Comparative Political Finance

Political finance regulations were inaugurated in the United States and the United Kingdom in the late-nineteenth century with the aim of combatting electoral corruption (Fisher Reference Fisher and Williams2000, 16; La Raja Reference La Raja2008, 18–33; Rix Reference Rix2008, 65). Uruguay initiated the first political finance subsidy in 1928, offering reimbursements for campaign expenses. Uruguayan reformers were concerned about the undue influence of donors and argued that subsidies would insulate parties and encourage broader representation and enhanced competition (Casas-Zamora Reference Casas-Zamora2005, 95). Costa Rica introduced post-election reimbursements in 1956, becoming the second country to legislate campaign subsidies. According to Casas-Zamora (Reference Casas-Zamora2005, 73–74, 95–97), these pioneering subsidy schemes were uncontroversial because they regulated practices that citizens viewed with suspicion and introduced a measure of transparency through accounting requirements. When countries in Western Europe began to adopt subsidies, these same arguments were recycled.Footnote ¹ Germany introduced funding for political parties in 1959 and political subsidies diffused to most Western democracies in the following decades (Koß Reference Koß2010; Scarrow Reference Scarrow2004, 661).

Countries transitioning to democracy in the 1980s and 1990s often adopted political finance regulations from these early reformers in Latin America and Europe. For example, political finance reforms in Argentina and Mexico were patterned after Spain and Costa Rica, and were seen as a replacement for ad hoc and informal transfers between states and parties (Instituto de Investigaciones Jurídicas 2011).Footnote ² Post-communist countries modeled their political finance systems on European exemplars (Roper Reference Roper2002, 175). By the 2000s, world powers, especially the United States, and international organizations such as the European Union and the Organization of American States, were pushing comprehensive political finance subsidies and regulations on the developing world (Council of Europe 2003; Koß Reference Koß2010; United States Agency for International Development 2003).

These reform efforts share several features. First, legislation is often introduced in the wake of a major political scandal (Alexander and Shiratori Reference Alexander and Shiratori1994, 3; Carlson Reference Carlson, Norris and Abel van Es2016, 103; Williams Reference Williams2000, 2). For example, after decades of limited and unenforced campaign finance regulations, the United States introduced campaign subsidies and comprehensive regulations in 1974 in response to the Watergate scandal (Briffault Reference Briffault, Norris and Abel van Es2016, 180; La Raja Reference La Raja2008, 18–33; McSweeney Reference McSweeney and Robert2000, 40). Secondly, reformers typically appeal to the importance of curbing special interests, leveling the playing field and fighting corruption (Williams Reference Williams2000). Thirdly, public opinion surveys suggest ambivalence on the part of mass publics, who tend to support political finance regulations like spending limits but are skeptical when politicians legislate subsidies that support their own campaigns (Avkiran, Kanol, and Oliver Reference Avkiran, Kanol and Oliver2016; Bryan and Baer Reference Bryan and Denise2005, 21; Primo and Milyo Reference Primo and Milyo2006).

Most democracies now have political finance laws and our dataset documents that most of those laws include direct subsidies to parties or campaigns. Laws vary considerably and in many instances are quite new, the product of an accelerating pace of reform over the past several decades (Butler Reference Butler2010; International Institute for Democracy and Electoral Assistance 2014). However, there are some common components. Most political finance systems mix public and private money (Williams Reference Williams2000, 8). The only country that insists upon sole reliance on public funding – that is, the total elimination of non-public sources – is Bhutan, though Tunisia also banned private donations in the 2011 election (International Institute for Democracy and Electoral Assistance 2014, 49). A handful of mostly European countries rely primarily on public funding with some private donations (Butler Reference Butler2010, 7). In most countries, however, public subsidies comprise a minority of party and campaign budgets.

Political Finance and Political Corruption

The concept of political finance, as employed here, encompasses three elements: the legal and statutory regulation of money in politics (stipulating what practices are legal or illegal), government subsidies (in kind or monetary) to support political parties and/or candidates for public office, and enforcement mechanisms to ensure that laws regulating behavior and subsidies are adhered to (Gokcekus and Sonan Reference Gokcekus and Sonan2016; Rose-Ackerman Reference Rose-Ackerman1978). In principle, these factors could vary independently: regulatory burdens may be strong, weak or nonexistent; subsidies may be high, low or nonexistent; enforcement may be strong, weak or nonexistent. In practice, they tend to cohere (Evertsson Reference Evertsson2013; Norris and Abel van Es Reference Norris and Abel van Es2016). It is rare to find generous subsidies unaccompanied by extensive regulation, for example. Thus we consider regulation, subsidies and enforcement as components of a compound treatment – ‘reform’ – with multiple levels.

The outcome of interest, political corruption, is commonly defined as the abuse of public office for private gain (Rose-Ackerman Reference Rose-Ackerman1999, 91). Typically, it involves the exchange of money or goods for political influence (Della Porta and Vannucci Reference Della Porta and Vannucci1999; Evertsson Reference Evertsson2013). Sometimes this agreement is explicit – a quid pro quo arrangement (for example, a bribe) whereby a specific amount of money is transferred from a principal, A, to an agent, B, in exchange for B's action on an issue of concern to A (Butler Reference Butler2010, 2; Gokcekus and Sonan Reference Gokcekus and Sonan2016; Williams Reference Williams2000, 1–2, 8). At other times, the exchange is implicit – a general understanding that A would like B to ‘do something about Policy X’. We consider exchanges of this sort to be corrupt in the general sense of the term even if they break no existing laws, as they involve what most would regard as an abuse of public office for private gain (McMenamin Reference McMenamin2012, 5). A public official may also embezzle state resources by illegally redirecting them to a campaign or into their own accounts.

In considering the impact of political finance reform on political corruption, we focus on the incentives facing politicians, that is, those who hold elective office or aspire to do so. We assume that politicians are motivated to win election (Gerber Reference Gerber1998; Green and Krasno Reference Green and Krasno1988). We also assume that money plays an important role in winning elections in a context of multi-party competition, but that its marginal utility declines (Jacobson Reference Jacobson1978; Samuels Reference Samuels2001): a campaign's first dollar is worth much more than its last (Abramowitz Reference Abramowitz1988; Gerber Reference Gerber1998; Green and Krasno Reference Green and Krasno1988; Levitt Reference Levitt1994).

Funding for election campaigns and party activities may be raised privately or publicly. Private money is susceptible to corrupt exchanges of money for influence, as defined above (Evertsson Reference Evertsson2013; Gokcekus and Sonan Reference Gokcekus and Sonan2016). Public money, by contrast, is not. In making this assertion we assume that there are no corrupt bargains between the state, which provides funding, and the recipients of that funding. In other words, we assume that the bureaucrats distributing money do not arrange kickbacks with politicians who receive funds or condition disbursement on other corrupt behavior. Note that in most instances subsidy disbursement formulas are very specific and therefore difficult to manipulate. Occasionally, election observers report that money arrives late or not at all (International Institute for Democracy and Electoral Assistance 2014, 50). According to those same sources, however, delayed or cancelled disbursements are generally a product of bureaucratic ineptitude rather than a calculated attempt to sway behavior. We have not encountered allegations of corrupt bargains between parties or candidates and state agencies around subsidy disbursement.

Thus, where reforms reduce the role of private money in electoral campaigns, incentives for corrupt behavior should decrease. This may be accomplished by regulation, for example, limiting the amount of money that candidates or parties can obtain from private sources, the kinds of donors that can contribute to campaigns, the amounts that each donor can contribute, and the transparency of these transactions (Fisher Reference Fisher and Williams2000, 16; McSweeney Reference McSweeney and Robert2000, 39–40). It may also be accomplished by subsidizing campaigns with publicly provided funds, which may replace or supplement money that officials raise privately. A body of research finds that politicians and donors skirt spending limits and disclosure requirements, and suggests that public subsidies may be a more effective form of campaign finance reform (Butler Reference Butler2010; Hogan Reference Hogan2005; La Raja Reference La Raja2014; Mann Reference Mann2003). If public money is provided on a supplementary basis, private money persists. However, its importance in winning elections is likely to be mitigated because of the declining marginal utility of money in a campaign, as noted.Footnote ³ This, in turn, should reduce candidates' dependence on private donors, making it less likely that they will engage in quid pro quo arrangements.

We must also consider the possibility of legal sanction. All political finance laws are accompanied by an enforcement mechanism designed to punish those who contravene the law (Evertsson Reference Evertsson2013). Sanctions may include loss of office, fines or incarceration. To be sure, enforcement is uncertain, and it is difficult to gauge the strength and independence of a country's enforcement mechanism. The threat of punishment alone may not significantly reduce corruption (Gans-Morse et al. Reference Gans-Morse2018; Pinto-Duschinsky Reference Pinto-Duschinsky2002, 80). Nonetheless, the possibility of legal sanction is present when a political finance regime is in place and absent when it is not. Insofar as corrupt exchanges violate political finance law, politicians who engage in such activities put themselves in legal jeopardy (Hopkin Reference Hopkin2004). There are also secondary effects on a candidate's probability of winning election, given that the suggestion of scandal – even in the absence of a successful prosecution – may harm a candidacy or even end a political career (Balán Reference Balán2011, 459; Basinger Reference Basinger2013, 386; Hopkin Reference Hopkin2004).Footnote ⁴

The media often play a key role in identifying and publicizing violations of campaign finance law. For our purposes, what is important is that the media's role as a whistle-blower depends on the existence of a public finance regulatory regime. Otherwise – if all actions are legal, or if illegality is ambiguous – there is little to report on. Note that a scandal usually involves a public official who breaks the law, and news reportage is typically drawn from prosecutors' statements, public trials and interviews with those engaged in the purportedly illegal activity. It is difficult to imagine reporters cracking down on politicians engaged in illicit acts if they are not in violation of a statute. The clearer that statute is, and the more actively it is enforced, the greater the role the press is able to play.

For all these reasons, we anticipate that political finance laws may dissuade politicians from engaging in corrupt behavior. The impact of political finance reform should also extend to non-elected officials such as bureaucrats and judges. Note that in many polities, public officials are appointed through patronage networks with the goal of enlisting them in electoral campaigns – to help get out the vote, to advertise the party's message, to distribute funds or to contribute part of their salary to the campaign (Auyero Reference Auyero2001; Casas-Zamora Reference Casas-Zamora2005; Muñoz Reference Muñoz2014; Muñoz Reference Muñoz2018). Where political finance reforms are in place, this sort of patronage system is likely to be disrupted. Politicians are less dependent on privately raised funds and they face a greater threat of discovery (by the press) or prosecution (by the judiciary or a special election board) if they are caught flaunting the law. This, in turn, should reduce the likelihood that appointed public officials such as ministers and judges will accept or solicit bribes, embezzle or improperly use state resources for campaigning.

In summary, political finance reforms attenuate the likelihood that a politician will engage in corrupt activity insofar as they (a) reduce the role of private money in campaign finance, (b) clarify the legal status of campaign finance activity (that is, the line between what is legal and illegal) and (c) enhance the risk of discovery – and hence of negative publicity and possible legal sanction. Knock-on effects should discourage corruption among non-elected officials, as discussed.

To be sure, political finance laws are only one factor influencing political corruption. Natural resources, state monopolies, unstable economies and many other factors can also fuel the abuse of public power for private gain (Gans-Morse Reference Gans-Morse2018; Rose-Ackerman Reference Rose-Ackerman1999; Treisman Reference Treisman2000; Treisman Reference Treisman2007). Nonetheless, political finance may be an important and understudied factor.

Moreover, as a policy tool, political finance reform is easier to implement than many alternative anti-corruption measures. Political parties usually welcome public funding, citizens usually support the attendant regulations, as noted, and the overall costs are fairly minimal if considered in light of the possible returns to improved governance. Hence, our topic has important practical implications.

Measuring Political Finance

A great deal of activity has occurred over the past century in an attempt to reform systems of political finance. To capture this activity in a systematic fashion we construct a dataset that covers the world at annual intervals from 1900 to 2015. This is a considerably longer period of time than existing political finance datasets such as that maintained by the International Institute for Democracy and Electoral Assistance (International IDEA) or the Political Party Database and, unlike the V-Dem measure, is based on a factual coding of laws and reports rather than expert perceptions.

Our effort builds upon the cross-sectional International IDEA database of public finance laws, which tracks answers to a number of questions related to current public campaign finance for most countries that currently hold elections (International Institute for Democracy and Electoral Assistance 2017). Helpfully, International IDEA publishes its sources – primarily, laws that are currently in force. Additional sources used in our coding include supranational reports, election observer reports, non-governmental organization (NGO) reports and academic articles. Of particular note are reports produced by the European Union's Group of States Against Corruption (GRECO), which provide detailed information on the history, provisions, implementation and public–private dimensions of campaign finance for about sixty countries. We also draw on two reports sponsored by the Organization of American States that cover Central and South America and the Caribbean. For African countries, we rely on the NGO Electoral Institute for Sustainable Democracy in Africa (EISA). We enlist election observer reports from the EU, the Commonwealth, the African Union, EISA and the Carter Center. Where laws or implementation were ambiguous, we sought advice from country experts.

We use these sources to code relevant laws establishing different types of political finance regimes in 175 countries that held national elections between 1900 and 2015. Countries that do not hold regular national elections are excluded from the dataset. Further information on coding protocol and sources is provided in Appendix A. Appendix B contains detailed information on each country, including the dates of adoption of major public finance initiatives and country-specific sources.

In selecting variables to code, we lean primarily upon the provision of public subsidies for candidates and parties. Public subsidies are one of the strongest political finance interventions and also serve as a proxy for other regulatory actions that are harder to measure in a systematic fashion like monitoring and enforcement mechanisms (Norris and Abel van Es Reference Norris and Abel van Es2016, 15). We are not aware of any public subsidy that is unaccompanied by additional regulatory measures.

We code five binary variables, as described in Table 1. (Further details are in Appendices A, B and C.) Each is coded across 175 countries and 115 years, generating 12,380 country-year observations.Footnote ⁵

Table 1.

PFSI components

Note: components of the Political Finance Subsidy Index (PFSI), each coded dichotomously (no = 0, yes = 1).

A principal component analysis demonstrates that a single dimension explains 72 per cent of the variance across the five variables. The first component also loads similarly on all five variables. This suggests that these five variables are capturing a single underlying concept that can be meaningfully combined into a unidimensional Political Finance Subsidy Index (PFSI).

We generate an additive index ranging from 0 to 5. We prefer the additive index to one formed from the first component of a principal components analysis because it is easier to interpret and to decompose. Both versions of the index correlate at r = 0.99 and render similar results, as shown in Appendix C.

Using the additive index (PFSI), we are able to visualize changes in political finance subsidies globally over time. Figure 1 graphs the average PFSI score globally from 1900 to 2015. In 1960, 98 per cent of countries in the dataset did not have any political finance subsidies. By 1980, 11 per cent had at least a subsidy provision on the books and by 1990, the number had more than doubled (to 23 per cent). By 2010, 63 per cent of countries had a law on the books establishing subsidies and most distributed those subsidies to candidates and/or parties. In 20 per cent of countries, public funds supplied the majority of the money used for political campaigns.

Figure 1.

Average PFSI score, aggregated globally, 1900–2015

Figure 2 graphs the average political finance subsidy index score by region from 1900 to 2015. Latin America led the world in subsidies until it was overtaken by Europe in the 1970s. Other regions followed suit in the 1980s. The drop in Europe's average around 1990 corresponds to the former USSR countries gaining independence; by 2000, most of these countries had added some political finance subsidies and the regional mean recovers. In the twenty-first century, political finance subsidies emerged as a global norm. Although political finance subsidies are now common in all regions, Africa, the Middle East and North Africa, and Asia lag behind.

Figure 2.

Average PFSI score, aggregated by region, 1900–2015

Case Study: Paraguay's 1996 Political Finance Reform

In this section we develop a case study to illustrate the argument laid out in the second section and to shed light on the mechanisms (M) that may lie between the apparent cause, X (political finance reform), and the outcome, Y (corruption). To serve this pathway function, we searched for a case that exhibits the relationship of interest without any obvious potential confounders (Z). Assuming for the moment that X causes Y across the population, the pathway case is a country where X co-varies with Y and Z is constant or exhibits a trend that is inconsistent with observed variation in Y (Gerring Reference Gerring2017).

We began by eliminating cases from our global sample with little or no variation in their political finance regimes like Suriname and Bangladesh. We then ran time-series analyses for the remaining countries to see where there was a strongly negative relationship between political finance subsidies and corruption, thus offering a possibility of elucidating the causal pathway of theoretical interest. Looking closely at the details of the countries that passed this test, we determined that Paraguay was the least subject to potential confounders (other events coincident in time with political finance reform that might also have affected political corruption), and therefore an appropriate choice for intensive analysis. While many countries legislate political finance subsidies as part of a large reform package or a constitutional reform during a regime change, Paraguay introduced its political finance reform several years after a new constitution and after its transition to democracy, thus sidelining these factors as potential confounders.

In addition to its pathway role, Paraguay is also arguably a crucial case for our hypothesis. Paraguayan politics are notoriously corrupt: former President Federico Franco remarked to us, ‘Paraguay is known for its triple border, narco-trafficking, car trafficking, wood smuggling, all sorts of illegal money.’Footnote ⁶ Paraguayan institutions have a legacy of corruption stemming from the Stroessner years, when the dictator openly encouraged public corruption (Hetherington Reference Hetherington2011; Nickson Reference Nickson, Lambert and Nickson1997, 24).

To develop the case study, one co-author spent three months in Asunción, Paraguay, conducting elite interviews. Informants included top political figures who were active when the 1996 reform was debated and adopted, current officials, and NGO workers and academics who focus on political finance and corruption issues. We use the full names of elected officials who spoke with us on the record because they are public figures and thus do not require confidentiality according to Institutional Review Board guidelines. The names of other informants – appointed officials, political operatives, NGO workers and academics – have been replaced with pseudonyms in accordance with Institutional Review Board guidelines on confidentiality. Appendix D contains a protocol of questions and a list of interview dates and names (for elected officials whose anonymity is not protected).

Background

Paraguay experienced institutionalized corruption under the Stroessner dictatorship, from 1954 to 1989. During the authoritarian era, General Stroessner and his Colorado party controlled the state and encouraged corruption as a way to pad salaries (Nickson Reference Nickson, Lambert and Nickson1997, 24). High-level officials were also involved in drug trafficking (Nickson and Lambert Reference Nickson and Lambert2002, 166). Although corruption has declined in recent decades, Paraguay remains the fifth most corrupt country in Latin America, according to the V-Dem corruption index. Bribery is often expected from public officials, embezzlement is commonplace, and few corruption cases make it through the courts (Hetherington Reference Hetherington2011; Lambert and Nickson Reference Nickson and Lambert2002, 166). Figure 3 illustrates Paraguay's decline from extreme corruption during the Stroessner dictatorship to moderate corruption under the current regime, using the V-Dem corruption index. Brief declines in the country's corruption score earlier in the century correspond to reforms that were passed and then rolled back under the Liberal Party (1908), General Rafael Franco (1936–38) and General Stroessner (1959).

Figure 3.

Corruption in Paraguay, 1900–2015

Note: the vertical line marks the initiation of political finance reform (1996). Corruption is measured by the V-Dem political corruption index.

Following a coup that returned the country to civilian rule, Paraguay introduced a new constitution (1992) and held its first democratic elections (1993). Colorado politicians, in consultation with representatives from the largest opposition party, the Liberal Party, introduced political finance legislation following the 1993 election. Both parties wanted legitimate state funds for their campaigns and party organizations. Additionally, the Colorados were confident that they could maintain their national majority against a better-funded opposition, particularly since subsidies are allocated based on how many seats each party has in Congress, an arrangement that continues to favor the Colorados. The law explicitly outlawed the use of state resources and government facilities for campaigns and required parties to publish financial accounts. The law also introduced substantial subsidies for campaigns and yearly subsidies for parties; public subsidies now cover roughly half of a campaign's expenses, though the proportion of public to private money is dropping. The Colorados and Liberals fund day-to-day operations with a mix of private and public money, while smaller parties rely almost entirely on public funding. In 1996, eighteen newly appointed electoral judges were charged with enforcing the law, which employs injunctions and confiscation to punish violations.

Subsidies were first distributed to parties in 1997 and to candidates in the 1998 elections. They appear to have had an immediate electoral effect. With a guaranteed funding floor, Liberal politicians were able to run viable campaigns and won additional seats in the 1998 elections, especially at local levels. Still, the Colorados maintained their legislative majority and have held the executive in all but one election.

Potential Confounders

Paraguay's decline in corruption might be credited to the departure of Stroessner and the subsequent transition to democracy. However, this would run against the experience of most countries, where democratic transitions are accompanied by an increase in corruption – especially electoral corruption, spurred by the fact that parties and politicians must now compete in multi-party elections, and often choose to do so with an expenditure of funds to buy votes or influence turnout (Keefer Reference Keefer2007; McMann et al. Reference McMann2017). We see considerable evidence of this in the Paraguayan case, vindicating the widely accepted view that the first-order effects of democratization on corruption are negative rather than positive. In the 1993 election, the Colorados were the only party with a developed party infrastructure and experienced personnel, as well as unrestricted access to state resources (Riquelme and Riquelme Reference Riquelme, Riquelme, Lambert and Nickson1997, 48). The 1993 election was marked by rampant vote buying and embezzlement, and Colorado candidates ran their campaigns out of state offices while using state resources. As a result, the Colorados easily beat the disorganized and underfunded Liberal opposition (Riquelme and Riquelme Reference Riquelme, Riquelme, Lambert and Nickson1997, 48).

To further address potential confounders, we analyze lagged dependent variable models of political finance and corruption in Paraguay in Table 2. These models have three specifications with different sets of controls, including GDP, growth, urbanization, polyarchy and polyarchy squared, as well as regular elections. The data come from a larger dataset described in the following section.

Table 2.

PFSI and corruption in Paraguay

Outcome: political corruption (V-Dem), forward lagged by five years. Estimator: ordinary least squares with a lagged dependent variable. *p < 0.1 **p < 0.05 ***p < 0.01 ****p < 0.001.

As Table 2 demonstrates, political finance is a negative predictor of corruption in Paraguay and is statistically significant at the p < 0.05 level in all three models. Even controlling for important potential confounders, political finance retains its significance – indeed, the estimated coefficient is virtually unchanged vis-à-vis the bivariate baseline specification (Model 1) – suggesting that Paraguay functions as a pathway case.

Panel Analysis

In this section, we evaluate the main hypothesis using a panel analysis incorporating sovereign states observed annually from 1900 to 2015. As a first step, we need to think carefully about the functional form of the relationship between political finance reform and political corruption. We presume that the impact of political finance subsidies accrues over time but with diminishing marginal returns. To operationalize this expectation, we construct a stock variable that adds the political finance subsidy index score registered for a country in each year (starting from 1900) with a 1 per cent depreciation rate (following Gerring et al. Reference Gerring2005). We further presume that the impact of a country's accumulated stock has diminishing marginal returns on corruption. The stock variable is therefore transformed by the natural logarithm. In this fashion, we attempt to track small, long-term effects over time.

To measure corruption, we rely on the corruption index developed by McMann et al. (Reference McMann2017) as part of the Varieties of Democracy (V-Dem) project (Coppedge et al. Reference Coppedge2017). The corruption index aggregates indicators of corruption in different arenas – executive, legislative, public sector and judicial. Underlying data is coded by country experts and integrated into a global dataset with a Bayesian latent variable measurement model (Pemstein et al. Reference Pemstein2015). The indicators measure the latent concepts of executive, legislative, public sector and judicial corruption by aggregating expert perceptions of these types of corruption and expert interpretations of primary and secondary sources from a given country over time. McMann et al. (Reference McMann2017) report that the V-Dem index covers more types of corruption than other corruption datasets and aligns with theoretical definitions of corruption. They also find that the individual indicators load heavily on a single dimension, justifying the construction of an aggregated index (McMann et al. Reference McMann2017, 9–12).

We recognize that all corruption indices are prone to measurement error (Jensen, Li and Rahman Reference Jensen, Li and Rahman2010; Olken and Pande Reference Olken and Pande2012; Sequeira and Djankov Reference Sequeira and Djankov2014). In particular, corruption perception indices are a controversial proxy for the level of corruption in a given country because perceptions and experiences often diverge (Donchev and Ujhelyi Reference Donchev and Ujhelyi2014; Razafindrakoto and Roubaud Reference Razafindrakoto and Roubaud2010). Unfortunately, the only global corruption experience survey, Transparency International's Global Corruption Barometer, offers only nine observations between 2003 and 2017. Likewise, questions about corruption experience do not overcome fundamental measurement problems, namely that respondents have incentives to hide or under-report the corruption that they engage in and may be unaware of what goes on outside their immediate purview (Della Porta and Vannucci Reference Della Porta and Vannucci1999; Olken Reference Olken2009). Thus, while imperfect, corruption perception indices represent our best guess about the level of corruption in different countries over time. Their validity is enhanced when measures of democracy and per capita GDP are included on the right side of a causal model, minimizing potential sources of measurement bias from these background factors.

The V-Dem index offers a substantial improvement in coverage relative to other indices, as yearly estimates extend from the turn of the twentieth century to the present, offering an unparalleled time series. Other cross-national measures such as Transparency International's Perceptions of Corruption index, the World Bank's corruption estimates, or the Political Risk Services Group's International Country Risk Guide, cover only the past several decades. Despite differences in measurement strategy, the Varieties of Democracy corruption index is highly correlated with other indices of corruption, such as the World Bank Corruption control index (Pearson's r = 0.91), Transparency International's Corruption Perceptions Index (0.87) and the International Country Risk Guide's corruption index (0.69). To test the robustness of our benchmark model we include pooled ordinary least squares (OLS) models with Transparency International's Corruption Perceptions Index and the World Bank's Control of Corruption estimates as dependent variables (Appendix Tables C22 and C23), and run our benchmark model using the International Country Risk Guide's corruption index (Appendix Table C26).

Indices based on respondent experiences (as opposed to perceptions) are even more limited in spatial and temporal coverage. Importantly, systematic error in measurement is most likely to be associated with cross-country comparisons insofar as corruption may be understood differently or manifested differently in different contexts, rendering estimates that are not cross-nationally equivalent. The expansive data provided by V-Dem allows for a time-series format (with country fixed effects), in which a country is compared to itself over time. This sort of comparison appears to be less prone to systematic error in measurement that might lead to biased estimates.

Expert perceptions indirectly measure underlying phenomena such as corruption with some level of error. Of particular concern for this project is the possibility that experts might expect political finance reforms to decrease corruption. Observing such reforms in the historical record, they may record a decrease in corruption in the following years. If so, the relationship is subject to circularity. We do not anticipate this to be the case since political finance reforms are often introduced in the wake of a scandal and scandals are likely to increase corruption perceptions (Alexander and Shiratori Reference Alexander and Shiratori1994, 3; Carlson Reference Carlson, Norris and Abel van Es2016, 103; Williams Reference Williams2000, 2). However, to systematically address this issue, we use V-Dem data to ‘clean’ our PFSI estimates in a robustness test explored below (Model 3, Table 5).

Initial Tests

Table 3 presents time-series cross-sectional analyses where corruption is regressed against a political finance subsidy index (PFSI). Model 1 employs OLS along with country and year fixed effects, with errors clustered by country. Right-side variables are lagged five years behind the outcome. (Appendix C contains comprehensive variable definitions, sources and descriptive statistics.) On the right side we include only the variable of theoretical interest, the PFSI, measured as a stock variable and transformed by the natural logarithm, as explained.

Table 3.

PFSI and corruption

Outcome: political corruption (V-Dem), forward lagged by five years. For GMM, analysis is conducted across 5-year intervals. Estimators: OLS (ordinary least squares), FE (country fixed effects), RE (random effects), GMM (generalized method of moments). Standard errors clustered by country. **p < 0.1 **p < 0.05 ***p < 0.01 ****p < 0.001.

Model 2 adds several covariates commonly regarded as causes of corruption – and, perhaps, of political finance reform. Democracies regulate political finance more heavily than non-democracies. Additionally, a number of studies have shown that democracy has non-linear effects on corruption (Treisman Reference Treisman2000; Treisman Reference Treisman2007; Rose-Ackerman Reference Rose-Ackerman1999) and perhaps on policy implementation, including political finance (Norris and Abel van Es Reference Norris and Abel van Es2016; Rose-Ackerman Reference Rose-Ackerman1999). We therefore include an index measuring the quality of democracy – the Polyarchy index (Teorell et al. Reference Teorell2016) – and its quadratic. Virtually all studies of corruption regard economic development, measured as per capita GDP (log) with historical data from Fariss et al (Reference Fariss2017), as a background factor (Mauro Reference Mauro1995; Treisman Reference Treisman2000; Treisman Reference Treisman2007). Owing to the ubiquity of these variables in studies of corruption, and their strong performance in our models, we regard the specification in Model 2 as the benchmark model.

In Model 3, we add several additional covariates that are less central in the literature but may nonetheless serve as confounders. These include urban population, GDP growth and regular elections (Fisman and Svenson Reference Fisman and Svensson2007; Mauro Reference Mauro1995; Treisman Reference Treisman2007). Model 4 replicates this model with a random effects estimator.

In Model 5, we enlist a dynamic panel model known as system generalized method of moments (GMM), developed explicitly to study sluggish variables (Blundell and Bond Reference Blundell and Bond1998). The benchmark specification includes a one-period lagged dependent variable along with a time trend (replacing the annual dummies). Data is aggregated at five-year intervals, rather than annually, to mitigate the problem of too many instruments (Roodman Reference Roodman2009). We also restrict the number of lags used for instrumentation to three (third to fifth lag). The model treats PFSI as endogenous, and performs well on all relevant specification tests. The Hansen J-test p-value is 0.92, the Ar(2)- and AR(3)-test p-values are, respectively, 0.46 and 0.39. There are thirty-seven instruments, considerably fewer than the 156 cross-section units. This suggests that Model 5 yields a consistent estimate of the relationship between PFSI and corruption.

Tests in Table 3 demonstrate that the relationship between PFSI and corruption is highly significant and fairly stable across all model specifications. Appendix C contains additional specification tests with covariates that measure other factors sometimes associated with corruption including media censorship, trade and natural resource income.

Figure 4 displays how predicted values for corruption vary with PFSI stock (log) in Model 2, the benchmark model, holding other variables at their means. Note that the confidence intervals increase at higher values of the PFSI because few countries have had subsidies in place for more than thirty years. To gain a more specific understanding of this relationship, let us consider the hypothetical case of a country with no prior political finance experience (PFSI = 0) that adopts a full set of measures (a maximum score on the index) in year 1 and maintains those regulations over the subsequent decade (PFSI = 4.02). Levels of corruption in this country are predicted to fall from 0.49 to 0.42, or 0.37 standard deviations.

Figure 4.

Predicted values for corruption as PFSI stock (log) increases

Note: Predicted values calculated holding all other variables in the benchmark specification (Model 2, Table 3) at their means, and surrounded by 95% confidence intervals.

Disaggregated Tests

Corruption takes many different forms, providing an additional point of leverage into the relationship between political finance reform and political behavior. While most corruption measures offer only a single variable, V-Dem offers several.

In the first section of Table 4 we interrogate the components of the V-Dem index of political corruption, which measure corruption in the executive, the legislature, the public sector and the judiciary, respectively. Each forms a dependent variable in subsequent regression tests, based on the benchmark specification (Model 2, Table 3). The estimated coefficient for PFSI is negative in all four models, and statistically significant in three. We suspect that small differences in the estimated coefficient for PFSI are the product of the different levels of year-to-year variability across these four outcomes rather than of differing underlying causal relationships.Footnote ¹⁸ In any case, our theoretical presumption that political finance reform has wide-ranging effects across different governmental bodies is confirmed.

Table 4.

PFSI and disaggregated corruption measures

Outcome: corruption in different sectors (V-Dem), forward lagged by five years. Estimator: ordinary least squares panel analysis with two-way fixed effects (unit and time) and standard errors clustered by country. *p < 0.1 **p < 0.05 ***p < 0.01 ****p < 0.001.

The final column of Table 4 tests corruption in the media. This variable, also drawn from the V-Dem project, is not included as a component of the public corruption index as the media is not – at least not formally – a part of the state. The V-Dem media corruption variable measures the likelihood that members of the media take money or receive favors in exchange for altering their coverage of political events. We have no reason to suspect that changes in political finance regulations would have any effect on levels of media corruption. This alternative dimension of corruption thus provides an important placebo test.

Note that when countries are on a reform trajectory it is reasonable to expect that lots of good things will happen at the same time, generating a problem of simultaneity between political finance reform and curbs on corruption. For example, leadership or a reform movement could be driving both political subsidy reforms and general improvements in corruption across many sectors of society. Our models attempt to deal with this potential problem by adopting a five-year lag structure, but lags do not necessarily suffice to overcome the problem of holism in politics (good things going together). Here, the placebo is helpful. If the covariation between PFSI and Public Corruption is spurious, we would expect it to have a (spurious) relationship with other variables in the V-Dem dataset that measure similar outcomes, like Media Corruption. Model 5 in Table 4 shows that there is no such relationship. The estimated coefficient for PFSI is positive, and nowhere near statistical significance. We regard this as strong corroborating evidence of a causal interpretation of the main finding.

Robustness

This section addresses several additional concerns about the quality of our data and the robustness of the analyses in previous tables.

In Model 1 of Table 5, we return to the benchmark model (Model 2, Table 3), this time with an alternate measure of political finance reform. This measure is drawn from V-Dem, where coding is performed by multiple country experts, aggregated in a latent-variable measurement model. (It is not logged because the V-Dem measurement model ensures that the underlying data approximate a normal distribution.) The V-Dem measure captures public campaign finance and involves a judgment of how central public finance is to national campaigns. Importantly, the V-Dem measure also includes perceptions of private money's importance in campaigns, which the PFSI does not. Estimates from this model confirm our finding that political finance reform is correlated with lower levels of political corruption. (We have more confidence in the accuracy of our factual coding than in V-Dem's expert coding, which is susceptible to systematic error. Note also that because the outcome variable is also drawn from the V-Dem dataset there is a potential problem of circularity.)

Table 5.

Political finance (variously measured) and corruption

Outcome: political corruption (V-Dem). DV is forward lagged by five years except public subsidies in Model 3. Estimator: ordinary least squares panel analysis with two-way fixed effects (unit and time) and standard errors clustered by country. *p < 0.1 **p < 0.05 ***p < 0.01 ****p < 0.001.

Model 2 introduces a measure of political finance disclosure regulations. V-Dem's disclosure variable asks experts to code if a country has political finance disclosure laws and, if so, the extent to which parties and candidates adhere to them. The theory outlined above suggests that disclosure requirements increase the possibility of sanctions and therefore are likely to reduce corruption. In conjunction with our political finance subsidy index, disclosure laws do appear to have a negative and statistically significant relationship with corruption, corroborating our expectation.

In the final model in Table 5 we address an additional problem of measurement. V-Dem experts coding levels of corruption in a country may regard political finance reforms as prima facie evidence that levels of corruption are decreasing in a given year. If so, this expectation should be captured in the V-Dem measure of political finance in that year. By including this covariate in our model – along with the PFSI – we should be able to purge the outcome (corruption) of this anticipatory bias. Model 3 excludes other covariates, which introduce problems of collinearity. (Appendix C, Table C25 contains additional models with more controls.) We find that the PFSI remains a negative and statistically significant predictor of corruption even after cleaning the estimate of potential coder bias. Indeed, the estimate from this model is remarkably close to the benchmark model in Table 3.

Further robustness tests are contained in Appendix C. There, we explore changes in model specification, estimator and sample, for example, alternative dependent variables, lagged dependent variable models, pooled OLS, varying lags of the right-side variables, and split-sample tests that restrict the sample to recent decades or drop influential cases or regions. We also show the results for the PFSI in the benchmark model when varying depreciation rates are used to construct the stock for this index, when the components of the PFSI are aggregated using the first component of a principal components analysis, when the PFSI is measured as a level (rather than stock) variable, and when the stock variable is measured in a linear, rather than logarithmic, form. We include controls for natural resources, imports and exports, and media censorship, all of which have been linked to corruption in the literature. We also evaluate alternative dependent variables using corruption data from Transparency International (1995–2015) and the World Bank (1996–2015) in pooled OLS models, as well as the International Country Risk Guide's corruption data (1984–2003) in the benchmark model with varying lags. In these tests, the PFSI shows a strong, negative and statistically significant relationship with political corruption.