An Insider–Outsider Theory of Popular Tolerance for Corrupt Politicians



This article addresses the puzzle of electoral support for corrupt politicians in emerging democracies by examining citizens’ varying attitudes toward political corruption. We make an important theoretical distinction between perceptions of and tolerance for corruption, and argue that these different attitudes vary across individuals depending on whether they are political insiders or outsiders. We test our theory using Afrobarometer survey data from 18 sub-Saharan African countries and find that individuals included within clientelistic networks simultaneously perceive corruption as ubiquitous and are more tolerant of malfeasance. Meanwhile, those individuals with partisan or ethnic ties to the incumbent are less likely to consider corruption as widespread. Finally, we explore whether variation in attitudes toward corruption influences citizens’ voting behavior, and find that insiders are less likely to “vote the rascals out.”


The recent wave of corruption scandals involving high-ranking officials in Turkey, Malawi, Greece, and Brazil spurred widespread public protests and condemnation. However, despite public outrage over political graft, corrupt politicians continue to command significant popular support. This contradiction, referred to as the “unpopular corruption and popular corrupt politician” paradox by Oskar Kurer (2001), remains one of the most intriguing puzzles in political science.

One dominant explanation attributes voters’ support for corrupt politicians to their lack of information (Ferraz and Finan 2008; Olken 2009; Winters and Weitz-Shapiro 2013). Yet, a recent wave of research shows that even when voters are knowledgeable about corruption, they are willing to excuse venality because they benefit from corrupt incumbents’ redistributive policies (de Sousa and Moriconi 2013; Manzetti and Wilson 2007). Building on these studies, we account for the corruption paradox by examining how voters’ relative positions in the political and economic system shape their attitudes toward corruption and influence their voting behavior.

Toward this end, we first distinguish between two parallel, yet different, public attitudes toward corruption: perception and tolerance. We define the perception of corruption as the degree to which citizens believe a political actor is engaged in corrupt practices. Tolerance of corruption, conversely, denotes citizens’ proclivity to condone politicians’ malfeasance. This conceptual distinction not only helps us reevaluate current empirical research focusing on the consequences of perceived corruption, but also provides us unique leverage in understanding why voters simultaneously condemn corruption yet support corrupt politicians.

As a second step in solving the corruption paradox, we propose an insider–outsider framework to understand the way in which voters form their corruption attitudes. Importantly, we draw the distinction between insiders and outsiders along two dimensions: cost–benefit instrumentality and affective identity. We define patronage insiders as those who belong to the patronage network, and identity insiders as those who share either partisan or ethnic affiliation with the incumbent.

Finally, we also apply this insider–outsider framework to account for the puzzling electoral success of corrupt politicians. We argue that voters are less likely to punish unscrupulous politicians at the ballot box when they share either an instrumental or identity bond with the incumbent. Put another way, insiders are less likely to “throw the rascals out” either because of their in-group bias or because the incumbent “brings home the bacon.”

Using survey data from 18 African multiparty regimes, we find that patronage insiders are more tolerant of political corruption despite perceiving higher levels of venality, while identity insiders are less aware of political corruption. Importantly, we find insiders’ electoral support of the incumbent is less affected by corruption. This finding is especially important in the context of Africa, where the electoral consequences of popular corruption attitudes have been underexplored. Our findings can also be generalized to other regions of the world where clientelism is pervasive and social identities and partisan cleavages are salient.

Perceptions of versus Tolerance for Corruption

The traditional literature has not adequately addressed the distinction between two fundamental attitudes toward corruption: corruption perceptions and corruption tolerance. Most scholars focus disproportionately on perceptions, while the few researchers that jointly examine perceptions and tolerance either conflate the two concepts or treat them as closely connected.

One reason for the lack of conceptual clarity stems from how corruption has been defined. One approach to define corruption relies on public opinion to determine what political graft is. Arnold Heidenheimer (2002), for instance, classifies corruption as white, gray, or black depending on elites and citizens’ willingness to punish the political act. Accordingly, perceptions of and tolerance for corruption may even be viewed as analogous as actions that are publicly deemed corrupt are, by definition, less tolerated. Another approach that contributes to the conceptual entanglement of perception and tolerance is the legalistic view. According to John Gardiner (2002, 29), “if an official's act is prohibited by laws established by the government, it is corrupt; if it is not prohibited, it is not corrupt even if it is abusive or unethical.” Hence, the legalistic approach relates an individual's perception of corruption to their views about acts punishable by law.

Recently, scholars have attempted to disentangle perceptions of from tolerance for corruption. Christopher Anderson and Yuliya Tverdova (2003) argue that the effect of corruption perceptions on political support is not consistent across citizens. Instead, corruption had a less harmful effect on the political support of citizens who voted for the incumbent. Jong-sung You and Sanjeev Khagram (2005) also highlight the distinction between tolerance and perceptions by showing that citizens in countries with high income inequality tend to be more aware of corruption and see bribe taking as acceptable.

Based on these recent contributions, we explicitly distinguish perceptions of corruption from tolerance for corruption. We define perceptions of corruption as the degree to which citizens believe a political actor is involved in fraudulent practices. Conversely, tolerance for corruption denotes citizens’ inclination to condemn a political actor's engagement in graft. Our conceptualization of corruption tolerance is adapted from research on political tolerance, which considers tolerance “as a willingness to put up with those things that one rejects” (Sullivan, Marcus, and Piereson 1979, 784). Individuals adjudicate tolerable from intolerable acts by weighing the consequences of the objectionable activity and their affective feelings toward the responsible persons. Specifically, materialistic interests and emotional ties can sometimes influence individuals to tolerate actions they normally would reject.

Distinguishing between corruption perceptions and tolerance serves both theoretical and empirical purposes. For instance, although studies have convincingly documented the pernicious effect of corruption perceptions on citizens’ support for democracy, it seems plausible that the corruption effect might also depend on the extent to which voters tolerate malfeasance. We test our conjecture empirically and our results suggest that corruption, indeed, matters less for citizens more tolerant of graft.1 Thus, the distinction between corruption perceptions and tolerance adds another layer of causal complexity that the literature has yet to fully explore. Importantly, distinguishing between perceptions and tolerance sheds light on the corruption paradox, as one can plausibly attribute voters’ seemingly self-contradicting electoral behaviors to their divergent attitudes toward corruption. Our analysis below illustrates this point clearly: We show that patronage insiders exhibit higher levels of tolerance for their corrupt patrons despite perceiving widespread corruption.

Finally, probing how voters develop their attitudes toward corruption brings to light potential biases plaguing the quality of subjective measures of corruption. Several scholars have demonstrated there is little empirical evidence linking perceptions of corruption to actual corruption (Olken 2009; Seligson 2002; Treisman 2007). Even experts tend to exaggerate the extent of graft, relying instead on their ideological positions to judge corruption levels (Razafindrakoto and Rouband 2010). Hence, understanding how individuals form their attitudes about corruption, and the role that context and relationships may play, is critical.2

Insider–Outsider Theory

Disentangling perceptions of from tolerance for corruption, however, is only the first step in understanding citizens’ attitude formation toward graft. We next propose an insider–outsider theory to examine how voters’ corruption attitudes can be shaped by their positions vis-à-vis the dominant political power structures within a country. We highlight the distinction between insiders from outsiders based on voters’ ability to access state institutions and resources via an affiliation with the incumbent.

To be sure, we are not the first to highlight the insider–outsider status, as research on democratic transitions (Bratton and van de Walle 1997), redistribution (Rueda 2005), and labor market economics (Lindbeck and Snower 1998) have identified differences among political actors as a principal analytical framework.

At the core of the insider–outsider distinction is the notion that insiders receive preferential access from the incumbent and, consequently, a range of excludable goods. In turn, this preferential access structures insiders’ attitudes and behaviors in ways that differ from outsiders. Specifically, we highlight two salient mechanisms by which voters can identify with the incumbent: cost–benefit instrumentality and affective identity. Cost–benefit instrumentality involves voters receiving tangible benefits from the incumbent, whereas affective identity refers to voters sharing a sense of affinity toward the incumbent. Based on these two mechanisms, we identify two types of insiders: Patronage insiders refer to those who belong to the patronage network of the incumbent, whereas identity insiders represent those who share either partisan or ethnic affiliation with the incumbent.

Notably, the distinction between identity and instrumentality is analytical, and not necessarily empirically distinguishable. Indeed, some individuals may fall into both categories. For instance, in Africa, ethnic insiders are frequently also patronage insiders, especially if the incumbent actively favors his co-ethnics (Posner 2005). Similarly, patronage insiders can also be linked by partisanship to the incumbent. Therefore, we view both concepts as ideal types occupying opposite positions on a continuum, while, in reality, most insiders are more likely to fall somewhere along the dimension between the two extremes.

Patronage Insider: Recipient of Patronage Benefits

We regard clientelism as an informal exchange relationship in which a patron offers benefits in return for the electoral allegiance of a client (Stokes et al. 2013). These benefits may be material or nonmaterial, ranging from money and food to employment, security, and preferential access to government services. Additionally, clients receive direct and personalized assistance, commonly referred to as “problem solving,” from their patrons to alleviate various personal or family-related issues clients usually encounter on a daily basis (Auyero 2000).

Clientelism tends to persist, and recent scholarship suggests that citizens remain committed to their patrons out of fear that their benefits or access will be revoked. We extend the benefit-based explanation and argue that the receipt of patronage benefits and the fear of losing such benefits are two critical parameters in citizens’ cost–benefit analysis. In particular, although the costs of corruption are borne by everyone, only insiders enjoy the proceeds from their patron's corrupt practices. Under such circumstances, patronage insiders are more likely to tolerate graft because they monopolize the benefits and only partially internalize the costs.

Moreover, membership in a patronage network can also increase one's perception of corruption. Generally, information about fraudulent transactions is limited because of the normative and legal ramifications that would follow if information were exposed. Nevertheless, we suggest that voters can infer corruption from patronage by linking politicians’ abuse of public office and their doling out of particularized goods. Particularly in African countries, as Michael Bratton (2007, 108) claims, “clientelism and corruption are best viewed as two sides of the same coin of distributive politics.” In fact, voters may have an incentive to gather in-depth knowledge of patronage transactions to evaluate the credibility of candidates’ future distributive promises (Chandra 2004). Therefore, we expect patronage insiders to have a better grasp of clientelistic network dynamics and the extent of graft relative to outsiders. Taken together, we formulate two hypotheses to account for corruption attitudes among citizens within clientelistic networks:

Hypothesis 1: Patronage insiders are more likely to tolerate corruption relative to patronage outsiders.

Hypothesis 2: Patronage insiders are more likely to perceive higher levels of corruption relative to patronage outsiders.

Identity Insider: Co-partisans or Co-ethnics of the Incumbent

Mass attitudes toward corruption are not only shaped by the expectation of material benefits, but also by citizens’ affective ties to the incumbent. Partisanship is one major source of such identity formation. Conventionally, partisanship represents a personal, psychological attachment to a political party based on a sense of shared identity with a social group (Bartels 2002). Other studies associate partisanship with citizens’ short-term considerations, such as performance evaluations of the incumbent, issue preferences, and political experiences (Fiorina 2002). Regardless of its source, party identification provides a perceptual frame through which voters process information and make political decisions (Zaller 1992).

We argue that voters with a partisan allegiance to the incumbent may exhibit a partisan bias and perceive lower levels of corruption involving the incumbent. Specifically, co-partisans are less inclined to search for discriminating information on corruption involving the incumbent compared with other citizens (Davis, Camp, and Coleman 2004). Moreover, when co-partisans are exposed to information implicating their party, they are less likely to accept it as the information would undermine the positive distinctiveness of the party. Co-partisans are also likely to dismiss corruption allegations concerning their partisan leaders as politically motivated attacks by opposition parties or the media (Anduiza, Gallego, and Muñoz 2013).

In some parts of the world, an equally important source of voters’ identification with the incumbent is ethnicity. Scholars suggest that ethnic identity formation is a dynamic and situational process subject to the influence of myriad social, economic, and political factors (Eifert, Miguel, and Posner 2010; Posner 2005). Ethnic bias can develop through a psychological process of group differentiation (Gibson and Gouws 2000) or through particularized trust of co-ethnics (Uslaner and Conley 2003). Regardless of how ethnic bias emerges, we argue that co-ethnics of the incumbent are more likely to perceive lower levels of corruption. We contend that ethnicity represents a schema for processing political information and making political judgments about the incumbent. Just like co-partisans, we expect co-ethnics to be predisposed to searching for information that confirms positive attributes of their co-ethnic leaders. Co-ethnics are also likely to be critical of the source and content of information that implicates their co-ethnic leaders. Of course, partisan and ethnic insiders may overlap in context where political parties make ethnic appeals to mobilize electoral support. Overall, we hypothesize that identity insiders—either through their partisan or ethnic affiliation with the incumbent—will turn a blind eye to their corrupt leaders.

Hypothesis 3: Identity insiders are more likely to perceive lower levels of corruption relative to identity outsiders.

Not only might affective identity alter perceptions of corruption, but it may also structure citizens’ tolerance for corruption. Goren's (2005) finding that partisanship affects one's moral tolerance is relevant to our understanding of corruption tolerance, as both concepts probe citizens’ acceptance of individuals whose actions they do not typically endorse. Partisan bias can also cause co-partisans to stomach corruption allegations involving their incumbent better than other citizens (Anduiza, Gallego, and Muñoz 2013). Similarly, we also expect ethnic bias to make co-ethnics more forgiving of their ethnic leaders. More generally, identity insiders may downplay their aversion to corruption because they value their social ties—partisan or ethnic—with the incumbent more than integrity and legality. Hence, we expect identity insiders to display higher tolerance of malfeasance as their values vis-à-vis corruption have been altered by their affiliation with the incumbent.

Hypothesis 4: Identity insiders are more likely to tolerate corruption relative to identity outsiders.

Empirical Analysis

We utilize data from Round 3 of the Afrobarometer survey (AB) to test these hypotheses. Our decision to focus on African countries was guided by several important considerations. First, many countries in Africa are affected by widespread political corruption. According to Transparency International, 90% of countries in sub-Saharan Africa fell in the bottom half of their 2014 Corruption Index. Meanwhile, the African Union estimates that over 25% of the gross domestic product in African states is lost to corruption each year. In addition, clientelism and politicized ethnicity are pervasive in many African countries (Posner 2005). This confluence of high rates of corruption, clientelism, and politicized ethnicity in Africa offers a ripe context for analysis. Second, while there have been notable studies on the sources and consequences of corruption attitudes in East Asia (Chang and Chu 2006), Latin America (Seligson 2002), and Eastern Europe (Anderson and Tverdova 2003), scarce studies have examined these dynamics in Africa. Third, AB contains a unique set of questions that tap into citizens’ corruption attitudes. To the best of our knowledge, comparable questions on corruption tolerance have not been available in other comparative public opinion data sources.3

Dependent Variables

Political corruption can involve various kinds of political actors. In most countries, however, few are as important as the president and national government officials. Therefore, we capitalize on AB's survey question asking citizens’ perception of corruption involving the president, appointees in the president's office, and other national government officials. We create a composite index of corruption perceptions by averaging across the responses to these questions, where 0 indicates “none,” 1 “some of them,” 2 “most of them,” and 3 “all of them.” The resulting corruption perception variable has a mean of 1.253 and a standard deviation of .816, suggesting that the average citizen considers at least some members of the executive to be corrupt. Preliminary analysis also reveals that corruption perceptions vary substantially within countries, reinforcing the need to understand why perceptions differ among voters.4

The second dependent variable taps into the notion of tolerance for corruption. We construct an indicator of corruption tolerance based on the definition of political corruption—the misuse of public office for personal benefit. AB provides a set of scenarios involving politicians’ abuse of power for personal gain, and asks respondents whether such actions are “not wrong at all,” “wrong but understandable,” or “wrong and punishable.” These scenarios include a government official giving a job to an unqualified family member, demanding a favor or special payment for some service that is part of his job, and locating a development project in an area where her friends and supporters live. For each scenario, we consider respondents tolerant of corruption if they answer “not wrong at all” or “wrong but understandable.” To ensure our results are not driven by the arbitrary choice of any one scenario, we construct a composite tolerance variable by averaging all items. The corruption tolerance variable indicates that approximately 25% of respondents see corruption as either purely legitimate or something that is wrong but understandable. Similar to the corruption perception variable, corruption tolerance exhibits substantial cross-country and within-country variation.

Independent Variables

Next, we operationalize our insider–outsider framework. To start, we attempt to empirically capture the theoretical concept of patronage insider. As AB does not directly ask respondents if they are beneficiaries of patronage, we rely on a proxy measure inquiring if a respondent is a member of a patronage network. Specifically, AB asks respondents how they would resolve various kinds of bureaucratic red tape and governmental harassment if they were waiting for a delayed government permit, had a family member wrongly arrested, or had family land wrongly seized. Respondents can answer “don't worry, things will be resolved given enough time,” “lodge a complaint through proper channels or procedures,” “use connections with influential people,” “offer tip or bribe,” “join in public protest,” “other,” or simply “nothing.” This set of questions provides unique empirical leverage to identify potential members of a patronage network as only people with connections to influential people can actually use them. Namely, those individuals who respond “use connections with influential people” to solve problems likely belong to a patronage network and have “strings to pull.” However, instead of stretching the claim that our measure effectively captures the concept of patronage insider, we opt for a more humble position and refer to our measure as network insiders, which takes the value of one if a respondent chooses to use connections with influential people to solve any of the above problems. We explicitly acknowledge that our measure of network insider might not directly tap into the nexus of patronage insider; however, we contend that our network insider measure serves as a next best solution for operationalizing this difficult concept.

In particular, there are several reasons why our measure of network insider should overlap with the concept of patronage insider. First, a majority of studies identify “problem solving” (Auyero 2000; Koter 2013), “request fulfilling” (Stokes et al. 2013), or “political fixing” (Berenschot 2011) as defining features of clientelism, especially in developing countries. In this form of politics, it is common practice that citizens rely on influential local brokers to solve day-to-day problems and then electorally reciprocate these services. Szwarcberg (2012, 231), for instance, defines “clientelism [as] a problem solving network where brokers solve voter problems by providing material and nonmaterial resources in exchange of political support.” Similarly, Stokes et al. (2013, 106) underscore the importance of problem fixing in Argentina where one-third of survey respondents say they would rely on an influential person if they were “facing a grave family problem related to a job or health,” while only 8% of the respondents say that they have received direct handouts from the brokers.

In Africa, influential people—for example, religious leaders, traditional chiefs, local politicians—often serve as intermediaries and brokers between the president and citizens in clientelistic networks (Koter 2013). Through their affiliation with the incumbent, brokers influence the distribution of state resources within their community. Given weak state capacity in many African countries, citizens depend on these brokers to provide material benefits and government services. Citizens primarily turn to brokers as problem solvers because they possess political or social authority unavailable to ordinary citizens.

Despite the influence of political brokers in their communities, not all citizens have access to them. Due to the scarcity of resources and the high premium placed on loyal supporters, brokers monitor citizens extensively to ensure that only those who have cultivated and maintained strong relationships with them (i.e., insiders) are able to get special treatment. To state it differently, problem solving is not a universal good; instead, it is a privileged service only available to network insiders, especially those insiders who, by their brokers’ own calculations, are likely to reciprocate these benefits by electorally supporting the incumbent and her party. As Herbert Kitschelt and Steven Wilkinson (2007, 10) put it, “it is the contingency of targeted benefits, not the targeting of goods taken by itself, that constitutes the clientelistic exchange.”

In short, given the prevalence of problem-solving politics in African countries, and the fact that political brokers selectively provide problem-fixing services to their core constituents, we are optimistic in our network insider measure's ability to capture the complexity of patronage insiders. Our logic is also consistent with Brusco, Nazareno, and Stokes’ (2004) attempt to identify the strength of patron–client networks by asking respondents in Argentina whether they have turned to “important persons” for help in the past year. They show that more than one-third of their respondents would turn to an important local patron for help if the head of her household lost their jobs. They also report that politicians’ clientelistic efforts are more effective in smaller communities.

Of course, there are weaknesses with this indicator; it is indirect and reflective of individuals’ attitudes to hypothetical situations. Nevertheless, we believe that our network insider measure improves on extant literature that has yet to develop a valid operationalization of membership in a clientelistic network. Prior approaches attempt to narrow the concept of clientelism to vote buying (Brusco, Nazareno, and Stokes 2004) or attitudes toward clientelism (Bratton 2007). However, these approaches either miss the fact that (1) politicians pursue multiple distribution strategies, including patronage to win support from a broader array of citizens than just core backers (Gans-Morse, Mazzuca, and Nichter 2014; Stokes et al. 2013); (2) clientelism is more than vote buying (Vicente 2014); (3) social desirability bias is present in answering direct questions on vote buying; or (4) attitudinal aspects do not equal factual aspects of clientelism. Contrastingly, our measure draws on citizens’ responses to a comprehensive range of plausible situations, and thereby has greater validity.

Another limitation of our proxy measure is the gap in levels of analysis between network insider (measuring clientelism at the local level) and our dependent variables (measuring corruption at the national level). One possible drawback to this incongruence is that supporters of opposition parties might claim national politics to be corrupt while also reporting that they rely on connections with influential local brokers to solve problems.5 While we fully acknowledge this measurement issue, for the majority of countries within our sample the incumbent ruling party actually has a monopoly and clear advantage over clientelistic distribution (Koter 2013; van de Walle 2007). Specifically, what Kitschelt and Wilkinson (2007) refer to as “monopolistic” or “unilateral” clientelism emerges in many African counties either because of dominant party systems or the underdeveloped nature of economies that make it difficult for rivaling elites to generate alternative patronage sources. Consequently, citizens who credibly reach out to influential people are likely to be incorporated within the patronage network of the incumbent. In fact, as we shall demonstrate below, insiders are indeed much more likely to support the incumbent than outsiders (see Figure 1 below). In addition, we find that our network insider measure and the attitudinal measure are strongly correlated—network insiders tend to express greater loyalty to their patron and form more favorable attitudes about clientelism.

The construction of the identity insiders variables is more straightforward. We identify co-partisans of the incumbent by constructing a dummy variable that takes on the value of one if a respondent feels close to the incumbent party. Similarly, we use a question from AB that asks respondents’ ethnic affiliation, and we consider a respondent a co-ethnic of the incumbent if she is from the same ethnic group as her president.

We also control for various factors that might shape individuals’ attitudes toward corruption. First, political sophistication, or the quantity and organization of a person's political cognition, affects virtually every dimension of public opinion (Zaller 1992). To a certain extent, theories of public opinion and political behavior are premised on voters’ mental capacity for understanding political issues. In turn, we expect sophisticated voters to be more aware of the existence of corruption and less likely to tolerate it as sophisticated voters are more sensitive to issue information and more likely to act on it (Dalton 2001). We operationalize political sophistication using an additive scale constructed from counts of correct answers to several political knowledge items. The variable is then rescaled from zero to one, where one indicates a respondent correctly answered all questions.

Studies suggest that citizens’ corruption attitudes are also influenced by media exposure (Davis, Camp, and Coleman 2004), although the literature is inconclusive on whether information access increases citizens’ awareness of malfeasance. It is possible that greater exposure to information increases voters’ ability to learn of corruption allegations or prompts voters to seek out media sources that confirm their existing partiality. We remain theoretically agnostic, but control for media exposure by examining how often respondents get news from the radio.

We also control for respondents’ experience with political corruption, as recent studies argue that citizens often use these experiences to gauge the degree of corruption (Treisman 2007). We construct an index of corruption experience that takes on the value of one if a respondent of AB has ever been asked to pay a bribe to receive basic government services. In addition, we also control for citizens’ assessment of their living conditions compared with others, as those individuals who feel socio-economically deprived might be more doubtful about the integrity of political elites and more resentful of corruption. To further control for unobserved attributes among individuals, we also include the Lived Poverty Index6 for citizens, and respondents’ education level and gender. Finally, we include country fixed effects to pick up any country-specific influence not accounted for in our model specification.

Empirical Results

Does the insider–outsider framework help explain how citizens formulate their perceptions of corruption (Hypotheses 2 and 3)? The results of the baseline regression model in Model 1 (Table 1) provide an affirmative answer. We find that those who share the same partisan or ethnic bond with the president consistently perceive lower levels of political corruption. Also squaring with our theory, network insiders perceive higher levels of graft. We interpret this finding to signify that network insiders develop a more accurate understanding of malfeasance through frequent interactions with their patrons. Because of this direct contact, network insiders are privy to information on corrupt practices.

Table 1. Determinants of Corruption Attitudes Formation
Model Dependent VariableModel 1PerceptionModel 2PerceptionModel 3PerceptionModel 4Tolerance
  1. Note: Afrobarometer lacks information on ethnicity in Cape Verde, Tanzania, and Zimbabwe. Coefficient estimates for country dummies are not reported. Robust standard errors in parentheses. *p < .1; **p < .05; ***p < .01. Two-tailed test.

Network insider.034**.027*.025*.029***
Identity insider  −.227***−.005
Information exposure .062***.049***−.021***
Political sophistication .065***.036*−.036***
Corruption experience .409***.442***.047***
Personal economic evaluation −.047***−.047***.011***
Lived Poverty Index .134***.139***.004
Education level .097***.103***−.008*
Female −.006−.009.012***

Model 2, which incorporates all of the control variables, still offers strong support for our theory. An individual's media exposure, political sophistication, direct experience with corruption, negative personal economic evaluations, higher levels of education, and poverty are all associated with a greater awareness of corruption. Importantly, all of our insider variables remain significant, reinforcing the robustness of the previous results.

Notice that Model 1 and Model 2 disaggregate identity insiders into co-partisans and co-ethnics with the president, and astute readers might question the extent to which these two models effectively test Hypotheses 3 and 4 regarding the effect of identity insiders. To alleviate this concern and to further minimize the potential gap between the theoretical hypotheses and the empirical design, we construct a new variable, identity insider, if a citizen shares either a partisan or ethnic attachment with the incumbent.7 Model 3 replaces the co-partisans and co-ethnics variables with the identity insider variable, and the results clearly demonstrate that identity insiders are less likely to perceive their incumbent as corrupt.

We repeated the modeling setup used in Model 3, in the interest of parsimony, to examine the determinants of tolerance for corruption in Model 4. The results in Model 4 show that network insiders, as previously speculated (Hypothesis 1), exhibit significantly higher tolerance for corruption. Essentially, individuals who benefit from their association with the incumbent are likely to be more accepting of corruption. A second and complementary, value-based interpretation is that network insiders are more likely to tolerate corruption simply because they attach similar values to being a network insider as they do to engaging in nominally corrupt activities, such as nepotism, favoritism, and bribes. Simply put, they consider clientelism and certain nominally corrupt activities as “politics as usual.”8

Contrary to our expectations in Hypothesis 4, the results in Model 4 reveal that citizens’ identity insider status has no recognizable effect on their patience for corruption. In other words, we find that identity insiders are no more tolerant of corruption. We also find that individuals with greater political sophistication, exposure to media, and negative personal economic evaluations tend to resent corruption. In addition, male respondents and those with higher education are less tolerant of corruption. One surprising finding is that victims of corruption actually tend to tolerate corruption. While we cannot surmise any reasonable explanation for this, we suspect that when bribes open the door to services previously foreclosed to individuals, they are more willing to see corruption as tolerable.

Why Do Voters Reelect Corrupt Incumbents?

Altogether, we find corroborating evidence that, indeed, individuals’ positions vis-à-vis the political power structures within a country shape their attitudes toward corruption. The results show that network insiders perceive higher levels of and are more tolerant of corruption compared with citizens outside of the clientelistic network. Meanwhile, identity insiders are less aware of graft than identity outsiders.

Remarkably, these findings provide the theoretical leverage necessary to disentangle the paradoxical coexistence of public disdain for corruption and popular corrupt politicians. Conventional wisdom attributes the electoral success of corrupt politicians to two informational dilemmas faced by the electorate: information deficiencies and coordination problems induced by information asymmetry. Proponents of the deficiency explanation argue that voters support corrupt politicians simply because they are poorly informed and unable to accurately observe politicians’ true type. This paucity of information account implies that voters will punish corruption once they are empowered with adequate information (Ferraz and Finan 2008; Olken 2009; Winters and Weitz-Shapiro 2013). Conversely, the coordination-problem perspective argues that as others’ intentions are unknown, a rational voter may reluctantly support the corrupt politician out of fear that she might be defecting alone and consequently face retaliation from the reigning incumbent (Kurer 2001). Overall, the informational approach assumes that voters genuinely dislike corruption, but that they ultimately support politicians “on the take” because voters are kept in the dark or trapped in coordination problems.

A more recent wave of research treats supporting corruption as a rational decision by well-informed voters who are willing to trade off their aversion to corruption for material benefits. For example, Luigi Manzetti and Carole Wilson (2007) contend that corrupt politicians in weakly institutionalized systems are able to remain electorally viable because of their capacity to exploit resources for patrimonial purposes. Similarly, Iannis Konstantinidis and Georgios Xezonakis (2013) show that Greek citizens are more forgiving of politicians charged with corruption who had a record of improving local economic conditions.

We extend these bodies of scholarship and apply the insider–outsider framework to elucidate the corruption paradox. We argue that identity insiders’ judgment, whether co-partisans or co-ethnics, of the incumbent is inevitably clouded regardless of the relevant information available. This in-group bias substantially attenuates the negative corruption–vote link. Moreover, considering the rational corruption perspective, we charge that network insiders may be willing to cut corrupt politicians slack so long as they benefit from the spoils. Ultimately, insiders’ electoral support for the incumbent, when compared with outsiders, is less affected by corruption. Importantly, we do not argue that insiders reward corruption; instead, our hypothesis is that corruption matters less for insiders’ decision to back the incumbent.

We explicitly test this hypothesis with AB data. The dependent variable examines voters’ electoral support for the incumbent, and takes on the value of one if a respondent indicates she would vote for the incumbent if an election were held tomorrow. The key independent variables are personal experience with corruption, insider status, and the interaction of the two. Specifically, we examine the effect of respondents’ experience with corruption, as we are concerned with the possibility that citizens’ corruption perceptions might be endogenously determined by their electoral support. Indeed, while respondents’ electoral support might plausibly shape their perception of corruption toward the incumbent, electoral support should have no bearing on respondents’ direct experience with corruption. We expect the coefficient for corruption experience to be negative. Additionally, to examine whether the insider–outsider status lessens the corruption–vote link, we construct a composite index for insider status that captures the extent to which a voter shares both instrumental and identity bonds with the incumbent. Finally, we create an interaction term between our measure of corruption experience and the composite insider status variable, and control for citizens’ Lived Poverty Index, respondents’ evaluation of the present economic conditions, and the extent to which respondents tolerate corruption. If our expectation that insiders weigh corruption less than outsiders in casting a ballot for the incumbent holds, the coefficient of the interaction term should be positive.

The results from our logistic regression show just that.9 Figure 1 illustrates that insiders are less inclined to punish corrupt incumbents at the polls. Insiders, either due to their affective identity or material considerations, are much more supportive of the incumbent compared with outsiders. Most importantly, when insiders experience corruption, their support for the incumbent does not decline, as the slope for insiders is virtually flat and statistically insignificant. Juxtapose this slope with the sharp downward trend seen for outsiders. This suggests that outsiders’ personal experience with corruption precipitously reduces their electoral support for the incumbent. Put another way, while corruption decreases outsiders’ support for the incumbent, graft has virtually no consequences for insiders’ electoral decision making. Therefore, unscrupulous politicians could remain in office so long as they are able to mobilize support from a sufficient proportion of insiders.

Figure 1.

Support for Corrupt Politicians

Note: Figure 1 shows that corruption matters more electorally for outsiders than for insiders.

To further compare the explanatory power of our insider–outsider theory with the conventional information theory, we include the media exposure variable and its interaction term with the corruption variable. As the results in Table 2 show, information exposure has no conditioning effect on the relationship between voters’ experience with corruption and their support for the incumbent, while the conditioning effect of our insider status variable remains significant.

Table 2. Insiders, Outsiders, and Corruption Voting
Model Dependent VariableModel 5 Support for the IncumbentModel 6 Support for the Incumbent
  1. Note: Coefficient estimates for country dummies are not reported. Robust standard errors in parentheses. *p < .1; **p < .05; ***p < .01. Two-tailed test.

Corruption experience−.900***−.955***
Insider status2.852***2.857***
Corruption experience ×Insider status.513***.508***
Evaluation of present economy.214***.217***
Information exposure −.173***
Corruption experience ×Information exposure .086
Lived Poverty Index.160***.138**
Corruption tolerance.118*.108*

Overall, these results provide a compelling rationale for the long-term electoral viability of corrupt incumbents. We suggest that citizens’ likelihood of acting on information depends critically on the nature of their affiliation with the incumbent. And our findings complement the rational decision thesis and offer a more plausible explanation for the paradox of why citizens, who are sufficiently informed about corruption, may not necessarily sanction incumbents. Furthermore, our results emphasize the importance of tolerance for corruption. While those individuals who directly suffer from corruption are less supportive of the incumbent, those individuals willing to tolerate venality are more likely to back the incumbent.


This article makes an important theoretical distinction between perceptions of and tolerance for corruption, and specifically argues that these two different attitudes toward corruption vary across individuals depending on whether they are political insiders or outsiders. We find that individuals who are members of patronage networks perceive higher levels of and are more tolerant of corruption than outsiders. Moreover, those individuals bound to the incumbent by either shared ethnicity or partisan ties are less likely to consider corruption widespread. In addition, our insider–outsider framework provides useful insight into the puzzle of why voters disdain corruption yet continue to elect corrupt politicians. The takeaway of this article is that citizens’ status as an insider or outsider has significant implications for how they form attitudes toward corruption and how they behave electorally.

One might raise a legitimate concern about the possibility of reverse causality, that is, whether citizens with higher perceptions of corruption may seek to join a patronage relationship to gain connections to influential people, rather than resorting to legal channels to solve problems. While this alternative interpretation of the empirical association between patronage insider and corruption perceptions is plausible, we remain confident that the effect of insider status is exogenous to corruption attitudes. First, we contend that while relying on influential people may be a rational strategy for citizens with elevated perceptions of corruption, actually doing so may be quite difficult as problem-solving is a privileged service granted at the discretion of the influential person. As we discussed above, problem solving is often restricted to citizens who are already locked in to a long-term, reciprocal relationship between themselves and the influential person. Therefore, we believe that citizens with high corruption perceptions may see reaching out to an influential person as a less viable strategy if he or she has not already established an enduring relationship with an influential person. In fact, across the AB sample, only 9% of respondents who perceive high levels of corruption would turn to influential people to overcome extended delays in obtaining a government permit, while 42% would lodge a complaint.

Another limitation of this article relates to our operationalization of patronage insider. Being aware of the inherent problems that scholars have traditionally encountered in measuring the concept of clientelism, we approach it from a different angle by considering the extent to which voters rely on connections with influential people to overcome bureaucratic red tape and governmental harassment. While our network insider indicator is not perfect, we believe that we have provided a novel operationalization that improves on dominant approaches in extant literature.

Measurement issues could also undermine our empirical results regarding the effects of identity insiders on tolerance for corruption. As elaborated earlier, both co-partisan and co-ethnic variables are constructed based on whether respondents share the same partisan or ethnic bond with the president. However, we created the variable of corruption tolerance with reference to general government officials rather than to the president specifically. Ideally, one would like to have a set of survey questions that tap into citizens’ tolerance for corruption by the president and members of the administration so there is a direct match between the response and explanatory variables. Unfortunately, no such data are available.

Our final concern has to do with the generalizability of our theory. To a great extent, we believe that the results based on sub-Saharan Africa are applicable to other developing democracies where ethnic identities and partisan cleavages are salient and clientelism is the most dominant linkage between citizens and elites. Nevertheless, our insider–outsider framework may not extend to established democracies where formal institutions outweigh informal ones and where the informal practices of clientelism have been replaced with programmatic redistributive politics.


We are grateful to Mike Bratton, Mike Colaresi, Miriam Golden, Eunyoung Ha, Philip Keefer, and Daniela Stockmann for their helpful comments on earlier drafts of this paper and to Marty Jordan for indispensable research assistance. Earlier versions of this article were presented at the Workshop on Conflict, Michigan State University, 2009, the 2009 annual meetings of the American Political Science Association, Philadelphia, and the Workshop on Accountability and Corruption, University of Michigan, 2013. The authors are grateful to helpful comments and suggestions given by the participants of these conferences. Any remaining errors are the responsibility of the authors. This research was supported by the National Research Foundation of Korea (2014S1A3A2044032).


  1. 1

    Results are not reported here due to space limitations, but are available on request.

  2. 2

    In the previous example of support for democracy, we find that once we control for citizens’ insider–outsider status, the effect of corruption perception is no longer significant.

  3. 3

    Although Afrobarometer has multiple rounds of data, we chose to utilize Round 3 because it is the only round that contains relevant survey items to operationalize the concepts of corruption perception, corruption tolerance, and patronage insider.

  4. 4

    The Appendix provides Afrobarometer questions for our key variables.

  5. 5

    We thank one of our reviewers for raising this important and legitimate concern.

  6. 6

    The Lived Poverty Index, developed by AB, measures the multidimensionality of poverty by assessing citizens’ access to basic human needs including cash income, food, medical care, and cooking fuel (Mattes 2008). Several studies regard the Lived Poverty Index as a more preferable proxy for individuals’ financial well-being in less economically developed contexts (and subsistence-based economies) where a significant proportion of the population does not receive cash-based wages, and nonmonetary income and household assets are difficult to assess and compare (Bratton 2008; MacLean 2010).

  7. 7

    In other words, we consider a respondent an identity insider if she is a partisan or ethnic insider.

  8. 8

    We thank one of our reviewers for pointing out this alternative explanation.

  9. 9

    See Table 2 for numerical results. To facilitate the interpretation, we recode our insider variable so that observations with values higher than the mean are considered as insiders.

Appendix: Operationalization of the Key Variables

VariableDescriptionExact Wording from the Afrobarometer
Corruption perceptionComposite variable that takes the average of citizens’ perception of corruption involving the president, his executive members, and national government officials.How many of the following people do you think are involved in corruption, or haven't you heard enough about them to say: (1) The President and Officials in his Office? (2) National government officials?
Corruption toleranceComposite variable that takes the average of various items that ask whether respondents find hypothetical scenarios involving abuse of public office for personal enrichment “nothing wrong at all” or “wrong but understandable.”For each of the following please indicate whether you think the act is not wrong at all, wrong but understandable, or wrong and punishable. (1) government official gives a job to someone from his family who does not have adequate qualifications, (2) government official demands a favor or an additional payment for some service that is part of his job, (3) public official decides to locate a development project in an area where his friends and supporters lived.
Support for the incumbentRespondents’ electoral support for the incumbent president.If a presidential election were held tomorrow, which party's candidate would you vote for?
Network insiderCoded 1 if a respondent chooses “use connections with influential people” to solve any kind of bureaucratic red tape and governmental harassment.What, if anything, would you do to try and resolve each of the following situations? (1) You were waiting for a government permit or license, but kept encountering delays. (2) The police wrongly arrested someone in your family. (3) Someone wrongly seized your family's land.
Co-partisansCoded 1 if a respondent feels close to the incumbent party.Do you feel close to any particular political party? Which party is that?
Co-ethnicsCoded 1 if a respondent is from the same ethnic group as the president of her country.What is your tribe? You know, your ethnic or cultural group.
Political sophisticationAn additive scale constructed from counts of correct answers to several political knowledge items.Can you tell me the name of: (1) Your Member of Parliament? (2) Your DCE? (3) The Vice President?
Information exposureHow often respondents get news from the radio?How often do you get news from the following sources: Radio?
Corruption experienceCoded 1 if a respondent has had to pay a bribe to receive basic government services in the past year.In the past year, how often (if ever) have you had to pay a bribe, give a gift, or do a favor to government officials to: (1) Get a document or a permit? (2) Get a child into school? (3) Get a household service? (4) Get medicine or medical attention? (5) Avoid a problem with the police?
Personal economic evaluationRespondents’ assessment of their relative living conditions compared with others.In general, how do you rate: Your living conditions compared with those of other countrymen?
Evaluation of present economyRespondents’ assessment of the present economic conditions.In general, how would you describe: The present economic conditions of this country?
Lived Poverty IndexComposite variable that takes average of various items capturing citizens’ access to basic human needs.Over the past year, how often, if ever, have you or your family gone without: (1) Enough food to eat? (2) Enough clean water for home use? (3) Medicines or medical treatment? (4) Enough fuel to cook your food? (5) A cash income?
Education levelCoded 1 if a respondent has at least attended some secondary/high school.What is the highest level of education you have completed?
FemaleCoded 1 if the respondent is a female.