gap in the value of a vote, because the population change enhances the value of a vote in the party's strongholds. For example, thanks to rapid population inflows into urban areas which gradually widened the value of a vote between rural and urban areas, the ruling, rural-based Liberal Democratic Party increasingly enjoyed seat premiums and for a long time was hesitant to reform the malapportioned legislature (Sugawara, Reference Sugawara and Takashi2009). Without adjusting for the gap between shares of seats and population size across different electoral jurisdictions, political leaders are able to bias election results in their favor.
When malapportionment is extensive due to these reasons, i.e., historical path dependency and demographic changes, political leaders may no longer rely heavily on every possible measure of blatant electoral manipulation, as Simpser (Reference Simpser2013) has suggested.Footnote 4 Put differently, leaders may no longer need to use both violence and cheating as complements to win parliamentary elections because the baked-in ‘structural feature’ of malapportionment provides seat premiums. Rather, when malapportionment is high, political leaders may become less dependent on blatant electoral manipulation in consideration of its risk. Under the condition of extensive malapportionment, given the relative flexibility of blatant fraud techniques, leaders may start thinking about striking a balance of election victory and political risk by doing without either violence or cheating. Namely, malapportionment may affect the extent of blatant electoral fraud as a substitute that lowers the simultaneous employment of violence and cheating in elections. Therefore, the second hypothesis is formalized as follows:
Hypothesis 2: Increasing malapportionment is likely to reduce the probability of simultaneous electoral cheating and violence.
4. Empirics
4.1 Sample and malapportionment data
To test the hypotheses proposed in the previous section, I conduct cross-national statistical analyses. My analysis includes 98 countries covering the period from 1993 to 2012. The unit of analysis is country-election year, and includes 248 legislative elections. As some levels of malapportionment and blatant electoral manipulation exist in both democracies and autocracies and thus arbitrarily selecting samples according to the levels of political development risks the danger of selection bias. However, limiting the sample to developing countries and thus excluding industrial democracies (i.e., old members of the Organization for Economic Co-operation and Development (OECD)) does not alter the main results (Appendix B4).
The main independent variable in this paper is the degree of malapportionment. I use an extensive cross-sectional time series dataset of malapportionment originally constructed by Kamahara and Kasuya (Reference Kamahara and Kasuya2014), complemented with Ong et al.'s (Reference Ong, Kasuya and Mori2017) cross-sectional data of malapportionment.Footnote 5 Malapportionment is defined as ‘the discrepancy between the shares of legislative seats and the shares of population held by geographical units’ (Samuels and Snyder, Reference Samuels and Snyder2001: 652). It is measured as an index that employs a measure of the Loosemore–Hanby index (Kamahara and Kasuya, Reference Kamahara and Kasuya2014: 4):
$$MAL_{t\comma j} = \displaystyle{1 \over 2}\sum {\vert S_{i\comma t\comma j}-V_{i\comma t\comma j}\vert } $$where i denotes a particular district, t a certain election-year, j a given country, s denotes the proportion of allocated seats in district i to all districts, and v the share of population or electorates in district i to the entire population or electorates. When MAL is zero, the distribution of seats does not favor any electoral districts in the country. As this value increases, the legislature consists of representatives selected from more malapportioned electoral districts.
The mean of malapportionment is 0.06, meaning that on average countries have 6 percentage point difference in votes vs seats obtained across electoral districts.Footnote 6 Countries adopting a proportional representation system with a nationwide district do not have any malapportioned districts (e.g., Netherlands, Slovakia, Israel, and Kazakhstan), whereas countries such as Chile (0.15), Spain (0.1), Gambia (0.27), Ghana (0.19), Togo (0.22), Tanzania (0.27), and Mongolia (0.14) maintain high levels of malapportionment.
4.2 Statistical models
4.2.1 Dependent variables
We have three dependent variables in the analysis: (i) only election violence, (ii) only electoral cheating, and (iii) both election violence and electoral cheating. The dependent variables (i) and (ii) are operationalized to test Hypothesis 1, whereas (iii) is to test Hypothesis 2.
Whether an election experienced only election violence is measured by using Hyde and Marinov's (Reference Hyde and Marinov2011) National Elections in Democracies and Autocracies (NELDA; Version 4). The NELDA dataset contains information on elections for national offices for all sovereign countries with populations greater than 500,000. The dataset is constructed by using various sources including newswire reports, newspaper archives, academic research, archives from specific countries, and reports from intergovernmental organizations. Following Hafner-Burton et al. (Reference Hafner-Burton, Hyde and Jablonski2014: 165), we code the occurrence of election violence if the government engaged in election-specific violence against civilians (coded from Nelda33) or harassed opposition members (Nelda15). Then, to measure the individual use of election violence, the dependent variable takes the value of 1 if election violence happens but election cheating does not occur and 0 otherwise. Of 248 country-election years, a total of 7.6% (19) of country-election years in the sample experienced only election violence (Appendix A2). In democracies, 6.3% (13) of country-election years had election violence, whereas election violence is observed in 14.6% (6) of autocratic country-election years.Footnote 7
The second dependent variable is only election cheating, measured by using the NELDA dataset. This variable comes from Nelda11 indicating whether there were ‘significant concerns that the elections will not be free and fair’. This variable captures ‘domestic or international concern’ about the quality of the election, including whether ‘elections were widely perceived to lack the basic criteria for competitive elections, such as more than one political party’ (Hyde and Marinov, Reference Hyde and Marinov2011). Code 1 indicates there are serious concerns that the elections will be fraudulent. Similar to the measure of only election violence, the individual use of election cheating is coded 1 when election cheating happened but election violence does not occur. Of 248 country-election years, 5.6% (14) of country-election years in the sample experienced only election cheating (Appendix A). In democracies, 4.3% (9) of country-election years had the above mean level of election cheating, whereas the same level of electoral cheating is recorded in 12.2% (5) of autocratic country-election years. Although Hafner-Burton et al. (Reference Hafner-Burton, Hyde and Jablonski2014) state that this variable measures ‘another prominent tactic of electoral manipulation’ distinct from election violence, one may wonder if the variable at least partially includes election violence. Therefore, I use an alternative measure of election cheating, based upon the Varieties of Democracy (V-Dem) dataset. The results are robust across those measures (Appendix B3).
Finally, the third dependent variable, the dual use of electoral violence and election cheating, is operationalized by coding whether a country-election year registers both election violence and electoral cheating simultaneously (on the NELDA dataset). In the sample, 10.0% of 248 parliamentary elections (25 country-election years) experienced both electoral violence and election cheating (e.g., Kenya, Moldova, Pakistan, Sri Lanka, and Ukraine), whereas 41.4% of autocracies (17 country-election years) and 3.8% of democracies (8 country-election years) experienced both violence and cheating, respectively.
4.2.2 Model specification
To test Hypotheses 1-a and 1-b, i.e., that malapportionment reduces the probabilities of election violence and electoral cheating individually, I regress each of these variables on malapportionment and control for other covariates introduced by Hafner-Burton et al. (Reference Hafner-Burton, Hyde and Jablonski2014). Hypotheses 1-a and 1-b predict the coefficients of the malapportionment variable to be negative and statistically significant for the individual use of election violence and electoral cheating. Hypothesis 2 predicts that malapportionment reduces the probability of the simultaneous use of electoral fraud and violence. Note that here I do not test the causal effect of malapportionment on these techniques of blatant electoral fraud. This regression analysis is to test a correlation – i.e., whether leaders' simultaneous use of violence and cheating becomes less likely as the levels of malapportionment become higher. If we find malapportionment is negatively correlated with the likelihood of the dual use of election violence (in Hypothesis 2) and cheating but not with each individually (in Hypothesis 1), the overall results suggest that political leaders become less dependent on a wide range of blatant fraud when the levels of malapportionment are already high.
The model specification of this study follows Hafner-Burton et al. (Reference Hafner-Burton, Hyde and Jablonski2014). They provide a well-founded baseline model explaining cross-national variations in electoral violence in particular and blatant electoral manipulation in general. According to them, political leaders are inclined to use election violence when their election victories are uncertain, but the use of election violence is constrained when checks and balances toward the executive exist. This conditional effect is operationalized by introducing an interaction term between NELDA's measure of victory uncertaintyFootnote 8 and Polity IV's strength of executive constraints.Footnote 9 Repressive regimes are more likely to use election violence and other fraud techniques. Without controlling for political repressiveness of countries, models run the risk of just estimating which regimes are more repressive in general. Therefore, my models include a physical integrity index (1-year lagged, 3-years moving average) to measure the levels of government repression in non-electoral periods.Footnote 10 To ensure that the estimation results are not spurious with the level of democracy, I include measures of political competitivenessFootnote 11 and executive recruitmentFootnote 12 from the Polity IV project. Similarly to the physical integrity index, these measures of democracy are measured as 1-year lagged, 3-years moving averages to control for non-election-specific components. Logged populationFootnote 13 and logged GDP per capitaFootnote 14 are also included as controls because wealth and population size influence the use of violence and cheating (e.g., Lehoucq and Molina, Reference Lehoucq and Molina2002; Fukumoto and Horiuchi, Reference Fukumoto and Horiuchi2011). Because political leaders may be more likely to use election fraud based upon their length of tenure or their experience, I include leader's tenure length and leader's age from Archigos version 4.1. Because civil conflict is associated with human rights abuses, I introduce a binary measure of civil war from the Peace Research Institute of Oslo (PRIO) dataset. Finally, I include the number of demonstrations, anti-government strikes, and riots to consider the possibility that civic mobilization encourages incumbents to use blatant electoral fraud (1-year lagged, from Banks' (2016) Cross-National Time Series Data Archive).
4.2.3 Estimator
This study has three binary dependent variables (only election violence, only electoral cheating, and both violence and cheating) and therefore employs logistic regressions. Because time-series of this dataset are far shorter (2.6 election-years on average) than its cross-section (98 countries) and some independent variables (e.g., the malapportionment variable) are highly sluggish over time,Footnote 15 employing fixed effects (FE) estimators yields higher variance than random-effects (RE) estimators (Clark and Linzer, Reference Clark and Linzer2015).Footnote 16 Therefore, I estimate RE logistic regressions with robust standard errors to consider country-level unobserved heterogeneity and heteroskedasticity. In a robustness check, I alternatively employ a multi-nominal probit regression in which I regard the four groups (no violence and no cheating, only violence, only cheating, and both violence and cheating) as distinct categories within a single categorical dependent variable, to find that the results are identical to those of the RE-logit estimator (Appendix Table B1). Each RE-logit model is formalized as follows:
$$Pr\lpar {OnlyViolence_{it} = 1} \rpar = f\lpar {\beta_1MAL_{it} + \phi X_{it} + \gamma_i + \psi_t + \epsilon_{it}} \rpar $$
$$Pr\lpar {OnlyCheating_{it} = 1} \rpar = f\lpar {\beta_1MAL_{it} + \phi X_{it} + \gamma_i + \psi_t + \epsilon_{it}} \rpar $$
$$Pr\lpar {Violence_{it} = 1\cap Cheating_{it} = 1} \rpar = f\lpar {\beta_1MAL_{it} + \phi X_{it} + \gamma_i + \psi_t + \epsilon_{it}} \rpar $$where X it is a vector of control variables, γ i is random-effects controlling for unobserved country-level heterogeneity, and ψ t is year-fixed effects. Equations (1) and (2) are for hypotheses 1-a and 1-b, respectively, while (3) tests hypothesis 2.
4.2.4 Estimation results
Table 1 reports the estimation results. In Model 1 (where the dependent variable is whether only election violence occurred), the coefficient of the malapportionment variable is positive and not statistically significant. Similarly, in Model 2 (where the dependent variable is whether only electoral cheating occurred), the malapportionment variable's coefficient is positive and again not statistically significant. These results suggest that malapportionment does not reduce each of election violence and electoral cheating individually, contrary to the expectation proposed in Hypothesis 1.
Blatant electoral fraud and malapportionment
Note: Robust standard errors in parentheses. ***<0.01, **<0.05, *<0.1.
Model 3 then tests Hypothesis 2, which asserts that political leaders refrain from using both election violence and electoral cheating if high levels of malapportionment are endowed. As prima facie evidence, Figure 1 presents a Jitter and Violin plot on the relationship between malapportionment and blatant electoral fraud. As the figure shows, malapportionment levels tend to be lower when countries experience both election violence and cheating. Countries with low levels of malapportionment, such as Tajikistan, Thailand, Nepal, Ukraine, Belarus, Armenia, Kazakhstan, Russia, and Moldova experienced both electoral violence and election cheating (lower part of the ‘both violence and cheating category’).Footnote 17 The regression analysis (Model 3) shows that the coefficient of the malapportionment variable is negative and statistically significant at the 5% level, indicating that the higher malapportionment is, the less likely political leaders are to rely on both repressive and non-repressive measures of blatant electoral manipulation at the same time. Figure 2 illustrates changes in the predicted probabilities of the simultaneous use of election violence and electoral cheating according to the levels of malapportionment. When the value of a vote is the same across all electoral districts, electoral cheating is accompanied by electoral violence by a high probability, approximately 17%. As the degree of malapportionment becomes larger, however, elections become less likely to experience election violence and electoral cheating simultaneously. For instance, when the malapportionment score increases to 0.15, the likelihood that the dual use of violence and cheating happens becomes less than 6% at the 5% significance level. This provides supporting evidence for Hypothesis 2.
Jitter–Violin plot.
Note: The big dots represent mean values of malapportionment in each category, whereas the small dots stand for observations. The outer shape represents kernel density plots showing distribution of malapportionment levels.
Predicted probabilities of dual use of election violence and cheating.
Note: The straight line represents the coefficient of the malapportionment variable, whereas the dashed lines stand for the 95% confidence intervals. The thin gray bars represent the distribution of malapportionment levels.
4.3 Robustness checks
I conduct a battery of sensitivity analyses to show that the main results are robust in the face of additional methodological issues such as (1) a different estimation method, (2) additional controls, (3) different measures of election violence and electoral cheating, (4) an alternative sample focusing on developing countries, (5) consideration of time dependence, and (6) potential outliers.
The outcome of interest, i.e., patterns of blatant electoral fraud, can be modeled differently. Here, I employ a multi-nominal probit regression with four different outcomes of blatant electoral fraud – no existence of both electoral violence and cheating, only electoral violence, only election cheating, and the dual use of election violence and cheating.Footnote 18 The estimation results show that high levels of malapportionment make the simultaneous use of violence and cheating less likely from any other patterns of electoral manipulation. When malapportionment becomes higher, leaders are more likely to shift their blatant electioneering strategies from dependence on both violence and cheating, to cheating alone, violence alone, or neither (Appendix B1). In contrast, malapportionment is again not associated with the individual use of election violence and electoral cheating. These findings again suggest that political leaders rely less on blatant electioneering as malapportionment increases, but malapportionment may not completely eliminate all forms of electioneering.
Previous research suggests that malapportionment is correlated with majoritarian electoral systems and fiscal transfers or expenditures (e.g., Horiuchi and Saito Reference Horiuchi and Saito2003; Snyder and Samuels, Reference Snyder, Samuels and Gibson2004). Furthermore, other research also finds that both majoritarian electoral systems and financial resources are related to the specific extent of blatant electoral manipulation (Birch, Reference Birch2007; Higashijima, Reference Higashijima, Norris, Frank and Martinez i Coma2015). To test whether the main results are robust to these possible confounders, I introduce electoral system typesFootnote 19 and fiscal expenditureFootnote 20 as controls. The results are not sensitive to the inclusion of those variables (Appendix B2).
To examine whether the results are sensitive to different measures of electoral cheating and election violence, I use the V-Dem dataset as an alternative source to dichotomously categorize both election cheating and election violence. To do this, following the V-Dem project, I first measure the extent of electoral violence and election cheating by applying an Item Response Theory technique to create their latent variables. The latent electoral cheating measure consists of (1) Voter registry (v2elrgstry), (2) vote-buying (v2elvotbuy), and (3) other voting irregularities (v2elirreg), whereas the latent election violence measure is made by (1) government intimidation (v2elintim) and other election violence (v2elpeace). Then, to make dummy variables of election violence and electoral cheating, I use the means of both measures as the thresholds above which elections are seen as those with electoral violence and cheating. Using these alternative measures of election violence and cheating does not affect the main results (Appendix B3).
The main analysis includes both developed and developing countries because of the reasons stated above, yet violence and cheating are more likely to occur in authoritarian regimes in particular and developing countries in general. Therefore, I limit my sample to non-OECD countries (62 countries) and run the same models to find that this alternative sampling does not affect the main conclusions (Appendix B4). Also, to control for time dependence of binary dependent variable models, I control for the time lapse since the last election violence and its time polynomials (Carter and Signorino, Reference Carter and Signorino2010). The inclusion of these variables does not alter the original results (Appendix B5). Finally, to take into account possible outliers that may derive from some extreme values of the malapportionment variable, I conduct jackknife analyses by excluding country and year one by one. The results remain robust.Footnote 21
4.4 Additional analyses on different features of blatant fraud and malapportionment
The interpretation of the estimation results so far is based upon the assumption that blatant electoral fraud and malapportionment have different features as electioneering strategies. To test this assumption, I conduct additional analyses on differences in the key features of blatant electoral fraud and malapportionment. The first major difference between the two electioneering strategies is that, whereas blatant electoral fraud often provokes post-election protest movements, high levels of malapportionment do not, given its almost entirely indirect, invisible features of biasing election results. To test this additional observable implication, I estimate the effect of malapportionment on the likelihood of popular protests. Again, the estimation method (RE logit) and model specifications are based upon Hafner-Burton et al.'s (Reference Hafner-Burton, Hyde and Jablonski2014) analysis of post-election protests, except that I add the malapportionment variable to their models. The dependent variable, popular protests, is binary and measured by using Nelda29, which indicates whether there were ‘riots or protests after elections’ that were ‘at least somewhat related to the outcome or handling of the elections’. The results are consistent with my theoretical expectation. The coefficient of malapportionment is highly uncertain (P = 0.402), suggesting that the levels of malapportionment are unrelated to post-election protests (Appendix C1).
The second important difference between blatant electoral fraud and malapportionment is that malapportionment is a more inflexible electioneering strategy than blatant electoral fraud. One observable implication of this feature is that the levels of malapportionment should change more slowly over time than both election violence and electoral cheating do. Comparing within-country standard deviations (S.D.) and the within-country mean of each variable, those of malapportionment are 0.006 (S.D.) and 0.06 (mean) and thus its correlation of variation (CV) is only 0.1. By contrast, the S.D. and the mean of election violence are 0.2 and 0.177 with a CV of 1.12, whereas those of electoral cheating are 0.185 and 0.157 with a CV of 1.18. These differences suggest that variances of blatant electoral fraud are much larger than that of malapportionment (Appendix C2). Malapportionment is an even more inflexible electioneering strategy for political leaders than election violence and electoral cheating.
5. Conclusions
This paper has investigated how malapportionment is related to blatant electoral manipulation. Political leaders choose their election strategies while considering the benefits and costs of each strategy. Although blatant electoral manipulation helps rulers to obtain election victories, such coercive measures may also backfire on rulers given the potential to undermine political legitimacy and thereby spark popular protests. Malapportionment, a large gap in the value of a vote across electoral districts, enables political leaders to maintain legislative dominance by allocating more seats to their strongholds without using blatant electoral manipulation. Therefore, when malapportionment is high, political leaders should refrain from using electoral cheating and election violence. The cross-national statistical analysis has shown that, although malapportionment itself seems not to contribute to the reductions of election cheating and electoral violence each, political leaders become less inclined to use both cheating and violence simultaneously when high levels of malapportionment are endowed. This suggests that political leaders may not be able to flexibly manipulate electoral districts as a complete substitute for ‘good, old-fashioned’ cheating and violence. Instead, they may carefully strike a balance between election results and blatant fraud by becoming less dependent on both electoral cheating and election violence when malapportionment has already made the electoral battlefield favorable to the incumbents.
This paper's findings may suggest several policy implications. Contemporary election monitoring tends to draw their attention toward overt electoral fraud and concludes that elections have no serious problems as long as the elections do not suffer blatant election fraud. However, the core implication of this paper is that, even if political leaders do not resort to cheating and violence, when malapportionment is high, such relatively ‘free and fair’ elections may derive not from leaders' respect for transparent elections, but rather from the fact that they still hold big electoral advantages brought about by high levels of malapportionment. Policymakers and international organizations might need to consider the trade-off between overt electoral fraud and malapportionment, so as to be able to design election monitoring schemes accordingly.
Second, that being said, this research also suggests that international organizations may find it difficult to encourage political leaders to adjust huge gaps in the value of a vote because the current level of malapportionment may be a manifestation of a political equilibrium of electoral interests among legislators. That is, among several politicians there can be a reluctant attitude whereby the potentially corrupting structures of misaligned and democratically unrepresentative regimes become ossified to the point where outright and blatant corruption transpires and even, in the worst cases, become normalized. Even if international civil society succeeds in correcting for malapportionment, what follows may be the eruption of blatant electoral manipulation by the leader who desires to hold onto power, followed by post-electoral popular protests by citizens, both of which may ultimately destabilize the country. In this sense, international support for improving the discrepancy in the value of a vote needs to be taken into account and more specifically, that consequences occur after their assistance should be considered.
Because this paper's empirical analysis is cross-national, it is generally difficult to test further observable implications at sub-national levels. For example, it may be that a high level of malapportionment is associated with the absence of election cheating and electoral violence in the electoral districts where gerrymandering does not involve serious coordination problems among ruling politicians or where population changes lead to favorable gaps in the value of a vote vis-á-vis ruling parties. To test such additional predictions, dis-aggregated, election-district level data on overt election fraud and malapportionment will be needed. Furthermore, the empirical analysis of this paper explores correlations between blatant electoral fraud and malapportionment. To test causal relationships between these electioneering strategies, future research may apply methods of causal inference.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1468109921000037 and https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/WG2HG7
Acknowledgements
Earlier versions of this paper were presented at the Shanghai University of Finance and Economics, the 2016 World Congress of International Political Science Association, and the 2016 annual meeting of Japanese Political Science Association. I appreciate feedback and comments from the participants of these conferences. Austin Mitchell and Yuki Yanai provided helpful comments at the final stage of this project. The three anonymous reviewers also made useful and constructive comments which contributed to further improving the paper. Finally, Yuta Kamahara and Yuko Kasuya generously shared with me their malapportionment data. This research is funded by a JSPS grant-in-aid (26285032).
