During the 2013 congressional elections in the Philippines, Gloria Arroyo defeated a challenger from the ruling Liberal Party, Vivian Dabuand, and secured re-election to a second term in Congress despite the fact that she was on trial for corruption. Voters were not helped in their decision by the confusing choice that they faced. Arroyo, who was a founding member of the KAMPI party, was now standing for the independent minority party Lakas, and was also honorary chairwoman of the ruling Liberal Party. In the Philippines, even the largest political parties are characterized by frequent party switching and overlapping membership, and voters are confronted with short-term coalition building, as well as numerous party dissolutions and re-emergences from one election to the next.Footnote 1 This lack of party system institutionalization not only complicates for voters the task of attributing responsibility for corruption, it also undermines electoral co-ordination, and thus the ability of voters to employ electoral choice effectively to oust corrupt incumbents and hold their representatives accountable. Filipino voters do not approve of corruption. Sixty-four per cent believe governmental corruption to be a serious problem; seventy-five per cent believe that citizens have a role to play in combatting it.Footnote 2 Still, the Philippines are the sixteenth most corrupt democracy in our sample of eighty.
Realities like this are at variance with the conventional model of democratic accountability, which proposes that elections in principle make the self-interest of public officials and the preferences of the electorate compatible. Regardless of their real motives and intentions, the dependence of politicians on popular support is thought to force them to take electoral preferences into account.Footnote 3 This model assumes that citizens can use their vote effectively to sanction representatives who do not serve the interests of the electorate. However, empirical evidence drawn from a wide range of democracies demonstrates that freely contested regular elections often fail to secure governance in the interest of the people. Elections can help corrupt politicians to power and voters may fail to punish incumbents who engage in or permit malfeasance.Footnote 4
The impact of party systems on the effectiveness of elections has been largely overlooked in the comparative literature, which describes corruption as conditioned primarily by political institutions (an exception is Keefer).Footnote 5 This literature examines how free elections and full democratization,Footnote 6 presidentialism and federalism,Footnote 7 and also electoral systems affect accountability and clean government.Footnote 8
The reality, however, is that electoral accountability in democracies is not just structured by institutions, but also by the party system.Footnote 9 In this study, we show that the institutionalization of party systems plays an additional and neglected role in conditioning the effectiveness of elections as tools for voters to control politicians and to secure clean government. In many democracies around the world, voters face the challenge of controlling their politicians in the context of poorly institutionalized and unstable party systems characterized by short-lived party organizations that compete in changing electoral alliances attributing blame for poor performance and removing under-performing incumbents are formidable and have a powerful impact on the scope for corruption.
The effect of party system institutionalization, we argue, goes beyond underlying contextual factors such as electoral rules and social cleavages. To be sure, social diversity and the permissiveness of electoral rules define part of the structural context in which party system institutionalization may (or may not) arise.Footnote 10 But the institutionalization of party systems, given such contextual features, varies greatly,Footnote 11 and reflects contingent political choices by politicians, voters and elites more broadly. Whether parties persist or perish, for instance, is shaped by party leaders’ choices and inclinations such as their investment in a strong party organization and their responses to legitimacy problems (for instance, in the post-communist context).Footnote 12 Opportunities for new parties to rise are contingent, reflecting, for example, the strategies adopted by other competitors, endorsements by organized interests beyond the state (such as trade unions), and the financial support available.Footnote 13 Similarly, the stability of patterns of party competition is shaped by agency and discretion: party elites can recast competition in a variety of ways by betraying coalition partners, reversing policy in response to exogenous shocks, or seizing on valence issues.Footnote 14 Moreover, such choices are typically hotly contested precisely because they recast patterns of competition and because it is difficult to predict whether voters will depart from prior behaviour to reward or punish such strategies. In sum, party system institutionalization can be expected to have an independent impact on the effectiveness of elections that extends beyond the influence of social cleavages and electoral rules.
We test this argument using a research design that combines the strengths of a controlled comparative analysis of corruption in eighty democracies with an examination of the process by which weak party system institutionalization attenuates accountability in a case study of Panama. In order to causally identify the effect of party system institutionalization, we pay particular attention to sequencing and reverse causation. Our results reveal powerful effects of the party system on the scope for corruption. Where voters’ ability to attribute responsibility and to punish is limited by weakly institutionalized party systems, the otherwise large electoral penalties that voters impose for corruption (up to 13 per cent in our case study) can effectively be reduced to zero. As a result, the system of repeated elections no longer functions as intended and democracies give greater latitude to corruption.
These findings have ramifications for two central areas of comparative politics. To the comparative literature on the political determinants of corruption, we contribute by showing that the scope for corruption cannot be understood as structured by institutions alone, but depends critically on the existence of party systems that allow voters to attribute responsibility and to punish incumbents who fail to address corruption. To the literature on electoral accountability, we contribute the insight that party systems may play a role in shaping the electoral nexus that rivals that of institutions, on which the extant debate has overwhelmingly focused.Footnote 15
Conceptualizing Corruption
We conceptualize governmental corruption as the misuse of public office for personal or political gain, as well as the acquiescence in such misuse by bureaucrats.Footnote 16 This definition encompasses all forms of governmental corruption – grand and petty forms of theft, bribery and illegal rent-seeking by public officials. Defined in this way, governmental corruption is a public policy outcome, for which politicians are in principle accountable – they may invest resources in fighting it, or tolerate it and possibly even engage in it.Footnote 17
Surveys around the world indicate that a majority of citizens perceive governmental corruption as a deleterious outcome. This is true even in countries where corruption is endemic so that citizens regularly experience it, and benefit from it in individual instances, as is often the case in highly clientelist systems. For instance, in African democracies, as in other polities in which corruption and clientelism are widespread, decisive majorities of citizens typically regard corruption as morally reprehensible.Footnote 18 The 2006 Afrobarometer makes clear that over 60 per cent of all African respondents regard corruption by public officials as wrong and punishable.Footnote 19 That is not surprising. Even when citizens are frequently involved in corrupt exchanges, and benefit from them in specific instances, this does not imply that they normatively approve of governmental corruption, nor does it imply citizen tolerance for illegal rent seeking by politicians.
Surveys also make clear that citizens perceive governmental corruption as an outcome that results when their political agents – elected politicians – fail to address malfeasance. In Transparency International’s 2013 Global Corruption Barometer, which polled 114,000 people in 107 countries and territories around the world, respondents ranked political parties as the most corrupt institution overall, and in eighty-eight countries a majority of the respondents perceived their government’s efforts to fight corruption as ineffective.Footnote 20 For citizens collectively, then, governmental corruption is a principal–agent problem that arises when politicians deviate from the electorate’s interests. Voters (the principals) delegate government power to politicians (their agents) to attain certain goals – such as clean government – but self-interested politicians may abuse their office or tolerate malfeasance by bureaucrats.Footnote 21 Under what conditions, then, do politicians act more or less effectively as agents of the electorate?
Shirking, Party System Institutionalization And The Scope For Corruption
The literature on shirking examines precisely this question. This work is based on the observation that politicians may deviate from voter preferences because expending effort to pursue their principals’ concerns is costly.Footnote 22 For example, fighting corruption requires time and resources, for which other policy priorities compete. Likewise, desisting from abuse of office implies forgoing the personal and partisan benefits that politicians may otherwise derive from corruption. When politicians’ interests are not well aligned with their electorate, the result can be shirking, defined as ‘behaviour that differs from what would be observed given perfect monitoring and effective punishment’.Footnote 23
A central finding in the literature on shirking is that representatives deviate from voter preferences when the costs of doing so are low.Footnote 24 These costs vary with the effectiveness of elections so that transformations of the electoral connection can impact dramatically on the extent to which politicians act in the interest of voters. For example, retirement decisions and constitutional term limits, which sever the electoral connection, have been shown to increase shirking in the context of the United States: exiting members of Congress represent their constituents less faithfully and participate less;Footnote 25 retiring legislators spend substantially more money on foreign travel than their colleagues;Footnote 26 and state governors who cannot stand for re-election because of term limits are less disciplined in their tax and spending policies.Footnote 27 Although most of this research concentrates on the US, the intuition extends to other cases.Footnote 28
We build on the insight that the effectiveness of the electoral connection shapes the incentives and policy choices of politicians, and apply it to the comparative context and to a different domain – corruption control. In a cross-national context, we argue, the effectiveness of elections as a means to control and incentivize politicians is critically shaped by party system institutionalization.
Party system institutionalization has two central constituent features – the regularity of competition and the organizational stability of parties within the system.Footnote 29 While well institutionalized party systems feature stable parties and alliances as well as predominantly constant party affiliations of politicians, poorly institutionalized systems are characterized by the frequent rise of new parties and alliances, the demise of pre-existing ones, a low average party age, high electoral volatility, and frequent party switching by politicians.Footnote 30
The level of party system institutionalization varies among both young and old democracies. While Taiwan, for instance, quickly established a relatively stable party system,Footnote 31 Russia throughout the 1990s, had a fluid system in which the identity of key competitors changed from election to election.Footnote 32 Similarly, older democracies can suffer crises of representation that result in the extensive de-institutionalizations of their party systems, as in Italy in the 1990s, and Bolivia, Ecuador, Venezuela and Peru during the 1990s and early 2000s.Footnote 33
This variation in party system institutionalization has implications for the effectiveness of elections as instruments of accountability.Footnote 34 Electoral accountability depends on two conditions, (i) citizens must be able to attribute responsibility for policy and (ii) ‘they must have a fair opportunity to cast a meaningful vote for or against the policymakers’.Footnote 35 Party system institutionalization affects both dimensions of accountability, which has implications for shirking and the scope for corruption.
In well institutionalized systems, in which party organizations persist over time and alliances display some continuity, parties are able to develop clear reputations. This (i) allows voters to distinguish clearly between incumbents and the opposition, and (ii) enhances the informational value of party labels as well as the credibility of policy promises.Footnote 36 Voters are thereby given the means to attribute responsibility and to unseat corrupt incumbents by co-ordinating on a challenger with a credible commitment to reduce corruption, which raises the costs of shirking.
In contrast, poorly institutionalized party systems attenuate electoral accountability. When party and alliance changes affect the relationship between politicians and voters, voters are presented with an unstable set of choices that may (i) cut across the incumbent–opposition divide established by the previous election, and (ii) conflate the reputations of individual incumbents (who may have switched party), parties (which may have switched alliance partners), and alliances (that may have changed in composition).Footnote 37 As a result, responsibility is obscured: alliance and party changes distance parties (and individual politicians) from their record in office, detach their reputations from those of their former coalition partners (and party), and pool them with new coalition partners (or a new party). This makes it unlikely that voters will be able to attribute responsibility and effectively co-ordinate their efforts to punish incumbents who have failed to curb corruption. The same party system features also undermine the credibility of policy commitments to reduce corruption by pooling the non-credible commitments of tarnished incumbents with those of new coalition partners (or a new party).Footnote 38 Through each of these mechanisms low party system institutionalization impedes the electoral punishment of corruption and lowers the costs of shirking.Footnote 39
Anecdotally, too, we see that voters are better able to identify and sanction poorly performing incumbents in well institutionalized party systems. For instance, Welch and Hibbing show that in the context of the United States, a corruption charge costs incumbents nearly 10 per cent of the vote, and leads to electoral defeat in 25 per cent of the cases – far higher than the rate of defeat among incumbents who had not been charged.Footnote 40 Poorly institutionalized party systems, in contrast, diminish the ability of voters to attribute responsibility and to sanction effectively. For instance, the party formed by Brazilian President Fernando Collor de Mello, who resigned in 1992 following charges of corruption and a threat of impeachment, dissolved in the months following his resignation.Footnote 41 Similarly, party switching in Poland compromised the ability of voters to hold legislators accountable for poor performance.Footnote 42
Other things equal, then, poorly institutionalized party systems can be expected to increase the scope for shirking and corruption by diminishing the effectiveness of elections as mechanisms of electoral control and accountability.Footnote 43 This analysis yields two complementary hypotheses. First, at the aggregate cross-national level, we anticipate that:
Hypothesis 1: Greater party system institutionalization reduces governmental corruption.
Second, we expect that the constituent features of party system institutionalization condition the electoral accountability of individual incumbents, so that:
Hypothesis 2: Politicians associated with corruption, who are affected by party and alliance changes, are punished less effectively in elections.
To test these expectations, we combine a cross-national comparative analysis with a nested case study of Panama. The two parts of this research design have complementary strengths. The cross-national analysis contributes external validity by enabling us to test the effect of party system institutionalization on corruption in a large sample of democracies, taking account of confounding factors. The case study contributes causal process observations. Following Lieberman’s recommendations, we select a case that is nested in the cross-national analysis and is well predicted by it, so that the hypothesized causal processes can be expected to operate.Footnote 44 Within-case variation enables us to test a crucial part of the causal process – the effect of party and alliance changes on the extent to which incumbents are held accountable for alleged corruption.
Comparative Analysis
In testing our hypothesis at the cross-national comparative level, we focus on full democracies, which rank 6 or higher on the Polity Index of Democracy.Footnote 45 Our unit of analysis is the country and the data cover eighty democracies, observed over a seven-year period 2003–2009 (Appendix 1 details the countries included).
Dependent Variable
One of the most widely accepted measures of corruption is the control of corruption dimension of the World Bank Governance Indicators.Footnote 46 These data measure the essentially hidden phenomenon of corruption via a range of surveys of residents of a country, international and domestic business people, and risk analysts. The indicator is designed to record the extent to which public power is exercised for private gain, including petty and grand forms of corruption, as well as ‘capture’ of the state by elites and private interests. The measure embraces perceptions of all the forms of corruption that we wish to capture given that we conceive of governmental corruption as a public policy outcome. The World Bank indicator aggregates these surveys, treating them as measures of a common latent variable, which is estimated using an unobserved components model. In our data, the measure varies from −1.46 (Bangladesh) to 2.35 (Denmark), with higher values corresponding to better corruption control (and thus lower levels of perceived corruption). Descriptive statistics for all variables included in the analysis are reported in the online appendix (Table SI 1).
The two critical advantages of this indicator are its breadth of coverage, which is unmatched by any alternative measure, and the variety of sources employed, which makes it less susceptible to poll-specific or question-specific idiosyncrasies. Despite these advantages, though, the data pose several challenges. First, the World Bank indicator’s use for longitudinal analysis remains disputed because of changes in the methodology used to construct the index over time.Footnote 47 This concern is particularly pertinent for the period 1996–2002 when the indicator’s country coverage, the data sources, and the weighting of those sources changes extensively.Footnote 48 Therefore, we do not use the index for the early years, but employ it from 2003 after which the sample, data sources and weightings are significantly more stable. Moreover, while we are sympathetic to Kaufmann, Kraay and Maztruzzi’s argument that the methodological changes (after Reference Kaufmann, Kraay and Mastruzzi2002) are substantively not very important, we apply caution and confine ourselves to cross-sectional analysis of these data.Footnote 49 Secondly, the measure records corruption perceptions rather than the frequency or seriousness of actual corruption and it is possible that corruption perceptions deviate from the underlying phenomenon. Given the covert and illicit nature of corruption, no measures of actual corruption exist for a large enough number of cases to enable cross-national analysis. Surveys that gauge corruption experiences come closest to providing such a measure, but tend to capture only petty rather than political and grand corruption and their coverage of countries and years is as yet too limited. Fundamentally, though, corruption perceptions reflect the underlying frequency of corrupt interactions. As Treisman reports, the correlation between the World Bank measure of corruption perceptions and the main survey measures of corruption experiences is high and statistically significant, with correlation coefficients that range from 0.66 to 0.79 (p=0.01).Footnote 50 To take account of the fact that spikes in corruption perceptions can occur through raised awareness in a particular country-year, we average corruption perceptions reported for each of the countries in our analysis across a seven-year period (2003–2009).
Independent Variable: Party System Institutionalization
Two measures of party system institutionalization are available for a large enough number of cases to enable comparative analysis: average party age (of the main political parties) and electoral volatility. Both measures of party system institutionalization are widely used in comparative studies and reflect the constituent features of party system institutionalization – the organizational stability of parties and the regularity of patterns of competition.Footnote 51 The average age of the main parties is affected by (i) the entry and exit of parties into the political system (i.e. their organizational stability), as well as by (ii) changes in their relative size – which conditions what group of parties is included in the measure (regularity of patterns of competition). Electoral volatility, likewise, reflects (i) the entry and exit of parties, and (ii) changes in their electoral fortunes.
Our main analysis relies on the average age of the first and second largest governing parties and the largest opposition party or any subset of these for which party age is known,Footnote 52 recorded in the Database of Political Institutions by Beck et al.Footnote 53 Because the marginal effect of an additional year can be expected to decrease as average party age rises, we take the natural logarithm of average party age (Appendix 2 details coding decisions and data sources for all variables).Footnote 54 To test the robustness of these results, we replicate the analysis using electoral volatility (full data available for a reduced sample of fifty-two countries). We employ Mainwaring, Espana and Gervasoni’s measure for total volatility aggregated over the period 1990–2002,Footnote 55 which captures the ‘absolute value of change in the percentage of votes gained or lost by each party from one election to the next’.Footnote 56 As anticipated, average party age correlates highly with volatility (r=0.64, p=0.000).
Control Variables
We employ two sets of control variables that have been shown to affect corruption in previous cross-national work – a more parsimonious set that includes other party system features, constitutional, economic, political, and regional factors, and a more extensive set that also includes social and electoral system characteristics.
Two features of party systems beyond the level of institutionalization can be expected to affect the scope for corruption: the number of parties and the degree to which competition is ideologically structured. Much of the literature focuses on the corruption-reducing effects of the effective number of electoral parties (ENEP). When competition is low the ability of opposition parties to mount a realistic challenge to the incumbent is reduced, and, as work on Japan and South Africa suggests, this can allow corrupt incumbents to survive in office for prolonged periods of time.Footnote 57 We take the natural logarithm of ENEP because the marginal effect of each additional party can be expected to decrease as the number of parties rises (see Appendix 2 for data sources). Secondly, programmatic party structuration can be expected to affect corruption. Less ideologically structured and more clientelist forms of competition correlate with higher levels of corruption.Footnote 58 To measure the ideological structuration of party systems, we have drawn on Keefer and Stasavage’s dataFootnote 59 and followed Keefer’s approach of calculating what proportion of the three largest government parties plus the largest opposition party in a country adopts programmatic policy positions.Footnote 60
Constitutional factors are viewed as a powerful influence on governmental corruption. Constitutions which decentralize power and those that feature executive presidents are characterized by greater competition between political actors and more extensive checks and balances, features that a range of scholars argue limit the scope for corruption.Footnote 61 Others, however, formulate the opposite expectation.Footnote 62 To control for either possibility, we measure decentralization using Beck et al.’s coding of the extent to which countries have autonomous, locally elected governments and employ an indicator for democracies that feature an executive president drawing on Svolik’s coding.Footnote 63
To take account of the nature of the economy, we control for economic development measured as the natural logarithm of real GDP per capita (in constant 2000 US$). Development can be expected to curb corruption because it ‘increases the spread of education, literacy, and depersonalized relationships – each of which should raise the odds that an abuse will be noticed and challenged’.Footnote 64 Additionally, the ability of officials to extract rents in the domestic market should be reduced when that market is open.Footnote 65 Tradeopenness is measured by the sum of a country’s imports and exports as a share of GDP.
The quality and longevity of democracy also affects corruption. We include these two closely related features of democracy as alternative controls – the polity score measuring the quality of democracy in a country (which has been shown to curb corruption), and the age of democracy.Footnote 66
The more extensive set of control variables additionally captures factors that have a theoretical link to party system features, and are expected to shape corruption via electoral choice. First, social cleavages, in particular ethnic and linguistic divisions, are often associated with greater corruption in the literature because corrupt rents can be extracted more easily in divided societies with internal norms and sanctions against betraying co-ethnics. To capture the degree of ethnolinguistic fragmentation we draw on Alesina et al.’s index.Footnote 67 Secondly, electoral rules are thought to affect corruption. One argument is that electoral systems, which encourage the cultivation of personal votes – such as plurality elections or open-list proportional representation (PR) – lead to greater corruption. The focus on personal reputations gives politicians incentives to differentiate themselves from their co-partisans and ‘to use illegal proceeds to fund electoral competition’.Footnote 68 Others disagree and propose that voting for individual candidates in plurality electoral systems (compared to PR lists) makes it easier for voters to attribute responsibility and punish corruption.Footnote 69 To account for either possibility, we use a series of indicator variables to record whether a country employs Proportional Representation, Plurality, and Open Lists and include an interaction for open list PR systems (Open List*PR).
Empirical Strategy and Results
In order to causally identify the effect of party system institutionalization on governmental corruption, we pay particular attention to causal sequencing and reverse causation. These are important concerns because party systems may, in part, be shaped by politicians who wish to protect corrupt practices by limiting the transparency and effectiveness of the choices available to voters. We address this concern in two ways. First, we lag all of our explanatory and control variables by a period of seven years, which significantly outstrips the average period incumbents spend in office in our data. The length of this lag is designed to rule out reverse causation and to separate the incumbent’s attributes and record in addressing corruption today (at time t) from the political choices that affected the party system in the past (at time t−7). Since our dependent variable is averaged over a seven-year period (2003–2009), we also average our explanatory and control variables over the corresponding lagged seven-year period (1996–2002). Secondly, we confront any lingering concerns about endogeneity and reverse causation by instrumenting for party institutionalization 1996–2002, using historical party age in the 1980s, and employ two-stage least squares instrumental variable regression as an estimation strategy.
Our dependent variable, the World Bank Control of Corruption Index, is continuous, which makes ordinary least squares (OLS) regression an appropriate model choice. All analyses account for unobserved regional sources of heterogeneity in the levels of perceived corruption by including region fixed effects for the Former Soviet Union, the Middle East, Central and Latin America, Asia, and Africa.Footnote 70
Table 1 presents the results of the analysis.Footnote 71 Models 1 and 2 use the parsimonious set of control variables (controlling for the quality and age of democracy respectively), while Model 3 includes the more extensive set of controls. All three models show that corruption control improves with party system institutionalization, as anticipated by Hypothesis 1 (recall that the World Bank index records corruption perceptions so that higher values indicate better corruption control and thus lower levels of perceived corruption).Footnote 72 This is consistent with our argument that institutionalization improves the attribution of responsibility and assists voters in punishing tainted representatives.
Table 1 Party System Institutionalization Effects on Corruption Perceptions (averaged 2003–2009)
Note: OLS regressions. The dependent variable is perceptions of corruption control. Table entries are regression coefficients, standard errors in parentheses; *p<0.10, **p<0.05, ***p<0.01 (two-tailed test).
The results of Models 1 and 2 are very similar. Model 1 indicates that the effect of party system institutionalization is strongest in young party systems. Thus, holding other variables constant at their means, a rise in average party age from just 4.6 years (Ukraine) by around 1 standard deviation to thirty-five years (Czech Republic) correlates with an improvement (i.e., a reduction) of 9.3 per cent of the corruption perception score. Predictably, a further 1 standard-deviation increase in institutionalization of an already well-institutionalized party system to sixty-eight years, the average party age in Sweden, has a diminishing marginal effect and is only associated with an additional improvement of 3.3 per cent in corruption perceptions.
Turning to the control variables, both models suggest that corruption scores also respond to other party system features. Ideological party system structuration has a powerful and statistically significant effect in reducing corruption perceptions, the coefficient for the effective number of parties, however, does not reach statistical significance. The control variables that capture constitutional features – decentralized constitutions and the existence of an executive president – both have the positive sign that Persson and Tabellini and Fisman and Gatti would expect, which indicates that they tend to correlate with improved corruption perception scores. But as previous research has suggested, these effects are fragile and neither coefficient is precisely estimated.Footnote 73 Wealth – as the extant research overwhelmingly suggests – is associated with significantly improved levels of perceived corruption. Trade openness never reaches conventional levels of statistical significance. Democracy (both quality in Model 1 and age in Model 2) is also associated with improved levels of perceived corruption, though the coefficient is not always significant in this sample, which consists only of high-quality democracies.Footnote 74
In Model 3, we include the expanded set of controls for ethnolinguistic cleavages and electoral system features to explore how far party system institutionalization retains an independent effect once these underlying structural features are accounted for. As discussed above, these factors have a theoretical connection to party system features, but despite these confounding influences, party system institutionalization retains its significance and substantively sizable effect. All additional controls have the expected signs but the effects are fragile, as the contradictory findings of previous work suggest, and none of these coefficients is estimated precisely enough to reach conventional levels of statistical significance.
Next, we confront the concern about the direction of causality and endogeneity in more detail by re-specifying our Models 1 and 2, as instrumental variable regressions to guard against the potential problem that OLS estimation produces biased results if the model includes an endogenous regressor. An instrumental variable approach makes causal identification possible when the instrument is correlated with the independent variable (party system institutionalization in 1996–2002) but could not be influenced by the dependent variable (perceived incumbent corruption in 2003–2009) or correlated with its other unobserved causes.Footnote 75 As Sovey and Green note,Footnote 76 lagged values of variables in the model can provide natural candidates for such instruments. We follow this approach and instrument for party system institutionalization in 1996–2002 using average party age in the 1980s. The length of the lag ensures that incumbents – ministers, president and legislators – serving on behalf of parties in the 1980s and 1996–2002 are almost never the same. The variable is, therefore, free from the criticism which might apply to temporally more proximate measures – that specific incumbent attributes (such as effectiveness in addressing corruption) have a reverse causal connection with party system institutionalization. It is implausible that incumbent corruption in 2003–2009 could have had an effect on the organizational stability of parties and the stability of patterns of party competition in the 1980s. It is also not likely that historical party system institutionalization is systematically related to corruption perceptions through other unobserved channels. Clientelism, for instance, is accounted for in the analysis as a party system feature that varies independently of party system institutionalization. So, too, are the quality and age of democracy, two fairly robust predictors of the level of perceived corruption.Footnote 77 Average party age in the 1980s is a strong instrument, which correlates highly with party system institutionalization in 1996–2002 (r=0.66, p=0.000).
Instrumental variable regression (IV2SLS) proceeds in two stages. At the first stage a new variable is estimated from the instrument (average party age in the 1980s) and the covariates to replace the endogenous regressor. These predicted values are then used in the second stage to isolate the exogenous effects of party system institutionalization on corruption. Table 2 reports the results of the instrumental variable equivalents of our main Models 1 and 2 in Table 1.Footnote 78 The first stage estimates (columns 1 and 2) show that average party age in the 1980s is a strong predictor of party age in 1996–2002. The coefficients for this variable are large and statistically significant; the partial R 2 statistics (0.397 and 0.354 respectively) indicate that its contribution to explained variance at the first stage is sizeable, and the F-tests confirm that the instrument cannot be excluded from either model (test statistics are reported at the bottom of columns 3 and 4).
Table 2 Instrumental Variable Analysis (Two Stage Least Squares Regression)
Note: The dependent variable (stage 2) is perceptions of corruption control; table entries are regression coefficients, standard errors in parentheses. *p<0.10, **p<0.05, ***p<0.01 (two-tailed test).
Columns 3 and 4 report the results of the second stage IV2SLS regressions in which party system institutionalization is replaced with the predicted values that were estimated at the first stage. The coefficients for party system institutionalization in these models represent unbiased estimates, purged of any potential endogenous effects. As the results make clear, party system institutionalization retains its statistical significance in the instrumental variable regressions in both models, which control for the quality of democracy (column 3) and its age (column 4) respectively. The coefficients for party system institutionalization are slightly larger respectively than in the original OLS models. Substantively, these results confirm that party system institutionalization has a powerful and robust exogenous effect in improving corruption perceptions.
Note that a Durbin–Wu–Hausman test for endogeneity does not reject the null hypothesis that average party age is exogenous in both models (bottom row Table 2). Put differently, econometrically there is no evidence of endogeneity. This makes OLS the more efficient form of estimation, and we focus on Model 2 (cf. Table 1) in probing the robustness of the results further.
Robustness
We test the robustness of the results from our main Model 2 (cf. Table 1) to four further alternative specifications (results are reported in the online appendix SI 2). First, we change our main explanatory variable, and employ electoral volatility instead of average party age as our measure of party system institutionalization. Party systems that display higher levels of volatility are less institutionalized and are, as anticipated, associated with significantly poorer corruption control. Secondly, we address the fact that party system institutionalization, the age of democracy and wealth are correlated by orthogonalizing these three variables and replacing party system institutionalization with the residuals from the auxiliary regression. This procedure attributes all shared variation between party system institutionalization, wealth and democracy to the two latter variables and allows us to examine the effect of party system institutionalization on perceived corruption, net of that shared variance.Footnote 79 As a result, the coefficients on wealth and age of democracy increase in size while the coefficient for party system institutionalization is slightly reduced, but the effect remains sizeable and statistically significant. Thirdly, we change our definition of democracy, and successively restrict the sample to focus on increasingly high quality democracies. In a first step we apply a polity threshold of 7, and in a second step a threshold of 8. The results are robust to all of these alternative specifications.
In sum, the comparative analysis suggests that party system institutionalization has a substantively large and empirically robust effect on perceived levels of corruption as Hypothesis 1 anticipates. In the next section we turn to the task of testing our conceptualization of the causal process that accounts for this effect.
Nested Case Study: Panama
Our theoretical argument suggests that the two constitutive aspects of weak party system institutionalization – irregular patterns of party competition and the organizational instability of parties – mediate and undermine electoral accountability, and thereby increase the scope for corruption. In this section of the article we change the level of analysis and focus on a single case to test our second hypothesis, which focuses on this mediation effect.
We employ a nested research design in which the results of the cross-national analysis are used to guide the selection of the case study.Footnote 80 When nested analysis is used for model testing, the case should be well predicted and representative of the larger sample (rather than an outlier), so as to be informative about the anticipated causal processes. Panama’s perceived corruption score (−0.30) is predicted exactly by Model 2 in Table 1. Table 3 reports the comparisons of Panama’s values on the explanatory and control variables in Model 2 with sample mean values in the comparative data. As is consistent with Panama’s poor corruption score, the lagged average party age (1996–2002) is just 12.5 years, around half of the sample mean of 24.6 years. Other control variables are close to the mean (modal) values in our cross-national data (though it is impossible to achieve a perfect match). Crucially, the case combines a weakly institutionalized party system with meaningful within-case variation in the extent to which that weakness affects the accountability relationship between individual members of parliament and voters. As far as possible, our case selection strategy and research design, therefore, seek to isolate the impact of party system institutionalization on accountability and corruption.
Table 3 Panama Compared to Sample Mean (and Modal) Values
† Eighty observations.
We begin by showing that all constitutive aspects of weak party institutionalization are in evidence during the period that we study.
Party System Weakness in Panama
We observe Panama between 1999 and 2009, after re-democratization in the wake of the 1989 American intervention that deposed General Noriega. In the early 1990s, the former ruling party’s (Partido Revolucionario Democratico, PRD) tainted status and ties to the twenty-year long dictatorship motivated the entry of a large number of new parties into the system. However, even as Panama’s democracy grew more mature, the party system remained poorly institutionalized.
Throughout the period of our study, the constitutive elements of weak party system institutionalization are manifest. First, patterns of party competition are unstable: between two and six parties exit the system in each election and an average of 3.23 new parties enter, significantly changing the choices that confront voters every five years.Footnote 81 Electoral alliances among parties are ephemeral. Table 4 reports the alliances in which parties contested elections between 1994 and 2009. While the PRD (which has its roots in the period of military rule) and the Panamenista Party (Partido Panamenisto, PPAN) typically form the core of two alliances, other parties switch frequently from one election to the next. In every election, therefore, voters are faced with different sets of alliances, which often combine previous incumbents with some of their erstwhile opponents. Secondly, party organizations are weak and rates of party switching among legislators are high. A sizeable share of these switches is driven by the dissolution of parties and the formation of new ones.Footnote 82 Since 1989, fully 22–35 per cent of Panama’s legislators have switched parties from one election to the next. A critical aspect of our case is variation in the extent to which weak party system institutionalization affects the relationship between voters and individual politicians. While some legislators are affected by party or alliance changes, others are not. We exploit that variation to test our second hypothesis that the constituent features of weak party system institutionalization – party and alliance changes – mediate and reduce electoral accountability for corruption.
Table 4 Electoral Alliances in Panama 1994–2009
Note: Governing alliances in bold; *PDC changed its name to Partido Popular (PP) in 2001; **In 2006, PLN and SOLID merged into Partido Union Patriotica (UP).
Sources: Guevara Mann (Reference Guevara Mann2011); Perez (Reference Perez2000); Elections in the Americas Data Handbook 2005; Panama Electoral Tribunal.
Alternative Explanations
Before proceeding to the analysis, three potential alternative explanations for corruption in Panama require attention: cultural tolerance of corruption, the level of public information about corruption and the electoral system.
Governmental corruption in Panama is not explained by cultural tolerance of it. Corruption has been a salient concern for voters since re-democratization. Panama’s poor rankings in the corruption perception indices compiled by the World Bank and Transparency International capture the local view of corruption well: 88 per cent of Panamanians see corruption as widespread and 50 per cent rate it as the country’s biggest problem.Footnote 83 Over 58 per cent believe that corruption is prominent among politicians.Footnote 84 Tolerance of corruption among citizens is low and on a par with significantly less corrupt Latin American countries such as Uruguay – 88 per cent of Panamanian respondents disagree that paying a bribe is sometimes justified.Footnote 85 Moreover, Panamanian voters translate their values and perceptions into action. While legislative incumbents who were associated with corruption allegations on average lost 1,149 votes between the elections that we focus on in this study (2004 and 2009), those who were not subject to such allegations gained an average of thirty-five votes.
Governmental corruption also cannot be attributed to a lack of public information about the problem. Voters can easily obtain information about corruption. The media in Panama are free.Footnote 86 The main national newspapers regularly report on political and governmental corruption and corruption has been a prominent issue in every election campaign. Indeed, governing parties have regularly launched anti-corruption initiatives in the run-up to elections.
Finally, Panama’s electoral system facilitates the attribution of responsibility and effective electoral sanctioning. Panama has a small, unicameral legislature (seventy-nine members in 2004, seventy-one in 2009) and employs a mixed electoral system that combines twenty-six single member districts with the election of legislators by open list proportional representation (PR), mostly in small districts (district magnitude 2 predominates among the thirteen PR districts, but individual districts are larger, the largest has a magnitude of 8). The average district magnitude is 1.8. Notwithstanding other disagreements about the incentives generated by electoral systems, previous studies agree that single member and small open list districts afford voters the ability to attribute responsibility and to punish.Footnote 87 The electoral system, therefore, should not present obstacles to voters in holding their representatives accountable for corruption.
The puzzle, then, is that despite the high levels of voter concern about corruption, its electoral salience, media information and an electoral system that makes possible effective punishment, Panamanian voters regularly re-elect a significant proportion of national legislators who are subject to corruption allegations.
Data and Dependent Variable
To what extent does low party system institutionalization account for this puzzle? To answer this question we examine the electoral performance of all legislators who served in the 1999–2004 and 2004–2009 legislatures and stood for re-election, 114 in total.Footnote 88 The data are organized as legislator-election panels. The analysis employs two equally plausible dependent variables, the vote share won by the incumbent and the percentage change in that vote share. Descriptive statistics for all variables included in the analysis are reported in the online appendix, Table SI 3.
Explanatory Variables
To measure the information that voters had about corruption among their representatives we focus on media coverage in the two major national daily newspapers with digital archives – El Panama America and La Prensa. Both dailies regularly report allegations of corruption. The availability of digital archives for these papers restricts the scope of our investigation to deputies who served in the 1999–2004 and 2004–2009 legislatures.
In constructing our measure of corruption, we searched all articles that fall into the two legislative periods for items that included both the legislator’s name and the word ‘corruption’. We then analysed the content of the article to ensure its relevance to our analysis. As is consistent with our definition of corruption, we counted all instances in which deputies were alleged to be involved in the misuse of public office for personal or political gain or to have tolerated corruption by bureaucrats. This search yielded a total of 390 articles, with an average of 5.7 stories per corruption allegation. In the two legislative periods El Panama America identifies thirty-nine incumbents as associated with corruption, the corresponding figure for La Prensa is thirty-three. Overlap in the corruption allegations reported by the two papers is high. Both papers jointly report allegations about thirty legislators, 21 per cent of all incumbents serving between 1999 and 2009. Because the two papers differ in their political leanings, we record corruption allegations in two different ways. Our first measure is a count only of those instances and allegations of corruption which are reported in both papers (corruption allegations (both papers)). The second measure counts all corruption allegations reported jointly and separately by the two newspapers (corruption allegations (either paper)). Counts of corruption allegations vary between zero and sixty-five per deputy.
Our second explanatory variable is an indicator variable for Party/Alliance Change (1, otherwise 0). It records whether the relationship between voters and each individual legislator in each election is affected by the constituent features of low party system institutionalization – either electoral alliance changes (unstable patterns of competition), or party switches (the weakness of party organizations), or both. In our data, party and alliance changes affect just under a quarter of all legislators (24 per cent). Testing Hypothesis 2, that accountability for corruption is mediated by these features of party system under-institutionalization, requires that we interact party/alliance change with corruption allegations.
Controls
Guided by the literature on incumbent re-election, we take account of four types of confounding influences on the vote share won by deputies.Footnote 89 First, we capture features of their constituency. Electorates in Urban Constituencies are larger, younger and more mobile, which may generate greater vote share swings. We control for the intensity of district level competition. As the Number of Candidates in a District rises, the vote share won by individual candidates should on average be reduced. Secondly, we record features of the party on behalf of which an MP has served. Serving with the main governing party typically carries an electoral penalty. Hence we include membership of the Core Governing Party, which controls the largest number of seats in the governing alliance.Footnote 90 We also account for the effect of membership of the leading opposition party on an MP’s vote share (Major Opposition Party). Thirdly, we control for incumbent-specific features, including whether they hold a Senior Legislative Position as party leaders or assembly president, whether they have a long established relationship with their constituency (Terms in Office), and their Age.Footnote 91 The analysis also takes account of a central performance measure – the Unemployment rate in the MP’s district at the time of re-election. In the regressions that focus on vote share instead of vote share change we control for previous incumbent support (Vote Share (previous election)). Appendix 2 details the coding choices and data sources.
Party System Institutionalization and Accountability for Corruption Allegations
We are now in a position to analyse the effectiveness of accountability for alleged corruption when deputies are (and are not) affected by alliance and party changes.Footnote 92 As indicated in Hypothesis 2, we expect voters to punish corruption allegations, but anticipate that their ability to do so is diminished in the presence of party and alliance changes. Hence, we expect corruption allegations to have a negative effect on the electoral performance of deputies and that party and alliance changes mitigate this negative effect.Footnote 93
We model our two dependent variables (incumbent vote share change and incumbent vote share) using ordinary least squares regression and report all coefficient estimates with robust standard errors, clustered on election.
Table 5 presents the results of the analysis. Models 1 and 2 examine how low party system institutionalization mediates the effects of corruption allegations on the change in the incumbent’s vote share from one election to the next, Models 3 and 4 focus on the straightforward vote share of the incumbent.Footnote 94 For each pair of models the first column (Models 1 and 3) employs only the count of corruption allegations reported in both papers (corruption alleg. (both)), the second model (Models 2 and 4) employs the count of all allegations reported jointly and separately by the two newspapers (corruption alleg. (either)).
Table 5 Effects of Corruption Allegations on Vote for the Incumbent
Note: Table entries are OLS regression coefficients, robust standard errors clustered on election year in parentheses. ***p<0.01, **p<0.05, *p<0.1. The Wald Tests report the p-value for the corruption allegations variable plus its interaction with party/alliance change.
The results indicate that voters punish alleged corruption when they can attribute responsibility and exercise effective electoral choice, but their ability to punish is diminished when party or alliance changes intervene. Note that corruption allegations damage electoral performance, as indicated by the negative coefficient, but the interaction term between alleged corruption and party/alliance change always has a positive sign, which detracts from that negative effect. The bottom row of the table sheds light on these interactions. It reports the p-values of Wald tests, which examine whether the net effect of corruption allegations is different from zero, given party or alliance changes. The tests show that when low party system institutionalization interferes, the statistical significance of corruption allegations drops below conventional levels. Put differently, party and alliance changes mediate and powerfully weaken electoral accountability for corruption allegations, as Hypothesis 2 suggests. As Table 5 shows, this result is robust across both specifications and dependent variables.
Figure 1 plots the marginal effect of a corruption allegation based on Model 1 for the two values of party/alliance change (i.e., 0 and 1).Footnote 95 The figure shows that the effect of a one-step increase in corruption allegations is negative, sizeable and statistically significant only when accountability of a deputy is not affected by party and alliance changes. Under these circumstances, deputies lose around 0.2 percentage points of their vote share for each additional corruption allegation. Thus, counts of twenty, forty and sixty-five corruption allegations (the maximum observed value in our data) give rise to predicted vote share penalties of −3.95 (p=0.082), −8.06 (p=0.037) and −13.19 per cent (p=0.021) respectively. However, when party and alliance changes attenuate electoral accountability, these penalties never reach conventional levels of statistical significance and are effectively reduced to zero. Low party system institutionalization, then, undermines electoral accountability for corruption precisely as Hypothesis 2 anticipates.
Fig. 1 Marginal effect of corruption allegation while varying party/alliance change Note: The figure shows the marginal effect of corruption allegation (with 95 per cent confidence intervals) while varying party/alliance change.
Other variables behave as expected. As electoral competition becomes more intense and additional candidates enter the race, MPs suffer greater vote share losses (Models 3 and 4). Legislators with well-established relationships with their constituencies secure an average vote share bonus of over 3 per cent for each additional term served (Models 3 and 4). Across the two elections, the lack of popularity of the major opposition parties leads to an average vote share loss of around 6 per cent for legislators affiliated with them (Models 1 and 2), and older MPs suffer greater electoral losses than their younger peers (Models 3 and 4). Other variables including the main effect for low party system institutionalization have no statistically significant effects on vote share changes.
These results lend unequivocal support to Hypothesis 2 – while Panamanians punish corrupt incumbents when they are able to attribute responsibility and to cast an effective vote, party and alliance switches reduce their capacity to sanction. The contrast demonstrates that poor party system institutionalization increases the scope for corruption via the process that we identified theoretically – electoral accountability.
In closing we return to the concern of reverse causation. Could corruption in Panama be the cause of the persistence of party system under-institutionalization? If this were the case we would expect to see a significant correlation between deputies against whom there are corruption allegations and those affected by party or alliance changes, but there is no correlation between these two groups (r=0.06, p=0.525). Party system under-institutionalization in Panama has long-running historical and social causes which have weakened parties as organizations and vehicles of effective political competition. The frequency with which parties form and dissolve, MPs change party affiliations, and parties change alliances has created an environment in which corruption can flourish, but it does not seem to be an environment that is caused by corruption in the time period that we study.
Conclusion
These findings contribute to two important literatures in comparative politics: work on democratic accountability and the literature on the political determinants of corruption. A central assumption of conventional models of democracy is that regular elections make politicians responsive to voter preferences. It is this mechanism by which democracy is supposed to contain corruption and to ensure that politicians govern in the interest of the electorate. Yet, in reality, governmental corruption varies tremendously even in established democracies. Only in a subset of these democracies are elections effective instruments of accountability in the hands of voters.
Our analysis in this article offers the first systematic empirical evidence to show that party system institutionalization critically conditions the scope for corruption by affecting the quality of the electoral connection between voters and politicians. Weak party system institutionalization, we propose, erodes the ability of voters to control their representatives through regular elections and undermines the incentives for politicians to curb governmental corruption and to act in the interest of their electorate. Giving careful attention to causal sequencing, reverse causation and the causal mechanisms, we test the observable implications of this proposition at the aggregate cross-national level and at the level of individual incumbent–voter relationships. The controlled comparative analysis suggests that our argument has external validity, and that party system institutionalization has a substantively large and robust effect on perceived levels of corruption across eighty democracies around the world. We then focus on a nested case study of Panama for a process-oriented test, which shows that irregular patterns of party competition and organizationally unstable parties have precisely the meditating effect on electoral accountability that we envisage theoretically. When party system institutionalization affects accountability relationships, it can reduce the very large electoral penalties that voters will otherwise impose for corruption (up to 13 per cent in the case of Panama) effectively to zero. Taken together, the results of both analyses suggest that the system of repeated elections requires institutionalized party systems in order to function effectively. It is through stable patterns of party competition and the organizational stability of parties that elections turn into effective mechanisms for voters to align the interests of their representatives with their own and to reduce the scope for governmental corruption.
From this finding derives the central implication of our work: the literatures on the political determinants of corruption and electoral accountability have stressed the importance of institutions – both constitutions and electoral systems – in explaining why accountability sometimes fails. Our analysis shows that the nature of the party system is an additional and neglected factor in explaining the effectiveness of democratic accountability. That is, in achieving a more nuanced and accurate understanding of how well accountability in democracies works, attention to the party system is crucial.
Appendix 1 Countries Included in Analysis
Appendix 2 Data Sources
