Introduction

Since the early 1990s, when the third wave of democratisation swept across developing countries, more than 500 national elections have been conducted across the African continent (Bleck & Van De Walle, 2018). Unfortunately, while some of these elections have been certified as free and fair and held in countries that respect the rule of law, still in many parts of Africa, elections continue to be plagued by corruption scandals, vote-buying and widespread irregularities that compromise their integrity (Bleck & Van De Walle, 2018; Bratton & van de Walle, 1997; Elklit & Svensson, 1997). For example, in Kenya and Malawi's 2017 and 2020 presidential elections, both countries' supreme courts annulled the elections and ordered a rerun after finding that widespread irregularities and fraud compromised their integrity (BBC, 2017, 2020). In Nigeria, during the presidential election of 2019, videos emerged showing politicians and parties sharing food and valuable items with voters in states such as Ekiti, Anambra and Edo (ACCORD, 2018Footnote 1; Amaechi & Stockemer, 2022).

Although African politics are not systematically different from other regions (Chabal & Daloz, 1999; Lemke, 2003), we still contend that African elections differ from other regions in observed levels, notably corruption and clientelism. Scholars have long suggested that corruption and clientelism negatively affect economic developmentFootnote 2 and individual well-being (Bentzen, 2012; Holmberg & Rothstein, 2011; Charron et al., 2014; Podobnik et al., 2008). Relatedly, Warren (2004) and Hicken (2011) contend corruption and clientelism can subvert democracy and democratic principles – undermining the ability of ordinary citizens to elect and hold government officials accountable. Corruption and clientelism are some of the worst problems in many parts of the world, particularly in Africa (Transparency International, 2019; Lodge, 2019). Therefore, it warrants asking how clientelism experiences and the general perception of corruption affect an individual-level voter turnout decision in Africa.

A few studies have examined the impact of clientelism and/or corruption on African political behaviour in single-country studies (Bratton, 2008; Inman & Andrews, 2009; Kramon, 2017; Lindberg & Morrison, 2008; Wantchekon, 2003). Still, the topic has received little cross-national attention to the best of our knowledge. In this article, we ask whether the effect of corruption and clientelism vary significantly across the continent, and if so, what explains such cross-national variations? Thus, our article extends previous research by exploring the conditional effect of corruption and clientelism on voting decisions across 34 African countries. We argue that the presence of corruption and clientelism might either mobilise or suppress turnout but that the effect is conditional upon the country's corruption and clientelism context. Overall, by investigating how voters’ perception of corruption and clientelism affects individual voter turnout and whether this presumed relationship differs depending on the context of corruption and clientelism in the country of residence, we offer a better understanding of the mechanisms that might suppress or mobilise voters on election day.

Corruption, clientelism and voter turnout in Africa

Voting is considered a key indicator of a functioning democracy. It allows citizens to articulate their demands and select political leaders while holding them accountable (Katz, 1997; Lanning, 2008). Based initially on studies in Western democracies, the explanation of individual voter turnout has emphasised at least two groups of factors: individual and institutional explanations.Footnote 3

First, scholars distinguish between socio-demographic and mobilisation models among the individual-level explanations. Looking at the socio-demographic determinants of vote choice, most studies focus on age, gender, and socioeconomic status. Beginning with age and drawing on life-cycle effects,Footnote 4 young people often turn out in lower numbers than older citizens. Next, concerning gender, it is essential to note that in established and some newer democracies of Western Europe, the US, Latin America, and East Asia, recent studies indicate that the gender gap seems to have reduced or even reversed, with women consistently being more likely to vote than men (Carreras, 2018; Kostelka et al., 2019; Liu, 2022). Also, income and education are positively associated with the turnout at the individual level, with those with higher socioeconomic status voting at higher rates (Verba et al., 1995). Second, studies have also examined the impact of mobilising agencies (political parties, neighbourhood organisations, labour unions) on turnout at the individual level. The mobilisation model focuses on how a voter's membership in social or informal networks influences an individual decision to vote and whom to vote for (Rosenstone & Hansen, 1993, p. 23). More importantly, evidence from sub-Saharan Africa suggests that membership in voluntary associations is correlated with the propensity to vote (Kuenzi & Lambright, 2011; Tambe, 2017).

Second, aside from individual-level explanations, studies have also emphasised the importance of institutional factors in predicting turnout focusing on factors ranging from the electoral system, concurrent elections, and competitiveness of elections (Stockemer, 2017; Cancela & Geys, 2016; Frank & Martínez i Coma, 2021). First, regarding the electoral system, studies suggest that proportional electoral systems are most likely to facilitate turnout compared to majoritarian or plurality systems (Blais & Aarts, 2006). According to Blais and Carty (1990), the proportional system (PR) reduces distortion in converting ballots into seats. Second, concerning concurrent elections, Cox and Munger (1989) argue that the possibility of many elections taking place at the same time will arouse more media coverage in at least one of the polls, which invariably boosts the amount of money spent on campaigning and, as such, increases the amount of information and public awareness about the issues at stake in these elections. Lastly, rounding up with competitive elections, the general expectation is that the more competitive an election appears, the more likely people will turn out on election day (Blais, 2000).

The preceding explanations significantly improved our knowledge about electoral politics and what motivates people to turn out on election day. However, with the newness of elections in AfricaFootnote 5 and the continent's distinct political and cultural context, notably weak institutions, the legacy of authoritarianism, and ethnic and cultural diversity, it may be that such explanations do not give a complete picture of individual-level turnout dynamics in the continent. Moreover, as we mentioned above, while we do not adhere that African politics are different from other geo-political regions, still, we contend that African elections are different from other regions in observed levels (i.e., corruption and clientelism). For this reason, we argue scholars working on African cases ought to include both these dominant explanations of turnout while also paying attention to factors likely more specific to political dynamics on the continent. Against this background, this article focuses on the conditional effects of corruption and clientelism on individual voting decisions across 34 African countries. The following section examines the theoretical mechanism linking these variables to individual-vote choice.

Corruption and voter turnout

Corruption is one of the slipperiest concepts in the social sciences. Nevertheless, in recent years, scholars and practitioners seem to have settled on one definition, derived from the work of Joseph Nye and popularised by Transparency International (TI), according to which corruption is the abuse of entrusted power for private gain. Still, as many scholars (Heywood, 2007; Hough, 2017; Johnston, 1999a, b; Rose & Peiffer, 2019) have pointed out, this standard definition of corruption is far from perfect, as most of its key concepts, such as 'abuse', 'public', 'private', "and even 'benefit' can be matters of considerable dispute" (Johnston, 1999a, b, p. 6).

However, to the extent that corruption is more prevalent in Africa than in any other region,Footnote 6 one wonders about its impact on political behaviour, including individual-level voter turnout. Research into the link between corruption and turnout falls into two main streams and varies in terms of the level of analysis. The first stream that focuses on individual-level analysis suggests that high corruption has a mobilising impact, generating incentives for citizens to cast corrupt leaders or officials from office. For instance, Bågenholm's (2013) study of European elections between 1981 and 2011 found that parties campaigning on an anti-corruption ticket in corrupt countries attracted more votes than those focused on other issues. Related to this, Inman and Andrews (2009) drew on individual-country studies and revealed that the perception of government corruption increases voting among Senegalese voters.

The second stream of research, drawn notably from macro-level studies, contends that corruption undermines voters' trust in the democratic process, weakening their desire to participate. For example, Kostadinova (2009) shows that corruption creates distrust and cynicism towards the incumbent and democracy, eroding willingness to cast a ballot. To substantiate this, Stockemer et al. (2013) performed analyses of turnout in 72 countries, including established and developing countries, and found that countries with better control of corruption tend to display higher turnout levels. Drawing from these two schools of thought, we examine how corruption affects electoral participation and ask whether the perception that corruption is widespread inspires or depresses turnout in the continent. Given the pervasiveness of corruption in many African countries and a more recent study that focuses on the effect of corruption on individual-level voter turnout confirms that corruption decreases turnout (Dahlberg & Solevid, 2016), we argue that the perception of corruption generally suppresses voter turnout. As such, we expect the following hypothesis to receive empirical backing:

  • H1a: An increase in perceptions of corruption decreases voter turnout.

However, there is reason to believe that the perception of corruption has a conditional impact on voter turnout depending on the institutional context (Birch, 2010; Dahlberg & Solevid, 2016). For example, in highly corrupt settings, an increase in the perception of corruption could dampen an activist spirit and bolster a sense that civic engagement will not be sufficient to change the status quo. In such settings, an individual who feels that the state and government institutions are steeped in corruption will have a lower probability of voting than an individual with the same corruption perception level but who happens to live in a low corruption setting. In fact, in countries where corruption is relatively low, an increase in the perception of corruption can encourage ordinary people to vote and 'kick the rascals out' (Johnston, 2005, p. 42). Based on this discussion, we expect the following hypothesis to receive empirical backing:

  • H1b: The negative impact of the perceptions of corruption will be more substantial in countries with high levels of corruption.

Clientelism and voter turnout

Africa is often described as a continent in which politics is structured by clientelism (Lemarchand, 1972; van de Walle, 2007), defined as the "instrumental friendship in which an individual of higher socioeconomic status (patron) uses his influence and resources to provide protection or benefits, or both, for a person of lower status (client) who, for his part, reciprocates by offering general support and assistance, including personal services, to the patron" (Scott, 1972: 92). Clientelism and corruption are closely related concepts and, indeed, mutually reinforcing phenomena, to the extent that they both "involve political actors manipulating public resources for personal gain (be it financial or political) (Singer, 2009, 2)". For this and other reasons (e.g., data availability), scholars tend to use measures of one concept as a proxy for the other (Keefer, 2007).

However, it will be a mistake to conceive clientelism as synonymous with corruption. Singer (2009, 3) rightly notes that "many forms of corruption have nothing to do with clientelism, such as when a public official steals money or extracts bribes to fund their own consumption". Moreover, the ongoing nature of the relationship between the client and patron distinguishes clientelism from corruption. In the words of Hicken (2011), ‘‘corruption is hopefully one-off interaction, with neither party having a strong expectation of interacting in the future. In contrast, clientelism is at its core an iterated interaction, with each side anticipating future interactions as they make decisions about their behaviour today’’.

Both phenomena can have sharply divergent effects on political behaviour at the individual level. In many developing countries, many forms of clientelism are open and, from voters' viewpoint, accepted forms of service performed by politiciansFootnote 7 for their constituents (Kitschelt & Altamirano, 2015). Trantidis and Tsagkroni (2017) demonstrate, using the case of Greece, that politicians can use clientelism and corruption as two distinct strategies of state capture when public resources are severely limited. It may be that ordinary Africans can make a cognitive distinction between clientelistic practices and corrupt conduct in ways that influence how they behave politically. Drawing from experimental data, Kramon (2017) and Wantchekon (2003) suggest that vote-buying positively impacts turnout. This leads us to expect the following hypothesis to receive support:

  • H2a: Voters who received material benefits from political parties or individual politicians are more likely to vote.

While clientelistic exchanges are common across the developing world, particularly in Africa, their impact on the behaviour of recipients appears to be shaped by other individual and country-level forces, particularly poverty (see Wantchekon, 2003). According to Stokes et al. (2013, p. 662), "poor people value a handout more highly than wealthy people; hence, if one is going to hand out goodies, one will target the poor …poor people are [also] risk-averse and hence value more highly a bag of goodies in hand today than the promise of a redistributive public policy tomorrow". This means that poorer individuals are more likely to seek and receive material rewards for their votes and are also more likely to demonstrate that they value gifts by actually keeping their end of the bargain and turning out to vote. Conversely, when citizens are materially well-to-do, disbursement of material rewards is likely to become more questionable as it might be seen to drain resources meant to provide public goods and services (Jensen & Justesen, 2014; Kitschelt & Altamirano, 2015).

Based on the factors mentioned above, we first expect that clientelist practices will be more prevalent in countries where a large proportion of people lack necessities such as food, water, fuel to cook food and medical supplies. Second, we expect that vote-buying experience to affect individuals differently depending on the material conditions of their country, with clients from poorer countries being more likely to vote than clients from wealthier countries (Kitschelt & Kselman, 2013): Based on the above discussion, our hypothesis reads as follows:

  • H2b: The impact of clientelism will be more substantial in poorer countries than in their wealthier counterparts.

Data, measures and estimation procedure

Data

Data for the empirical analysis are drawn from the Afrobarometer survey data. The survey used a 'clustered, stratified, multi-stage probability sampling design to obtain 'national probability samples that represent an accurate cross-section of the voting age population 'in the study countries.Footnote 8 Although the Afrobarometer survey is currently in the eight-round, because of missing data for one of the key explanatory variables (i.e., clientelism), we restrict our analysis to Afrobarometer round 5 data conducted between 2011- 2013 across some 34 African countries (see Table A1 supplementary material for details the number of countries included in the study).

Dependent variable: voter turnout

To measure voter turnout, we used a question that asked whether respondents voted in the most recent national election. Thus, we used a dichotomous measure for the respondents who voted in the most recent national elections: 1 = yes, voted; 0 = no, did not vote. Figure 1 illustrates turnout, as reported by the respondents in the Afrobarometer, compared with the official turnout derived from the International Institute of Democracy and Electoral Assistance (IDEA) Voter Turnout database.Footnote 9 The figure reveals the usual discrepancies between self-reported turnout in surveys and the official turnout rates for some countries ( Dahlberg & Persson, 2014). Still, given the absence of validated turnout measures at the individual level, we proceed with this variable. However, in the analysis section, we include a control variable (survey year) for the year the Afrobarometer survey data was conducted in each country included in our sample to partly handle the problem of over-reporting.Footnote 10

Fig. 1
figure 1

Self-reported turnout compared with Official turnout. Sources: Afrobarometer (https://afrobarometer.org/data) and International Institute for Democracy and Electoral Assistance – IDEA (https://www.idea.int/data-tools/data/voter-turnout) (Note: Namibia and Swaziland are excluded as there were no data for the official voter turnout.)

Full size image

Main independent variables

Turning to the key predictor variables, corruption is measured based on an additive index of similarly worded questions asking respondents to assess the level of corruption in various public institutions. The question reads as follows: How many of the following people do you think are involved in corruption, or haven't you heard enough about them to say: (a) office of the presidency, (b) members of parliament, (c) government officials, (d) local government councillors, (e) police and (f) judges and magistrates. The response options were: 0 = none, 1 = some of them, 2 = most of them, and 3 = all of them. On the other hand, clientelism was measured by a question that captured outright vote-buying as follows: During the last national election in (20xx), how often, if ever, did a candidate or someone from a political party offer you something, like food or a gift or money, in return for your vote? Respondents could answer using four categories: 0 = never (no experience of this in the past year), 1 = once or twice, 2 = a few times, and 3 = often. We coded clientelism to a dummy variable with 1 assigned to voters who have been offered a gift/bribe in return for their votes and 0 to people with no experience (i.e. who have never received gifts).

Control variables

We include several control variables drawn from studies on voting behaviour. These include socio-demographic controls such as age, gender, voter's location (rural and urban residence), education and live poverty/material wealth (a proxy for income). These variables were measured in the following ways: (1) age was kept as a continuous variable, (2) gender was coded as a dummy variable, and (3) place of residence was measured by a dummy with 0 = respondents living in urban areas and 1 = respondents living in rural areas. In addition, 4) education was measured with a standard question which asked respondents about the highest level of education they have completed. In addition, education is measured as a zero–nine index ranging from no education to post-graduate education. Finally, material wealth/poverty is measured by the Afrobarometer live poverty index, which averages an index of the five poverty items,Footnote 11 with scores ranging along a five-point scale, with a lower score (0) reflecting people living in good material conditions while a higher score (4) implies greater deprivation/a constant absence of all basic necessities (Mattes, 2020).

Also, we include several mobilisations and political and psychological controls, particularly party identification, associational membership, and political interest. Party identification is captured by a survey question that asks respondents whether they feel close to any political party. We collapsed the responses into a dummy variable: no, not close to any party, was coded as 0, while yes, feel close to a party, was coded as 1. Associational membership was measured with a question about a voluntary association or community group membership. We recoded this variable into three categories: 0 = not a member, 1 = an inactive member, and 2 = an active member/official leader.Footnote 12 Lastly, political interest is measured by asking respondents about their interest in public affairs.

Finally, we included a couple of contextual and institutional variables to explain why turnout varies by country and why the effect of the key predictors varies cross-nationally. First, GDP per capita is generally considered essential to understanding voter turnout. Still, we do not use this variable as it shows very limited variations in the group of countries we examine (see also Jensen & Justesen, 2014). Instead, we rely on a proxy for national wealth derived from the Afrobarometer data measure of live poverty index. Thus, our national wealth is measured by aggregating the live poverty index by country.Footnote 13 Second, political freedom is measured using the Polity IV score. Third, political corruptionFootnote 14 is derived from the V-Dem dataset, which captures the pervasiveness of corruption in the public, executive, legislative, and judiciary sectors. Again, political corruption is measured on a continuous scale from 0 to 1. The fourth election competitiveness is measured as the difference in vote share of the two largest parties in each election; we use a 10% level as a mark of competitiveness. The variable is then dummy coded (1 = competitive/close election) and (0 = not competitive/not a close election). Finally, we also include a variable that measures the type of electoral system. Electoral system type: information is taken from the Electoral System design database from the International Institute of Democracy and Electoral Assistance. The African electoral system comprises mixed, proportional, and plurality/majoritarian. Therefore, we created a dummy with a proportional and majoritarian system, with one of the other electoral systems being the reference category. The summary statistics of all our variables are displayed in supplementary materials (see Table A2 supplementary materials).

Estimation procedures

We employed a multilevel binary logistic model to estimate the effect of corruption and clientelism on voter turnout. First, this was appropriate given that the dependent variable has only two options: voted or otherwise. Second, our data consists of at least two levels: individuals nested in countries. Third, specifying a multilevel or variance components model enables us to simultaneously examine the effects of country-level predictors and their individual-level counterparts, yielding a potentially richer account of the variability in individual voting decisions in the continent. A common practice in multilevel modelling is partitioning the unexplained variance in the outcome variable according to the different levels of the data (Browne et al., 2005). This helps to determine whether there is enough variability at a higher level to warrant fitting the relatively complex variance components model.

However, calculating the proportion of variance that is due to differences between countries is not as straightforward in non-linear models as it is in their linear counterparts, where a normal error distribution is assumed (see Browne et al., 2005; Snijders & Bosker, 2012; Leyland & Groenewegen, 2020; Goldstein et al., 2002). Furthermore, the problem with modelling discrete responses, such as whether the person has voted, is that the variance partition coefficient (VPC) “must be calculated for specific values of the covariates included in a multilevel regression model” (Leyland & Groenewegen, 2020 p. 90). This means that it is possible for different covariates in the model to produce different estimates for the unexplained variance between levels.

Goldstein et al. (2002) offer four methods of calculating the VPC for discrete response models, the most widely used of which is the threshold model or the latent variable method. This assumes that a true underlying variable is continuous and “that the level 1 variance is fixed and independent of the predictor variables” (Browne et al., 2005, p. 604). As such, the method assumes a standard logistic distribution value of 3.29 at the individual level and calculates the VPC as the variance between countries divided by the sum of that value and the assumed value of 3.29. Thus, to use this method, we first need to run an intercept-only model and obtain the variance estimate, which is 0.333, and then add the fixed level 1 residual variance of 3.29, as shown below:

$$VPC = 0. 323/(0.323+ 3.29) = 0.09$$

These results indicate that the amount of variability in the outcome variable that is attributable to differences between countries is 9%. However, as we have noted, the above formula provides an approximate estimate of the proportion of unexplained variance attributable to the country level. Figure 2 plots each country's residual with its 95% confidence band against its rank and therefore shows a range of differential country effects on voter turnout. As the figure demonstrates, 27 out of 34 countries in the dataset have aggregate voter turnout figures (measured in log-odds) that are significantly different from the Africa-wide likelihood for turnout. Thirteen countries, including Benin, Liberia, Egypt, and Tanzania, have above-average voter turnout probability, while fourteen countries, including Algeria, Cameroon, Zambia, Ivory Coast, and Botswana, have below-average probabilities. Overall, the results indicate that the intercept is significantly different from the average in most countries, which justifies using a multilevel logistic regression model.

Fig. 2
figure 2

Log odds of voting across the African continent

Full size image

Results

In Table 1 (Model 1), we begin by assessing the impact of our two core explanatory variables, notably corruption perception and clientelism, and the individual-level variables known to affect voting. First, the coefficient for the perception of corruption is negative and statistically significant. This result suggests that voters who think corruption is more widespread in their country are less likely to vote. The marginal effect of voting decreases by 7 percentage points as we move from those who perceive widespread corruption (0.76) to those who perceive little or no corruption (0.83). This result confirms previous findings that corruption erodes civic engagement (Monyake, 2018) and, more specifically, voting (Dahlberg & Solevid, 2016). Next, the effect of clientelism is positive but nonsignificant. The difference in the marginal effect of voting increases by 2 percentage points between voters who declared having received a material gift, favour or money (0.81) in exchange for political support compared to voters who report no such experience (0.79). While the difference is not huge, the result still confirms clientelism as strategy parties and politicians can use to mobilise voters on election day.

Table 1 The effect of Corruption and Clientelism and Voter Turnout in Africa
Full size table

Second, moving to the individual-level control variables, the result largely collaborates with previous studies on the determinants of individual-level voter turnout across the African context. Rural voters, older people, having an interest in politics and being a member of a social and political organisation (i.e., political party and associational group) significantly increase the chances of voting (Tambe, 2017; Tambe & Kopacheva, 2023). Except for this, we confirmed previous priors that women and educated people are less likely to vote in national elections (Kuenzi & Lambright, 2011; Tambe & Kopacheva, 2023; Isaksson et al., 2014; Coffe & Bolzendahl, 2011; Erin, 2020).

Models 2–3 focus on the cross-country variability in the impact of our key predictors. There is relatively strong contextual variability in the effect of electoral clientelism. This implies that the variable is sensitive to the national context, having a stronger effect in some countries than they do in others. We can also confirm a relatively smaller but statistically significant country-to-country difference in the relationship between corruption perceptions and the propensity to vote. In Fig. 3, we plot the between-country variance as a function of the perceptions of corruption variable. The graph suggests that (measured and unmeasured) country differences in the probability of voting are largest among individuals with the highest perceptions of corruption and lowest among those with lower perceptions of corruption. Put simply, individuals with strong perceptions of corruption are most likely to display vastly different voter turnout patterns depending on where they live.

Fig. 3
figure 3

Between country variance as a function the perceptions of corruption

Full size image

The corruption perceptions variable has the strongest positive impact in seven of the 34 countries, including South Africa, Botswana, Zambia and Tunisia, and a negative impact in Liberia, Egypt, Niger, Mozambique, and Madagascar, among others (see Fig. 4). This perception appears to be most depressing on turnout in the North African country of Algeria. Indeed, Algeria has a history of low voter turnout, as recently shown by the less than 25% turnout rate in the referendum on constitutional amendments. In the presidential election of 2019, preceded by a series of anti-corruption protests, turnout averaged 33%, as tens of thousands of Algerians boycotted the poll, alleging corruption by the political establishment (Khetab, 2019). Corruption perception is likely to have a smaller but positive effect in Malawi, Cape Verde, Benin, and Guinea, to mention a few.

Fig. 4
figure 4

Intercept and slope residuals for the association between the perception of corruption and the probability of voting

Full size image

In Model 4, the individual-level variables and country-level variables are estimated at the same time. First, the coefficient for corruption perception is still negative and statistically significant. All else being equal, the marginal effect of voting decreases from 0.83 to 0.76 as we move from those who perceive corruption as less rampant to those who perceive corruption as widespread. On the other hand, the effect of electoral clientelism on voting is trivial as the likelihood of voting only increases by 1 percentage point for those who reported receiving material benefit compared to those who reported no such experience. Overall, we can still confirm that the first two hypotheses (i.e., H1a and H2a) increase in perceptions of corruption decreases voter turnout, and receiving material benefits increases turnout. Next, most of the results for the individual-level variables remain unchanged. For instance, older people, men, less educated people, poor and rural voters, those interested in politics and those who identify with social and political group organisations are more likely to vote. In particular, women are 4 percentage points less likely to vote than men. Less educated voters are 5 percentage points more likely to vote compared to those with higher levels of education. Those who declared membership in social group organisations are 6 percentage points more likely to vote. Likewise, those interested in politics are 7 percentage points more likely to vote. Finally, party identification yields the highest voter turnout difference. Indeed, those who identify as party members are 13 per cent more likely to vote than non-party members (see Table 2).

Table 2 Marginal effect for voter turnout at the minimum and maximum values for each independent variable at the individual level
Full size table

Second, regarding the contextual level variables, the electoral competitiveness variable is negative and nonsignificant, suggesting people are less likely to vote in countries where the elections are considered competitive or a closed race. The marginal effect of voting decreases by 5 percentage points as we move from countries where the election was not considered competitive (0.82) to those where the elections were considered competitive (0.77). Also, the electoral system's coefficient is quite surprising when we compare proportional and majoritarian systems. Specifically, people are 8 percentage points less likely to vote in countries with a proportional electoral system. In contrast, in countries with a majoritarian electoral system, voters are 5 percentage points more likely to vote on election day. The measure of political freedom (political score) has a positive and insignificant effect on voting. The difference in the marginal effect of voting increases by 13 percentage points as democratic quality increases. At the same time, the country-level measure of national wealth is negative and insignificant. Finally, the measure of political corruption obtained from the Varieties of Democracy (V-Dem) dataset is positive and statistically significant. An increase in political corruption increases the probability of turning out to vote.

Model 5–6 introduces the cross-level interactions to try and account for the differential impact of each of the three variables (political corruption, clientelism and national wealth). Model 5 shows a statistically significant and negative interaction between the individual-level corruption perception variable and the country-level political corruption variable. Individuals who live in countries that V-DEM experts perceive as highly politically corrupt, such as Kenya, Uganda, or Nigeria, are less likely to vote when they think corruption is widespread in both state and government institutions. In essence, as demonstrated by Fig. 5, a corrupt national context seems to have a more substantial negative effect on turnout among people who already perceive it as being rampant in society. Conversely, as the Figure demonstrates, an individual who lives in a relatively clean country but perceives corruption to be widespread is nearly as likely to vote as a person who thinks otherwise. Indeed, in low corruption settings, the difference in the probability of turning out between individuals with the highest score on corruption perception and those with the lowest is small, albeit statistically significant. The pattern of results shown in Fig. 5 is consistent with the result illustrated in Fig. 4, which indicates that context-level factors substantially impact individuals with high perceptions of corruption more than their low-perception counterparts. Overall, these results are consistent with hypothesis 1b and correspond with the view that an increase in corruption perception may erode civic engagement in high-corruption settings (Peiffer & Alvarez, 2016).

Fig. 5
figure 5

Cross-level interaction between perceptions of corruption and political corruption

Full size image

To test hypothesis 2b, we aggregated the lived poverty index and generated a new country-level poverty variable (i.e., national wealth). We consider this aggregated variable a more accurate measure of country-level material conditions than the more commonly used per capita GDP.Footnote 15 We examine the cross-level interaction between the national wealth variable and clientelism. The results are shown in Model 6, the cross-level interaction between these variables is statistically significant and positive, suggesting that those who received material rewards in exchange for political support are increasingly likely to vote when they live in countries with a large proportion of people lacking basic necessities such as food and water (see Fig. 6). This is in line with hypothesis H2b and the literature on clientelism (Jensen & Justesen, 2014). Quite strikingly, as Fig. 6 demonstrates, the country’s material conditions seem to have no impact on the voting propensity of the group of individuals who have not received any material resources as a quid pro quo for their votes.Footnote 16

Fig. 6
figure 6

Cross-level interaction between Clientelism and country-level poverty

Full size image

Conclusion

In this article, we examine the effects of corruption and clientelism on individual voter turnout decisions using cross-national survey data from the Afrobarometer for some 34 African countries. Our article makes two important contributions to research on voter turnout. First, perceptions of corruption reduce the likelihood of voting, contingent on the voter's country. In fact, we find that an increase in the perception of corruption has a more corrosive effect in countries characterised by high levels of political corruption. Where the institutional environment is not characterised by high levels of political corruption, variations in perceptions of corruption do not affect turnout. Second, we show that clientelism tends to increase voter turnout. But importantly, the analysis suggests that the country’s material conditions shape the impact of personal experience clientelism. Voters who declared having received material benefits, gifts or money in exchange for their vote are more likely to live in poor African countries.

However, we acknowledge that some issues, notably response bias, might have influenced our analysis. First, the spectre of response bias, notably the issue of using a self-reported measure for voter turnout capture in surveys, is an over-reporting concern. Over-reporting may, in turn, bias the relationship between the variables of interest and turnout. In Table A3Footnote 17 (see supplementary material), we include several social desirability bias controls, notably whether the Afrobarometer pollster was perceived as honest or misleading and their perception of the survey sponsors to test the robustness of our results.Footnote 18 For instance, as reported in Table A3, respondents who perceived the survey was organised by a government agency, a political party/politician, and NGO was less likely to have voted than those who perceived the survey sponsor as an Afrobarometer national partner. However, the coefficient was not statistically significant except for those who perceived the government organised the survey. Relatedly, respondents perceived as honest by the survey interviewer were more likely to vote than those perceived as misleading. These results, therefore, bolster our findings' general thrust and confidence.

Overall, our findings and conclusion hold important academic and policy implications. First, this study not only complements previous studies on voter turnout but, more importantly, on a theoretical level; this paper highlights the need to consider factors that have not been adequately studied or incorporated in the literature on voting behaviour in new democracies, particularly those of Africa. The result showcases the need for turnout models to consider corruption and clientelism as relevant variables and, more importantly, how these variables interact with each other and other institutional variables, such as economic conditions and corruption, in explaining political behaviour in Africa.

Second, these results equally have important policy implications. Given the pervasiveness of corruption tend to repel voters from the poll. Still, long-term corruption might also be detrimental to the government's legitimacy and the overall democratic trajectory of the region. In the phase of this visible threat, African governments need not only to combat corruption but also there has a responsibility to effectively inform the general public about progress in uprooting corruption and other malpractices that might undermine citizens from turning out on election day. In addition, given that turnout is higher among those with more experience with electoral clientelism (which is especially strong in poorer countries), this poses a serious problem where citizens may be induced to vote or keep leaders in office for their narrow interests. To ensure citizens only vote for leaders on what they can do or their ideologies, there is an urgent need for governments across the region to dedicate substantial efforts towards creating jobs and reducing poverty.