The Price of Poverty: Inequality and the Strategic Use of Clientelism in Divided Democracies

Abstract

This article investigates the political cost of poverty in democracies marked by deep social divisions. We develop a probabilistic voting model that incorporates clientelism as a strategic tool employed by elite political parties to secure electoral support from non-elite voters. Unlike models based on ideological proximity, our framework conceptualizes party competition as structured by the socioeconomic composition of their constituencies. We demonstrate that in contexts of high inequality and widespread poverty, elite parties face structural incentives to deploy clientelistic strategies rather than universalistic policy agendas. Our model predicts that clientelistic expenditures by elite parties increase proportionally with both inequality (GINI index) and poverty levels, rendering clientelism a rational and cost-effective mechanism of political control. Empirical evidence from a cross-national panel (2013–2019) confirms the theoretical predictions: an increase of the 1 percent in the GINI index increase a 1.3 percent in the clientelism, even after accounting for endogeneity and dynamic effects. These findings suggest that in divided democracies, poverty is not merely a condition to be alleviated, but a political resource that elites strategically exploit. Consequently, clientelism persists not as a cultural residue or institutional failure, but as a rational response to inequality-driven constraints within democratic competition.

1. Introduction

The price of poverty is not limited to the material deprivation or social exclusion experienced by the most vulnerable sectors; it also extends to the ways in which poverty is strategically exploited in the political realm. In contexts of structural inequality, political parties resort to clientelism not merely as a tool for electoral mobilization, but as a means of securing governability in deeply divided democracies. This article explores how conditions of poverty and social fragmentation enable clientelist practices, turning them into a rational and functional instrument for political actors who face institutional constraints and high levels of uncertainty.

Clientelism constitutes a deeply entrenched political phenomenon that not only perpetuates unequal power relations but also exacerbates some of the most critical social problems of our time. Among its most detrimental effects are the progressive weakening of democratic systems, the structural reproduction of inequality, and the consolidation of persistent cycles of poverty. From this perspective, far from being a mere exchange of favors, clientelism undermines the foundational principles of democratic citizenship by conditioning access to rights and opportunities on political alignment or personal loyalty (Guerrero-Sierra et al., 2024).

Moreover, the impact of clientelism transcends the institutional sphere, generating concrete effects on social stability. In regions such as Latin America and Africa—where clientelism has historically operated in contexts of exclusion and weak state presence—it has been identified as a factor contributing to dynamics of violence. This occurs both through competition between clientelist networks and through their entanglement with illegal economies and armed structures. Thus, beyond representing a deviation from democratic ideals, clientelism functions as a mechanism that reproduces inequality, injustice, and conflict in settings marked by institutional fragility (Anderson & Tverdova, 2005; Hicken, 2011).

Accordingly, this article conceptualizes clientelism as a form of political exchange based on the selective provision of private goods, privatized benefits, or individual favors by political parties in return for electoral support. Unlike the principles that guide the universal and equitable provision of public goods under strong institutional arrangements, clientelism is characterized by conditionality, informality, and interpersonal dependence (Kitschelt & Wilkinson, 2007; Stokes, 2005). Through both empirical and theoretical analysis, this study seeks to understand how inequality acts as a catalyst for these practices, revealing the true democratic and social cost of poverty in divided democracies.

Within this framework, this article presents a model of probabilistic voting under clientelism in a democracy divided into two social groups (Thachil, 2014b). On one hand, there is the elite group, composed of citizens in the most privileged and influential socioeconomic positions (see Gibson, 1996; Mosca, 1939; Thachil, 2014a). On the other, there is the non-elite group, primarily comprising individuals situated in the middle and lower socioeconomic strata of society (see Converse, 2006). Each group is represented by a political party that channels its citizens’ demands—an elite party and a non-elite party, respectively. Both parties compete for votes in an electoral process consistent with the liberal democratic tradition (Strom, 1989). As in (Thachil, 2014a), the party typology employed here is not based on a left–right ideological axis but rather on the social composition of their main constituencies. Consequently, each party’s electorate is primarily composed of citizens from the social group it represents.

This is not a minor matter. The elite political party understands that the votes it secures in elections do not depend solely on elite voters. It recognizes the need to attract non-elite voters, as they belong to a social group that the party does not represent. This theoretical approach marks a significant departure from conventional models. An elite party does not obtain votes solely through clientelism, nor exclusively through political platforms. Given the party typology, the model assumes that the elite party has incentives to employ clientelistic practices for a substantive methodological reason: while it controls substantial political and economic resources, it also represents the smallest social group in terms of voter numbers—the elite.

It is important to note that not all elite parties necessarily engage in political clientelism. There are numerous cases in which elite parties refrain from such practices. The theoretical model aims to identify the necessary and sufficient conditions under which an elite party resorts to clientelism. In many countries, both developed and developing, elite parties often pursue both strategies simultaneously: they buy votes while also providing public goods (see Szwarcberg, 2013).

On the other hand, the theoretical model assumes that non-elite parties do not engage in clientelism. However, this assumption does not imply a loss of generality. Empirical evidence indicates that, in contexts of institutional change—where access to power is broadened and non-elite parties gain entry into executive or legislative bodies—electoral support is primarily based on opinion votes rather than clientelistic exchanges.

Moreover, once institutional change has occurred and non-elite parties begin to accumulate significant social and political capital, they cease to be non-elite in the strict analytical sense. They may instead be understood as elite parties representing an emerging elite. The Venezuelan case illustrates this dynamic: the United Socialist Party of Venezuela (PSUV), although initially a non-elite party, no longer represents the interests of the broader non-elite citizenry. Methodologically, within the framework of our model, the current PSUV constitutes an elite party that allocates substantial resources to clientelism in order to secure support from non-elite voters.

In sum, a non-elite party is initially composed of political actors lacking economic and political capital. However, once such a party gains access to power and, over time, amasses substantial resources, it transitions analytically into the category of elite party. Under these new conditions, the model explores the circumstances under which an elite party engages in clientelism and the intensity of such practices.

What is noteworthy is that this theoretical model enables us to explain cases in which non-elite parties achieve electoral victories and drive institutional change, even when confronting resource-rich elite parties. Although the model is, as acknowledged, a simplification of complex political realities, the assumption does not entail a loss of generality.

The results of the model’s characterization are multifaceted.

The elite party anticipates that the non-elite party will adopt a similar political platform in terms of public policies. This is because, during elections, parties tend to converge toward the center to avoid being distinguishable from each other and to maximize their potential electoral support base (Proposition 1: a version of the median voter theorem).

Consequently, if the elite party knows that its electoral base is composed primarily of citizens from the elite social group, it also understands that the median voter from the non-elite group is unlikely to support it based on partisan affinity. The non-elite median voter will prefer the non-elite party unless the elite party undertakes three specific actions: refrains from vote-buying, adheres strictly to the median voter strategy, and focuses its policies mainly on the elite social group. Under these conditions, clientelism is not merely optional for the elite party—it becomes a strategic necessity. If the party limits itself to offering public goods, it is likely to lose the election, given that the elite social group represents a demographic minority (Propositions 2 and 3).

Therefore, the elite party opts to offer monetary payments to the non-elite median voter in order to override their party preferences and secure their vote, thereby increasing its probability of electoral success. Moreover, the elite party is aware that the per capita cost of providing public goods is consistently higher than the cost of vote-buying (Proposition 4). In fact, the probability of securing a non-elite voter’s support increases more with each dollar spent on vote-buying than with each dollar allocated to the provision of public goods, which entail a longer time horizon and require action across multiple levels of government (see Lo Bue et al., 2021).

What is the clientelistic expenditure undertaken by the elite party to secure the votes of non-elite citizens? The answer to this question underpins the main results of our article (Propositions 5 and 6). The greater the income inequality or the deeper the poverty in a divided democracy, the higher the elite party’s spending on clientelism during election periods. This finding aligns with a growing body of research that identifies clientelism as a more prominent phenomenon in poorer countries (Stokes, 2021). It further suggests that governing parties have limited incentives to promote redistributive policies when poverty increases the effectiveness of clientelistic strategies (Lo Bue et al., 2021; Robinson & Verdier, 2003; Stokes, 2021).

In Section 3, we conduct a dynamic panel estimation at the country level to analyze the effect of income inequality—measured by the Gini index—on political clientelism. The results indicate a positive and persistent relationship between income inequality and clientelism across all countries and time periods included in the analysis.

This empirical finding is consistent with the recent literature, which shows that clientelism is more prevalent in low-income settings (Bacchus & Boulding, 2022; Fergusson et al., 2022; Liu et al., 2021; Stokes, 2021; Yuda, 2021; Vilchez et al., 2021; Wood, 2018). Under such conditions, poverty and inequality correlate positively with the intensity of clientelistic practices.

Our contribution lies in developing a mainstream probabilistic voting model that incorporates clientelism as a variable, demonstrating theoretically that the prevalence of clientelism within a political regime is a function of income inequality and poverty levels. Crucially, our results imply that clientelism is invariant to changes in the electoral system when inequality and poverty are sufficiently high. In other words, electoral reform aimed at curbing clientelism through institutional engineering will be ineffective in democracies characterized by high levels of income inequality (Lo Bue et al., 2021; Robinson & Verdier, 2003; Stokes, 2021).

The objective of this article is to strengthen and deepen our understanding of how income inequality influences the functioning and intensity of electoral clientelism practiced by political parties within electoral systems. The analysis contributes substantially to the study of electoral and democratic dynamics at the subnational level. It is crucial to understand how and why income inequality serves as a key incentive for political parties to increase clientelistic practices in subnational electoral systems. Existing research on this topic remains limited—largely descriptive and often confined to single-country case studies.

The proposed rational choice model provides a foundation for future comparative research, both across countries and within subnational units of a single country.

The article is structured as follows. Section 2 presents the probabilistic voting model with vote-buying, with particular emphasis on the theoretical characterization discussed in Section 2.1. Section 3 introduces the empirical evidence, while Section 4 concludes the article with key insights and reflections on the broader literature concerning clientelism and inequality.

2. A Probabilistic Voting Model with Vote-Buying

We define a dynamic game with perfect and complete information, involving non-elite voters $n_N$, elite voters $n_E$, and two political parties: A (elite) and B (non-elite). The parties move simultaneously in the first stage of the game. Party A promises to spend $x_A\in {\mathbb{R}}_{+}$ on public goods for the set $\mathcal{E}=1,2,\dots ,n_E$ of elite voters, and $g_A\in {\mathbb{R}}_{+}$ for the set $\mathcal{N}=1,2,\dots ,n_N$ of non-elite voters. In addition, party A can promise a clientelistic payment $c_i\ge 0$ to each non-elite voter $i\in \mathcal{N}$, such that

$$\mathit{c}=\left(\begin{array}{c}{c}_1\\ c_2\\ ⋮\\ c_{n_N}\end{array}\right)\in {\mathbb{R}}^{n_N}$$

is its payment profile. Party B only provides public goods to non-elite voters through a public expenditure $g_B$ and does not offer public goods to the elite or vote-buying payments. In the second stage of the game, voters observe the actions $(x_A,g_A,\mathit{c})$ and $g_B$ taken by parties A and B, respectively, and decide which party to vote for.

2.1. Voters

We now define voters’ preferences in two dimensions: their orientation toward political parties and their evaluation of public expenditure.1

2.1.1. Elite Voters

Preferences over public goods. Following Pani (2011), public goods are produced by a competitive sector of firms operating with a common technology that exhibits constant marginal costs. Individual consumers cannot purchase public goods directly due to free-rider problems; instead, producers sell them to the government, which then provides access to all citizens. Although public goods may be supplied in varying quantities, once acquired by the government, they are fully available to everyone. Let

$${\mathit{x}}^{\top }\left(n_E\right)=\left(x^1\left(n_E\right),x^2\left(n_E\right),\dots ,x^{\mathit{l}}\left(n_E\right)\right)\in {\mathbb{R}}_{+}^{\mathit{l}}$$

be a vector of public policies, where $x^l\left(n_E\right)$ denotes the amount of public good $l=1,2,\dots ,\mathit{l}$ that party A offers to elite voters. Each component depends on $n_E$ such that $\partial x^l\left(n_E\right)/\partial n_E>0$ for all $l=1,2,\dots ,\mathit{l}$. Let

$${\mathit{p}}_E^{\top }=\left(p_E^1,p_E^2,\dots ,p_E^{\mathit{l}}\right)\in {\mathbb{R}}_{+}^{\mathit{l}}$$

be the vector of prices that the government pays to firms for the units of public goods purchased for the elite, where $p_{\mathrm{E}}^l$ is the price of one unit of public good $l=1,2,\dots ,\mathit{l}$. Let ${\phi }_{\mathit{l}}:{\mathbb{R}}_{+}^{\mathit{l}}\times {\mathbb{R}}_{+}^{\mathit{l}}⟶{\mathbb{R}}_{+}$ be a bijection such that

$${\phi }_{\mathit{l}}\left(\mathit{x}\left(n_E\right),{\mathit{p}}_E\right)={\mathit{x}}^{\top }\left(n_E\right)·{\mathit{p}}_E=x\in {\mathbb{R}}_{+},$$

representing the public expenditure associated with the public policy vector $\mathit{x}\left(n_E\right)$ for the set of elite voters $\mathcal{E}$, given the price vector ${\mathit{p}}_E$.

Notice that economic development is measured by the capital stock of public goods. Consequently, an increase in economic development implies a higher capital stock of public goods and, therefore, a reduction in the marginal cost of producing an additional unit of a public good (see Okada et al., 1996). In democracies with high levels of economic development—that is, with a high accumulation of public goods capital—the prices of public goods are relatively low. Since ${\phi }_{\mathit{l}}$ is a bijection, it follows that for every public expenditure x there exists one and only one public policy vector $\mathit{x}\left(n_E\right)$ such that ${\mathit{x}}^{\top }\left(n_E\right)·{\mathit{p}}_E=x$. It holds that $\partial x/\partial x^l\left(n_E\right)=p_E^l$ for each $l=1,2,\dots ,\mathit{l}$.

Let $x⟶u_i\left(x\right)=-{\left(x-x_i\right)}^2$ be the utility function of voter $i\in \mathcal{E}$, where $x_i$ is their ideal level of public expenditure. Let ${\mathit{x}}_i^{\top }\left(n_E\right)=\left(x_i^1\left(n_E\right),x_i^2\left(n_E\right),\dots ,x_i^{\mathit{l}}\left(n_E\right)\right)$ be the vector of ideal amounts of each public good for voter i such that

$${\mathit{x}}_i^{\top }\left(n_E\right)·{\mathit{p}}_E=\left(x_i^1\left(n_E\right),x_i^2\left(n_E\right),\dots ,x_i^{\mathit{l}}\left(n_E\right)\right)·\left(\begin{array}{c}{p}_E^1\\ p_E^2\\ ⋮\\ p_E^{\mathit{l}}\end{array}\right)=x_i\in {\mathbb{R}}_{+}$$

where $x_i^l\left(n_E\right)$ represents the ideal amount of public good $l=1,2,\dots ,\mathit{l}$ demanded by voter i in elections. This expression characterizes the ideal total public expenditure $x_i$ of elite voter $i\in \mathcal{E}$.

The utility function $u_i$ satisfies $u_i\left(0\right)=-x_i^2$ and attains its maximum value $u_i\left(x_i\right)=0$ at the ideal expenditure level $x=x_i$, where preferences are ordered such that $x_i<x_{i+1}$. Assumption 1.A posits that for every $i\in \mathcal{E}$, there exists no $x>0$ such that $u_i\left(x\right)<u_i\left(0\right)$, reflecting the fact that public goods are in short supply.

Assumption 1.A (A.1.A). For each $i\in \mathcal{E}$, we have that $u_i\left(x\right)>u_i\left(0\right)$ if and only if $x>0$, where the domain of the utility function $u_i$ is given by $D\left[u_i\right]=[0,2x_1)$ for every voter $i\in \mathcal{E}$.

Partisan preferences. Each voter $i\in \mathcal{E}$ has well-defined partisan preferences over the set $A,B$ of political parties, represented by the utility function ${\mathit{v}}_i^E:A,B⟶{\mathbb{R}}_{+}$, such that ${\mathit{v}}_i^E\left(A\right)=v_{iA}^E$ and ${\mathit{v}}_i^E\left(B\right)=v_{iB}^E$. We assume that $A{\succ }_{i}B$ if and only if $v_{iA}^E>v_{iB}^E$; $A{\sim }_{i}B$ if and only if $v_{iA}^E=v_{iB}^E$; and $A{\succsim }_{i}B$ if and only if $v_{iA}^E\ge v_{iB}^E$. There are two types of elite voters: $\mathcal{E}\left(A\right)=i:A{\succ }_{i}B$, the more common type; and $\mathcal{E}\left(B\right)=i:B{\succsim }_{i}A$, individuals guided by a criterion of solidarity and empathy with the needs of non-elite individuals. According to the literature surveyed, $\left|\mathcal{E}\right(A\left)\right|>\left|\mathcal{E}\right(B\left)\right|$, that is, the average voter strictly prefers party A to party B. In effect, if

$$n_E=\left|\mathcal{E}\left(A\right)\right|+\left|\mathcal{E}\left(B\right)\right|$$

and $\left|\mathcal{E}\left(A\right)\right|>\left|\mathcal{E}\left(B\right)\right|$ then

$$\left|\mathcal{E}\left(A\right)\right|+\left|\mathcal{E}\left(A\right)\right|>\left|\mathcal{E}\left(B\right)\right|+\left|\mathcal{E}\left(A\right)\right|=n_E$$

Hence, $\left|\mathcal{E}\left(A\right)\right|>n_E/2$. Given the weights, we define the partisan bias $v_{iA}^E-v_{iB}^E=v_i^E$ of voter $i\in \mathcal{E}$ in favor of party A, such that if $v_i^E>0$, we say that there is a bias in favor of party A; if it is negative, the individual exhibits a bias against it. We interpret the magnitude $∥v_i^E∥$ of the partisan bias as the degree to which partisan preferences influence the voting decision. If $∥v_i^E∥=0$, it means that voter i considers solely and exclusively the policy $x_A$ of party A when deciding which party to vote for.

Assumption 2.A. (A.2.A). The partisan weights $v_{iA}^E$ and $v_{iB}^E$ are privately known only to voter $i\in \mathcal{E}$. Therefore, the information available to political parties about the partisan bias $v_i^E$ is such that $v_i^E$ is modeled as a random variable satisfying

$$v_i^E\sim U\left[{\displaystyle \frac{b_E}{f_E}}-{\displaystyle \frac{1}{2f_E}},{\displaystyle \frac{b_E}{f_E}}+{\displaystyle \frac{1}{2f_E}}\right]$$

for each $i=1,2,\dots ,\mathit{l}$ such that $b_E⟶{\varphi }_E\left(b_E\right)$ is a function in which the degree of cohesion $f_E={\varphi }_E\left(b_E\right)$ of partisan biases depends on $b_E$, such that ${\varphi }_E\left(b_E\right)=(k_E-b_E)/{\overline{x}}^2$, where $0<b_E<k_E$, $0<k_E\le 1/2$, and $\overline{x}=(x_1+x_2+\dots +x_{n_E})/n_E$. We have $\overline{x}>b_E/f_E$. The density function is common knowledge.

Given the definition of the function ${\varphi }_E$ in Assumption 2.A, the mean of the elite voter set $\mathcal{E}$ strictly prefers party A over party B whenever $b_E\in [0,1/2]$, and the quantity $b_E/f_E$ corresponds to the mean of the underlying distribution. The function ${\varphi }_E$ captures a fundamental insight into the relationship between the degree of cohesion $f_E$ among elite voters, their average ideal level of public expenditure $\overline{x}$, and the extent to which their partisan bias conditions their voting behavior. The average ideal public expenditure $\overline{x}$ preferred by elite voters is low, insofar as their consumption of public goods—such as education, health, and even security—can be privately insured through their substantial endowments of physical and human capital, thereby reducing their reliance on state provision. This is a well-established result in political economy: “the rich prefer too few public goods while the poor prefer too many” (Persson & Tabellini, 2000).

In the definition of the function ${\varphi }_E$ we have that if the average ideal public expenditure $\overline{x}$ takes very small values, the degree of cohesion $f_E={\varphi }_E\left(b_E\right)$ of the partisan bias of elite voters will be very high. The elite has a high capacity for coordination and collective action, since it comprises a small percentage of the population, who occupy the most influential positions of power (Hafner-Burton et al., 2013). In terms of classical theory of elites, the ones who govern social organizations are few, cohesive, and capable of exercising power, even arbitrarily and violently, over the majority (Campati, 2022; Mosca, 1939; Pareto, 1991).

Elite voters, despite exhibiting a high degree of partisan cohesion, do not base their electoral decisions primarily on partisan identity per se, but rather on the perceived capacity of a governing party to address threats to their privileged positions within the social and institutional hierarchy. Specifically, in the definition of the function ${\varphi }_E$, it is evident that the parameter $b_E$ tends to assume very low values as the degree of cohesion $f_E$ increases.2 Although elites do not constitute a homogeneous bloc, their shared interest in preserving a mutually beneficial social order compels them to engage in coordination and mitigates the intragroup conflicts that typically arise from power rivalries (Best, 2010; Higley, 2010; Meisel, 1958). This dynamic of elite interaction was described as antagonistic cooperation (see Best & Hoffmann-Lange, 2018).

2.1.2. Non-Elite Voters

Preferences over public goods. Following Pani (2011), public goods are produced by a competitive sector of firms operating under a common technology that exhibits constant marginal costs. Due to the presence of free-rider problems, individual consumers cannot purchase public goods directly. Instead, these goods are sold to the government by the producers, and subsequently made available to all citizens. Although public goods can be supplied in varying quantities, once acquired by the government, they are non-excludable and fully accessible to the entire population. Let ${\mathit{g}}^{\top }\left(n_N\right)=\left(g^1\left(n_N\right),g^2\left(n_N\right),\dots ,g^{\mathit{m}}\left(n_N\right)\right)\in {\mathbb{R}}_{+}^{\mathit{m}}$ be a vector of public policies, where $g^m\left(n_N\right)$ denotes the quantity of public good $m=1,2,\dots ,\mathit{m}$ consumed by non-elite voters. We assume that the provision of each public good is increasing in the size of the non-elite population, so that $\partial g^m\left(n_N\right)/\partial n_N>0$ for all $m=1,2,\dots ,\mathit{m}$. Let

$${\mathit{p}}_N^{\top }=\left(p_N^1,p_N^2,\dots ,p_N^{\mathit{m}}\right)\in {\mathbb{R}}_{+}^{\mathit{m}}$$

be the vector of prices paid by the government to firms for the provision of public goods destined for non-elite voters, where $p_N^m$ denotes the price of one unit of public good $m=1,2,\dots ,\mathit{m}$. Let

$${\phi }_{\mathit{m}}:{\mathbb{R}}_{+}^{\mathit{m}}\times {\mathbb{R}}_{+}^{\mathit{m}}⟶{\mathbb{R}}_{+}$$

be a bijection such that

$${\phi }_{\mathit{m}}\left(\mathit{g}\left(n_N\right),{\mathit{p}}_N\right)={\mathit{g}}^{\top }\left(n_N\right)·{\mathit{p}}_N=g\in {\mathbb{R}}_{+},$$

which represents the total government expenditure on the vector of public policies $\mathit{g}\left(n_N\right)$ targeted to the set $\mathcal{N}$ of non-elite voters. Hence, for every level of public expenditure g, there exists a unique vector of public policies $\mathit{g}\left(n_N\right)$ such that ${\mathit{p}}_N^{\top }·\mathit{g}\left(n_N\right)=g$. It follows that the marginal cost of providing an additional unit of public good m is given by $\partial g/\partial g^m\left(n_N\right)=p_N^m$ for each $m=1,2,\dots ,\mathit{m}$. As previously noted, economic development is proxied by the capital stock of public goods. An increase in economic development corresponds to a larger stock of public goods, which in turn leads to a reduction in the marginal cost of producing an additional unit of any given public good (Okada et al., 1996). In democracies characterized by low levels of economic development and minimal public capital accumulation, the prices of public goods tend to be high.

The preferences of voter $i\in \mathcal{N}$ are represented by the utility function $g⟶u_i\left(g\right)=-{\left(g-g_i\right)}^2$, where $g_i$ denotes voter i’s ideal level of total public expenditure. Let the ideal bundle of public goods for voter i be given by the vector

$${\mathit{g}}_i^{\top }\left(n_N\right)=\left(g_i^1\left(n_N\right),g_i^2\left(n_N\right),\dots ,g_i^{\mathit{m}}\left(n_N\right)\right)$$

which depends functionally on the size of the non-elite population $n_N$. The total expenditure associated with this bundle is

$${\mathit{g}}_i^{\top }\left(n_N\right)·{\mathit{p}}_N=\left(g_i^1\left(n_N\right),g_i^2\left(n_N\right),\dots ,g_i^{\mathit{m}}\left(n_N\right)\right)·\left(\begin{array}{c}{p}_N^1\\ p_N^2\\ ⋮\\ p_N^{\mathit{m}}\end{array}\right)=g_i\in {\mathbb{R}}_{+},$$

where $g_i^m\left(n_N\right)$ denotes the ideal quantity of public good $m=1,2,\dots ,\mathit{m}$ for voter i, as a function of $n_N$. The utility function satisfies $u_i\left(0\right)=-g_i^2$, and $u_i\left(g\right)=0$ if and only if $g=g_i$, attaining a global maximum at the ideal public expenditure $g_i$, with $g_i<g_{i+1}$ for each $i\in \mathcal{N}-\left\{n_N\right\}$.

Assumption 1.B states that there exists no $g>0$ such that $u_i\left(g\right)<u_i\left(0\right)$ for any $i\in \mathcal{N}$, reflecting the fact that utility is strictly increasing in public expenditure up to the ideal point and that utility losses due to overprovision are always smaller than those due to total absence, owing to a structural shortage of public goods.

Assumption 1.B. (A.1.B). For each $i=1,2,\dots ,n_N$, we have $u_i\left(g\right)>u_i\left(0\right)$ if and only if $g>0$. The domain of the utility function $u_i$ is given by $D\left[u_i\right]=[0,2g_i)$, for each voter $i\in \mathcal{N}$.

Partisan preferences. Each voter $i\in \mathcal{N}$ holds well-defined partisan preferences over the set of political parties $A,B$, represented by the utility function ${\mathit{v}}_i^N:\left\{A,B\right\}⟶{\mathbb{R}}_{+}$, such that ${\mathit{v}}_i^N\left(A\right)=v_{iA}^N$ and ${\mathit{v}}_i^N\left(B\right)=v_{iB}^N$. We say that $B{\succ }_{i}A$ if and only if $v_{iB}^N>v_{iA}^N$; $A{\sim }_{i}B$ if and only if $v_{iB}^N=v_{iA}^N$; and $B{\succsim }_{i}A$ if and only if $v_{iB}^N\ge v_{iA}^N$.

Analogously to the structure above, we classify non-elite voters into two types: $\mathcal{N}\left(A\right)=i\in \mathcal{N}:A{⪰}_{i}B$, which is the less frequent type; and $\mathcal{N}\left(B\right)=i\in \mathcal{N}:B{\succ }_{i}A$, representing those biased toward the needs of their group. Therefore, it holds that $\left|\mathcal{N}\right(B\left)\right|>\left|\mathcal{N}\right(A\left)\right|$—i.e., the average voter strictly prefers party B over party A. In effect, if

$$n_N=\left|\mathcal{N}\left(A\right)\right|+\left|\mathcal{N}\left(B\right)\right|$$

and $\left|\mathcal{N}\left(B\right)\right|>\left|\mathcal{N}\left(A\right)\right|$, then

$$\left|\mathcal{N}\left(B\right)\right|+\left|\mathcal{N}\left(B\right)\right|>\left|\mathcal{N}\left(B\right)\right|+\left|\mathcal{N}\left(A\right)\right|=n_N$$

Hence, $\left|\mathcal{N}\left(B\right)\right|>n_N/2$. Given the utility weights, we define the partisan bias of voter $i\in \mathcal{N}$ as $v_i^N=v_{iB}^N-v_{iA}^N$, which captures the relative preference for party B. If $v_i^N>0$, we say that voter i exhibits a bias in favor of party B; if $v_i^N<0$, the bias is against party B.

We interpret the magnitude $∥v_i^N∥$ as the degree to which partisan preferences influence the voting decision of non-elite voter i. If $∥v_i^N∥=0$, this implies that voter $i\in \mathcal{N}$ does not consider partisan affiliation in their decision. Therefore, as $∥v_i^N∥$ increases, partisan preferences gain importance in shaping the electoral choice.

Assumption 2.B. (A.2.B). The partisan weights $v_{iA}^N$ and $v_{iB}^N$ are private information for each voter $i\in \mathcal{N}$. Consequently, the partisan bias $v_i^N=v_{iB}^N-v_{iA}^N$ is modeled as a random variable with distribution

$$v_i^N\sim U\left[{\displaystyle \frac{b_N}{f_N}}-{\displaystyle \frac{1}{2f_N}},{\displaystyle \frac{b_N}{f_N}}+{\displaystyle \frac{1}{2f_N}}\right]$$

for all $i=1,2,\dots ,\mathit{m}$, where $b_N$ is a location parameter and $f_N={\varphi }_N\left(b_N\right)$ denotes the degree of cohesion in partisan preferences among non-elite voters. The function ${\varphi }_N:\left(0,k_N\right)⟶{\mathbb{R}}_{+}$ is defined by

$${\varphi }_N\left(b_N\right)={\displaystyle \frac{\left(k_N-b_N\right)}{{\overline{g}}^2}}$$

where $0<b_N<k_N$, $0<k_N\le 1/2$ and $\overline{g}={\displaystyle \left(g_1+g_2+\dots +g_{n_N}\right)/n_N}$ is the average ideal public expenditure among non-elite voters. It is assumed that $\overline{g}>b_N/f_N$, and the distribution of $v_i^N$ is common knowledge among political parties.

The function ${\varphi }_N$, introduced in Assumption 2.B, guarantees that the average voter in the set $\mathcal{N}$ strictly prefers party B over party A. In addition, ${\varphi }_N$ encapsulates an important intuition regarding the inverse relationship between the degree of cohesion $f_N$ among non-elite voters and their average ideal public expenditure $\overline{g}$.

In contrast to the elite, non-elite voters are heterogeneous, exhibit low levels of social integration, and have limited access to positions of power. They are highly atomized, and their collective political agency is weak; as a result, the median voter within the non-elite is particularly susceptible to elite influence (Daby, 2021; Graziano, 2019; Holcombe, 2021; Inglehart, 1997; Schwander, 2019; Stokes et al., 2013; Wren & McElwain, 2011). The asymmetric access to public goods and opportunities characteristic of non-elite groups undermines their cohesion and contributes to broader patterns of social disarticulation (Hicken, 2007; Husermann & Schwander, 2012; Robinson & Verdier, 2003; Stokes, 2007).

A defining feature of non-elite voters is that their average ideal public expenditure $\overline{g}$ tends to be high, due to their limited endowments of both physical and human capital. Unlike elite actors, they face substantial difficulties in privately meeting their needs for public goods such as education and health. As a result, when $\overline{g}$ is large, the degree of cohesion $f_N={\varphi }_N\left(b_N\right)$ is necessarily low.

2.1.3. Comparative Analysis: Elite and Non-Elite Voters

The structure of our model rests on a fundamental asymmetry between elite and non-elite voters in terms of their preferences for public goods, the cohesion of their partisan biases, and their capacity for political coordination. These differences are not merely empirical regularities; rather, they emerge from structural inequalities embedded in the distribution of economic resources, political access, and social capital.

Table 1. Structural asymmetries between elite and non-elite voters in divided democracies.

Dimension	Elite Voters	Non-Elite Voters
Average ideal public expenditure	Low ($\overline{x}$)	High ($\overline{g}$)
Partisan cohesion (f)	High ($f_E$)	Low ($f_N$)
Collective action capacity	Strong (coordination, organization)	Weak (atomization, low integration)
Dependence on state provision	Low (self-sufficiency via private provision)	High (state-dependent for public goods)
Access to power	High (control of institutions/resources)	Low (structurally excluded)
Dominant partisan bias	Toward party A (elite)	Toward party B (non-elite)

Source: Own elaboration based on (Pareto, 1991; Persson & Tabellini, 2000; Robinson & Verdier, 2003).

summarizes the main dimensions of this asymmetry.

The contrast between the two voter groups reveals a deep segmentation in the structure of political demand. Elite voters tend to demand fewer public goods, as they can privately secure access to education, healthcare, and security. This results in a low average ideal public expenditure $\overline{x}$, which—through the function ${\varphi }_E\left(b_E\right)$—yields a high degree of partisan cohesion $f_E$. In addition, their small size, internal cohesion, and control over institutional resources enable them to act strategically in defense of their political and economic interests, often by supporting parties that preserve the existing distribution of power and wealth.

Non-elite voters, by contrast, are socially fragmented, economically dependent, and politically disarticulated. Their high demand for public goods $\overline{g}$ results in a broad dispersion of preferences and lower cohesion $f_N$, as captured by the function ${\varphi }_N\left(b_N\right)$. This lack of cohesion undermines their collective capacity to articulate and sustain demands for redistributive policies, rendering them more susceptible to clientelistic strategies.

The asymmetry in cohesion and political organization between elite and non-elite groups is not merely a sociological condition, but a fundamental mechanism through which democratic representation becomes systematically skewed in favor of entrenched elites.

This structural asymmetry also enables party A, as the partisan expression of elite interests, to strategically instrumentalize the non-elite electorate. Fully aware of the social fragmentation, low partisan cohesion $f_N$, and high demand for public goods $\overline{g}$ that characterize the voter set $\mathcal{N}$, party A deploys clientelistic mechanisms to selectively co-opt segments of this group.

Rather than addressing the aggregate redistributive demands of non-elite voters through universal public policies, party A exploits their atomization by offering targeted benefits in exchange for electoral support. This instrumental use of clientelism does not stem from ideological alignment; rather, it functions as a political technology aimed at neutralizing potential opposition and reinforcing elite dominance within democratic competition.

The effectiveness of this strategy relies precisely on the asymmetry in organizational capacity between $\mathcal{E}$ and $\mathcal{N}$, further distorting democratic representation by subordinating broad-based public goods provision to selective, transactional exchanges.

2.2. Political Parties

Both political parties are fully informed of Assumptions 1.A, 1.B, 2.A, and 2.B; moreover, this information is common knowledge. Political parties are assumed to be rational actors, and this rationality is also common knowledge.

2.2.1. Elite Political Party

The elite-aligned political party A is aware that the majority of non-elite voters prefer party B, and that these voters constitute the numerical majority of the electorate. In order to maximize its chances of winning the election, party A adopts a dual strategy:

1. It offers a positive amount $g_A>0$ of public goods to non-elite voters, seeking to approximate the ideal point of the median voter within the B-aligned electorate. This offer is made in addition to the provision of $x_A>0$ units of public goods targeted to elite voters. 2. It offers a clientelistic payment equal to $c_i$ such that $v_{iA}^N+c_i>v_{iB}^N$.

2.2.2. Non-Elite Political Party

The non-elite party offers only the public good demanded by non-elite individuals, such that $g_B>0$.

Assumption 3. (A.3.). Society $\mathcal{S}=\mathcal{N}\cup \mathcal{E}$ is divided into elite individuals ($n_E$) and non-elite individuals ($n_N$). If $n_E<n_N$, then $n_N$ is always at least $1/\left(2b_N\right)$ times larger than $n_E$. Formally, this implies that $2b_N·n_N>n_E$. Moreover, it also holds that $2b_N·(n_N+1)>n_E+1$.

The division within society $\mathcal{S}$ should not be considered negligible: non-elite voters must constitute an overwhelming majority. This is an empirically observable pattern in most democracies, where the elite typically represents around 1% of the total population (Alvaredo & Gasparini, 2015; Friedman et al., 2015; Williams & Filippakou, 2010). The magnitude of the division between elite and non-elite groups may vary depending on the type of government in power.

2.3. The Characterization Result

We present below the results of the model, developed in detail as theoretical predictions that are subsequently tested in the econometric evidence section. These predictions also outline an empirical research agenda concerning the relationship between socioeconomic inequality and clientelism—one that may be pursued by the authors or by other scholars in the field.

Proposition 1.

Under rational behavior by political parties, and given Assumptions 1.A, 1.B, 2.A, 2.B, and 3, both parties offer the same level of public goods in the Nash equilibrium of the probabilistic voting game. That is,

$$g_A^{*}=g_B^{*}=\left(g_1^{*}\left(n_N\right),g_2^{*}\left(n_N\right),\dots ,g_{\mathit{m}}^{*}\left(n_N\right)\right)·{\mathit{p}}_N=\overline{g},$$

where $\overline{g}$ is the level of public expenditure that maximizes the expected number of votes from non-elite voters. Consequently, $\overline{g}$ is both the average ideal expenditure of non-elite voters and the public expenditure offered by political parties during the electoral campaign. In contrast, the public expenditure offered by the elite party to elite voters is given by

$$x_A^{*}=\left(x_1^{\ast }\left(n_E\right),x_2^{\ast }\left(n_E\right),\dots ,g_{\mathit{l}}^{\ast }\left(n_E\right)\right)·{\mathit{p}}_E=\overline{x}$$

which maximizes the expected number of votes obtained from elite voters. As a result, the expected vote shares for parties A and B, denoted $V_A^{\ast }$ and $V_B^{\ast }$, respectively, when party A engages in clientelistic strategies, are given by

$$\left(\begin{array}{c}{V}_A^{\ast }\\ V_B^{\ast }\end{array}\right)=\left(\begin{array}{c}{n}_N·\left({\displaystyle \frac{1}{2}}-b_N\right)+n_E·\left({\displaystyle \frac{1}{2}}+b_E+f_E·{\overline{x}}^2\right)+f_N·\mathit{c}\\ n_N·\left({\displaystyle \frac{1}{2}}+b_N\right)+n_E·\left({\displaystyle \frac{1}{2}}-b_E-f_E·{\overline{x}}^2\right)-f_N·\mathit{c}\end{array}\right)$$

where $\mathit{c}$ represents the vector of clientelistic transfers, and the expressions capture the expected electoral payoff of each party under asymmetric strategies.

Proof. 

If both parties have common knowledge of Assumptions 1.A—3., and of players rationality, they will try to anticipate the best response of voters in the second stage of the game. They know that the probability

$$P_A^{iE}=\mathrm{Pr}\left(v_{iA}^E-v_{iB}^E\ge x_A^2-2x_A·x_i\right)$$

that voter $i\in \mathcal{E}$ votes for party A is

$$P_A^{iE}={\displaystyle \frac{1}{2}}+b_E-f_E·\left(x_A^2-2x_A·x_i\right)$$

such that

$$V_A^E=n_E·\left({\displaystyle \frac{1}{2}}+b_E-f_E·x_A^2\right)+2·f_E·x_A·\sum _{i=1}^{n_E}{x}_i$$

(1)

is the number of votes that party A expects from elite voters. The party must determine public expenses $x_A$ that maximize the expected votes:

$${\displaystyle \frac{\partial V_A^E\left(x_A\right)}{\partial x_A}}={\displaystyle \frac{\partial }{\partial x_A}}\left(n_E·\left({\displaystyle \frac{1}{2}}+b_E-f_E·x_A^2\right)+2·f_E·x_A·\sum _{i=1}^{n_E}{x}_i\right)=-2·n_E·f_E·x_A+2·f_E·\sum _{i=1}^{n_E}{x}_i$$

such that

$${\displaystyle \frac{\partial V_A^E\left(x_A^{\ast }\right)}{\partial x_A}}=0⟺-2·n_E·f_E·x_A^{\ast }+2·f_E·\sum _{i=1}^{n_E}{x}_i=0⟺x_A^{\ast }=\overline{x}$$

Therefore, it is common knowledge that party A chooses an amount of public goods $x_A^{\ast }=\overline{x}$ and obtains elite votes equal to

$$V_A^E=n_E·\left({\displaystyle \frac{1}{2}}+b_E+f_E·{\overline{x}}^2\right)$$

The number of elite votes $V_B^E=n_E-V_A^E$ expected by party B would be

$$V_B^E=n_E·\left({\displaystyle \frac{1}{2}}-b_E-f_E·{\overline{x}}^2\right)$$

It is also common knowledge that the non-elite voter $i\in \mathcal{N}$ votes for party A if

$$v_{iA}^N+c_i+u_i\left(g_A\right)\ge v_{iB}^N+u_i\left(g_B\right)$$

such that

$$P_A^{iN}={\displaystyle \frac{1}{2}}-b_N-f_N·\left(g_A^2-g_B^2-2·g_i·\left(g_A-g_B\right)-c_i\right)$$

is the probability $P_A^{iN}=\mathrm{Pr}\left(v_i^N<g_B^2-g_A^2-2·g_i·\left(g_B-g_A\right)+c_i\right)$ that the voter $i\in \mathcal{N}$ votes for party A. Therefore,

$$V_A^N=n_N·\left({\displaystyle \frac{1}{2}}-b_N-f_N·\left(g_A^2-g_B^2\right)\right)+2f_N·\left(g_A-g_B\right)·\sum _{i=1}^{n_N}{g}_i+f_N·\sum _{i=1}^{n_N}{c}_i$$

(2)

is the number of expected non-elite votes. Therefore, party A must determine the amount of public goods $g_A$ that maximizes its expected votes:

$$\begin{array}{ccc}{\displaystyle \frac{\partial V_A^N\left(g_A\right)}{\partial g_A}}& =& {\displaystyle \frac{\partial }{\partial g_A}}\left(n_N·\left({\displaystyle \frac{1}{2}}-b_N-f_N·\left(g_A^2-g_B^2\right)\right)+2f_N·\left(g_A-g_B\right)·\sum _{i=1}^{n_N}{g}_i+f_N·\sum _{i=1}^{n_N}{c}_i\right)\\ & =& -2·n_N·f_N·g_A+2·f_N·\sum _{i=1}^{n_E}{g}_i\end{array}$$

such that

$${\displaystyle \frac{\partial V_A^N\left(g_A^{\ast }\right)}{\partial g_A}}=0⟺-2·n_N·f_N·g_A^{\ast }+2·f_N·\sum _{i=1}^{n_E}{g}_i=0⟺g_A^{\ast }=\overline{g}$$

Therefore, it is common knowledge that during the elections the party A offers a public policy agenda $g_A^{\ast }=\overline{g}$ to maximize the non-elite votes, such that

$$V_A^N=f_N·\mathit{c}+n_N·\left({\displaystyle \frac{1}{2}}-b_N\right)$$

is the number of votes party A expects from its optimal strategy $g_A^{\ast }=\overline{g}$, such that $\mathit{c}={\sum }_{i=1}^{n_N}{c}_i$ is total expenses in clientelism. The number of non-elite votes $V_B^N=n_N-V_A^N$ expected by party B is

$$V_B^N=n_N\left({\displaystyle \frac{1}{2}}+b_N\right)-f_N·\mathit{c}$$

(3)

Therefore, the number of votes for every party is common knowledge:

$$\begin{array}{c}{V}_A^{\ast }=n_N·\left({\displaystyle \frac{1}{2}}-b_N\right)+n_E·\left({\displaystyle \frac{1}{2}}+b_E+f_E·{\overline{x}}^2\right)+f_N·\mathit{c}\end{array}$$

(4)

such that $V_A^{\ast }=V_A^N+V_A^E$ are the equilibrium votes for party A, and

$$\begin{array}{c}{V}_B^{\ast }=n_N·\left({\displaystyle \frac{1}{2}}+b_N\right)+n_E·\left({\displaystyle \frac{1}{2}}-b_E-f_E·{\overline{x}}^2\right)-f_N·\mathit{c}\end{array}$$

(5)

such that $V_B^{\ast }=V_B^N+V_B^E$ are the equilibrium votes for party B. □

Proposition 2 builds upon the idea that it is public knowledge that most voters prefer the non-elite party. The elite party knows that they are in the minority, even after matching the $\overline{g}$ average public expenditure offered by the non-elite party during the political campaign. Therefore, the elite party knows that it has to engage in clientelism to win the elections. This idea is presented in Proposition 6, where ${\left.V_p^{\ast }\right|}_{\mathit{c}=0}$ are the expected votes of every party $p\in \left\{A,B\right\}$ without the term $f_N·\mathit{c}$.

Proposition 2.

Without clientelism, the number of votes expected by the elite party is strictly less than the votes expected by the non-elite party. Formally, if $\mathit{c}=0$, then ${\left.V_B^{\ast }\right|}_{\mathit{c}=0}>{\left.V_A^{\ast }\right|}_{\mathit{c}=0}$.

Proof. 

We want to demostrate that if $\mathit{c}=0$, then $V_B^{\ast }{\mid }_{\mathit{c}=0}>V_A^{\ast }{\mid }_{\mathit{c}=0}$. If $\mathit{c}=0,$ and $\overline{g}$ is the public expenses promised by the parties during the campaign, we have

$$\left(\begin{array}{c}{V}_A^{\ast }{\mid }_{\mathit{c}=0}\\ V_B^{\ast }{\mid }_{\mathit{c}=0}\end{array}\right)=\left(\begin{array}{c}{n}_N·\left({\displaystyle \frac{1}{2}}-b_N\right)+n_E·\left({\displaystyle \frac{1}{2}}+b_E+f_E·{\overline{x}}^2\right)\\ n_N·\left({\displaystyle \frac{1}{2}}+b_N\right)+n_E·\left({\displaystyle \frac{1}{2}}-b_E-f_E·{\overline{x}}^2\right)\end{array}\right)$$

giving the vector of expected votes if the elite party does not engage in clientelism. In fact, by hypothesis, $b_E+f_E·{\overline{x}}^2<{\displaystyle \frac{1}{2}}$ and $n_N>{\displaystyle \frac{1}{2b_N}}·n_E$. Therefore,

$$n_N·b_N>{\displaystyle \frac{n_E}{2}}>n_E·\left(b_E+f_E·{\overline{x}}^2\right)$$

Consequently,

$$\begin{array}{cccc}& n_N·b_N& >& n_E·\left(b_E+f_E·{\overline{x}}^2\right)\\ ⟺& n_N·b_N+n_N·b_N& >& n_E·\left(b_E+f_E·{\overline{x}}^2\right)+n_E·\left(b_E+f_E·{\overline{x}}^2\right)\\ ⟺& n_N·b_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)& >& -n_N·b_N+n_E·\left(b_E+f_E·{\overline{x}}^2\right)\\ ⟺& {\displaystyle \frac{n_N}{2}}+n_N·b_N+{\displaystyle \frac{n_E}{2}}-n_E·\left(b_E+f_E·{\overline{x}}^2\right)& >& {\displaystyle \frac{n_N}{2}}-n_N·b_N+{\displaystyle \frac{n_E}{2}}+n_E·\left(b_E+f_E·{\overline{x}}^2\right)\\ ⟺& n_N·\left({\displaystyle \frac{1}{2}}+b_N\right)+n_E·\left({\displaystyle \frac{1}{2}}-b_E-f_E·{\overline{x}}^2\right)& >& n_N·\left({\displaystyle \frac{1}{2}}-b_N\right)+n_E·\left({\displaystyle \frac{1}{2}}+b_E+f_E·{\overline{x}}^2\right)\end{array}$$

We have demostrated that $V_B^{\ast }{\mid }_{\mathit{c}=0}>V_A^{\ast }{\mid }_{\mathit{c}=0}$. □

Proposition 3 clarifies that the elite party has to engage in clientelism—it is not an option. The problem is to determine the amount to be spent on clientelism. Proposition 3 presents this feature.

Proposition 3.

To obtain enough non-elite votes such that $V_A^{\ast }$ is strictly more than $V_B^{\ast }$, the elite party spends the following on clientelism: ${\mathit{c}}^{\ast }={\displaystyle \left(b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)\right)/f_N}$.

Proof. 

We have that (4) > (5) if, and only if,

$$\mathit{c}\ge {\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{f_N}}$$

(6)

such that

$${\mathit{c}}^{\ast }={\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{f_N}}$$

(7)

gives the lowest possible clientelism expenses that allow party A to win the elections, with the public goods expenses as promised during the campaign equal to $\overline{x}$ for the elite, and $\overline{g}$ for the non-elite, and clientelism expenses equal to $\mathit{c}$: $\left(x_A,g_A.\mathit{c}\right)=\left(\overline{x},\overline{g},\mathit{c}\right)$ □

Proposition 4 establishes a cost comparison between clientelism and the provision of public goods as electoral strategies targeting non-elite voters.

Proposition 4.

The average cost of clientelism ${\mathit{c}}^{\ast }/n_N$ is strictly lower than the average ideal public expenditure $\overline{g}$ promised during the political campaign.

Proof. 

We must demonstrate that ${\displaystyle {\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{f_N·n_N}}}<\overline{g}$. However, by hypothesis,

$$\begin{array}{ccc}{\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{f_N}}& <& {\displaystyle \frac{b_N}{f_N}}\end{array}$$

and ${\displaystyle \frac{b_N}{f_N}}<\overline{g}$. □

In equilibrium, the average cost of clientelism per vote is strictly lower than the average ideal public expenditure $\overline{g}$ demanded by non-elite voters (Proposition 4). This result has direct implications for the strategic behavior of the elite-aligned party. Given that non-elite voters exhibit low partisan cohesion (i.e., low $f_N$) and a high demand for public goods (i.e., high $\overline{g}$), as summarized in Table 1, they become particularly susceptible to clientelistic strategies.

The atomization of the non-elite group and its limited capacity for collective action make the selective transfer of private benefits—a hallmark of clientelism—a more cost-effective electoral strategy for the elite party than the universal provision of public goods.

By contrast, the high cohesion and lower public goods demand of elite voters imply that this group can be more efficiently targeted through programmatic, aggregated policies, without relying on clientelistic inducements.

This asymmetry in the cost-effectiveness of electoral strategies reflects a deeper asymmetry in political representation under divided democracies: while the elite group is integrated through policy-based channels, the non-elite group is managed through selective, transactional mechanisms. This cost advantage arises precisely because the low cohesion $f_N$ reduces the marginal cost of inducing defection through selective benefits, as formalized in Proposition 4.

Proposition 5 establishes that an increase in the average ideal demand for public goods, $\overline{g}$, leads to a more-than-proportional increase in the cost of implementing clientelistic strategies. A higher level of $\overline{g}$ reflects, ceteris paribus, deeper material deprivation among non-elite individuals, which in turn heightens their susceptibility to clientelistic inducements.

The elite-aligned party understands that greater socioeconomic inequality and poverty amplify the relative effectiveness of clientelism as an electoral tactic. Accordingly, the cost structure of clientelism becomes increasingly sensitive to shifts in redistributive demand.

Proposition 5.

Under the assumption that all other factors remain constant (ceteris paribus), if the average ideal public expenditure $\overline{g}$ increases— i.e., $△\overline{g}>0$—then the elite party’s clientelism expenditures ${\mathit{c}}^{\ast }$ increase more than proportionally. In symbols, $△{\mathit{c}}^{\ast }>△\overline{g}$.

Proof. 

Recall that

$${\mathit{c}}^{\ast }={\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{f_N}}={\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{k_N-b_N}·{\overline{g}}^2}$$

And in fact

$$\begin{array}{cccccc}& b_E+f_E·{\overline{x}}^2& <& {\displaystyle \frac{1}{2}}& & \mathrm{by}\mathrm{hypothesis}\\ ⟺& n_E·\left(b_E+f_E·{\overline{x}}^2\right)& <& {\displaystyle n_E/2}\\ ⟺& -n_E·\left(b_E+f_E·{\overline{x}}^2\right)& >& -{\displaystyle n_E/2}\\ ⟺& b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)& >& b_N·n_N-{\displaystyle n_E/2}\\ ⟺& {\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{k_N-b_N}}& >& {\displaystyle \frac{b_N·n_N-{\displaystyle n_E/2}}{k_N-b_N}};& & k_N-b_N>0\end{array}$$

Therefore,

$${\displaystyle {\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{k_N-b_N}}>\frac{2b_N·n_N-{\displaystyle n_E}}{2\left(k_N-b_N\right)}}$$

(8)

However,

$$\begin{array}{cccccc}& n_N+1& >& {\displaystyle \frac{1}{2b_N}}\left(n_E+1\right)& ;& \mathrm{by}\mathrm{hypothesis}\\ ⟺& 2b_N·\left(n_N+1\right)& >& n_E+1\\ ⟺& 2b_N·\left(n_N+1\right)-n_E& >& 1\\ ⟺& 2b_N·n_N+2b_N-n_E& >& 1\end{array}$$

And therefore

$$2b_N{n}_N-n_E>1-2b_N$$

(9)

and

$$\begin{array}{cccccccccc}& 0& <& b_N& <& k_N& <& 1/2& ;& \mathrm{by}\mathrm{definition}\\ ⟺& 0& <& k_N-b_N& <& 1/2-b_N\end{array}$$

Consequently

$$0<2\left(k_N-b_N\right)<1-2b_N$$

(10)

Using Inequalities (9) and (10) in Inequality (8) we have

$${\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{k_N-b_N}}>1$$

(11)

We have proved that

$${\displaystyle \frac{\partial {\mathit{c}}^{\ast }}{\partial \overline{g}}=2·}{\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{k_N-b_N}·\overline{g}}>1$$

if $\overline{g}>1$. □

Proposition 5 shows that as the average ideal demand for public goods $\overline{g}$ increases—an indicator of deeper structural inequality—the optimal clientelistic expenditure ${\mathit{c}}^{\ast }$, which maximizes the expected number of votes for the elite party, also increases more than proportionally. This dynamic does not reduce the strategic appeal of clientelism; on the contrary, in contexts of greater poverty and exclusion—where voters lack effective access to universal public goods—clientelism becomes a particularly efficient tool for inducing political support through selective benefits.

In addition, we have one of our most important results in Proposition 6. Higher inequality in democratic regimes is conducive to higher clientelism.

Proposition 6.

If inequality increases in a divided democracy, ceteris paribus, the expenses of clientelism for the elite party will increase. On the other hand, a decrease in the average cost of producing public goods will determine ceteris paribus a reduction in clientelism.

Proof. 

If inequality increases in a divided democracy, the elite party expenses in clientelism also increase. Additionally, we have that ${\displaystyle \frac{\partial n}{\partial n_N}}={\displaystyle \frac{\partial n_N}{\partial n_N}}+{\displaystyle \frac{\partial n_E}{\partial n_N}}=0$. We must demonstrate that

$${\displaystyle \frac{\partial {\mathit{c}}^{\ast }}{\partial n_N}}={\displaystyle \frac{\partial }{\partial n_N}\left(\frac{b_N{n}_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{f_N}\right)}>0$$

It holds that

$$\begin{array}{cc}& {\displaystyle \frac{\partial }{\partial n_N}}\left({\displaystyle \frac{b_N{n}_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{f_N}}\right)\\ =& {\displaystyle \frac{\left(b_N-\left({\displaystyle \frac{\partial n_E}{\partial n_N}}·\left(b_E+f_E·{\overline{x}}^2\right)+n_E·{\displaystyle \frac{\partial \left(b_E+f_E·{\overline{x}}^2\right)}{\partial n_N}}\right)\right)·f_N-{\displaystyle \left(b_N{n}_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)\right){\displaystyle \frac{\partial f_N}{\partial n_N}}}}{f_N^2}}\end{array}$$

However

$$\begin{array}{ccccc}{\displaystyle \frac{\partial \left(b_E+f_E·{\overline{x}}^2\right)}{\partial n_N}}& =& {\overline{x}}^2·{\displaystyle \frac{\partial f_E}{\partial n_N}}& +& f_E·{\displaystyle \frac{\partial {\overline{x}}^2}{\partial n_N}}\\ & =& {\overline{x}}^2·{\displaystyle {\displaystyle \frac{\partial f_E}{\partial \overline{x}}}·{\displaystyle {\displaystyle \frac{\partial \overline{x}}{\partial n_E}}·}{\displaystyle \frac{\partial n_E}{\partial n_N}}}& +& f_E·2\overline{x}·{\displaystyle {\displaystyle \frac{\partial \overline{x}}{\partial n_E}}·{\displaystyle \frac{\partial n_E}{\partial n_N}}}\\ & =& {\overline{x}}^2·{\displaystyle {\displaystyle \frac{\partial f_E}{\partial \overline{x}}}·{\displaystyle {\displaystyle \frac{\partial \overline{x}}{\partial n_E}}·}\left(-1\right)}& +& f_E·2\overline{x}·{\displaystyle {\displaystyle \frac{\partial \overline{x}}{\partial n_E}}·\left(-1\right)}\end{array}$$

In other words

$${\displaystyle \frac{\partial \left(b_E+f_E·{\overline{x}}^2\right)}{\partial n_N}}=\left(-1\right)·\overline{x}·{\displaystyle {\displaystyle \frac{\partial \overline{x}}{\partial n_E}}}·\left(\overline{x}{\displaystyle {\displaystyle \frac{\partial f_E}{\partial \overline{x}}}{\displaystyle +}2}{f}_E\right)$$

Therefore

$$\begin{array}{cc}& {\displaystyle \frac{\partial n_E}{\partial n_N}}·\left(b_E+f_E·{\overline{x}}^2\right)+n_E·{\displaystyle \frac{\partial \left(b_E+f_E·{\overline{x}}^2\right)}{\partial n_N}}\\ =& {\displaystyle \left(-1\right)·\left(b_E+f_E·{\overline{x}}^2\right)+n_E·\left(-1\right)·\overline{x}·{\displaystyle {\displaystyle \frac{\partial \overline{x}}{\partial n_E}}}·\left(\overline{x}{\displaystyle {\displaystyle \frac{\partial f_E}{\partial \overline{x}}}{\displaystyle +}2}{f}_E\right)}\\ =& {\displaystyle {\displaystyle \left(-1\right)·\left(b_E+f_E·{\overline{x}}^2+n_E·\overline{x}·{\displaystyle {\displaystyle \frac{\partial \overline{x}}{\partial n_E}}}·\left(\overline{x}{\displaystyle {\displaystyle \frac{\partial f_E}{\partial \overline{x}}}{\displaystyle +}2}{f}_E\right)\right)}}\end{array}$$

And consequently,

$$\begin{array}{cc}& {\displaystyle \frac{\partial }{\partial n_N}}\left({\displaystyle \frac{b_N{n}_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{f_N}}\right)\\ =& {\displaystyle \frac{\left(b_N+b_E+f_E·{\overline{x}}^2+n_E·\overline{x}·{\displaystyle {\displaystyle \frac{\partial \overline{x}}{\partial n_E}}}·\left(\overline{x}{\displaystyle {\displaystyle \frac{\partial f_E}{\partial \overline{x}}}{\displaystyle +}2}{f}_E\right)\right)·f_N-{\displaystyle \left(b_N{n}_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)\right){\displaystyle \frac{\partial f_N}{\partial n_N}}}}{f_N^2}}\end{array}$$

However, it holds that

$$n_N>{\displaystyle \frac{1}{2b_N}}{n}_E⟺b_N·{\displaystyle \frac{n_N}{n_E}}>{\displaystyle \frac{1}{2}}$$

and

$${\displaystyle \frac{1}{2}}>b_E+f_E·{\overline{x}}^2$$

is verified by hypothesis. In consequence,

$$b_N·{\displaystyle n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)>0}$$

(12)

In addition, we have

$${\displaystyle \frac{\partial f_N}{\partial n_N}}={\displaystyle \frac{\partial }{\partial n_N}\left(\frac{k-b_N}{{\overline{g}}^2}\right)=-2·\left(\frac{k-b_N}{{\overline{g}}^3}\right)·\frac{\partial \overline{g}}{\partial n_N}<0}$$

(13)

and

$$\overline{x}·{\displaystyle \frac{\partial f_E}{\partial \overline{x}}}+2f_E=\overline{x}·{\displaystyle \frac{\partial }{\partial \overline{x}}\left(\frac{k-b_E}{{\overline{x}}^2}\right)}+2f_E=-2·{\displaystyle \left(\frac{k-b_E}{{\overline{x}}^2}\right)}+2f_E=0$$

(14)

Therefore

$$\begin{array}{cc}& {\displaystyle \frac{\partial }{\partial n_N}}\left({\displaystyle \frac{b_N{n}_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{f_N}}\right)\\ =& {\displaystyle \frac{\left(b_N+b_E+f_E·{\overline{x}}^2\right)·f_N-{\displaystyle \left(b_N{n}_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)\right){\displaystyle \frac{\partial f_N}{\partial n_N}}}}{f_N^2}}& >& 0\end{array}$$

We have proved that ${\displaystyle \frac{\partial {\mathit{c}}^{\ast }}{\partial n_N}}>0$. On the other hand, public goods for non-elite citizens are produced by competitive firms sharing a common technolgoy with constant marginal costs. Consequently, average cost equals marginal cost. If the average cost of production of public goods decreases with their quantity, their price will also decrease in their quantity. Given that

$$\begin{array}{ccc}{\mathit{c}}^{\ast }& =& {\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{k_N-b_N}}·{\overline{g}}^2\\ & =& {\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{k_N-b_N}}·{\left({\mathit{p}}_{\mathrm{N}}^{⊺}·\mathit{g}\left(n_N\right)\right)}^2\\ & =& {\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{k_N-b_N}}·{\left(p_1^{\mathrm{N}}·g_1\left(n_N\right)+p_2^{\mathrm{N}}·g_2\left(n_N\right)+\cdots +p_{\mathit{m}}^{\mathrm{N}}·g_{\mathit{m}}\left(n_N\right)\right)}^2\end{array}$$

if the average cost of production decreases in at least one public goods industry, clientelism will decrease. □

Proposition 6 highlights a fundamental mechanism linking structural inequality with political strategy in divided democracies. When inequality increases—operationalized as a larger proportion of non-elite individuals ($n_N$)—the model shows that clientelistic expenditures by the elite party increase accordingly. This is because the marginal cost of securing non-elite votes through targeted transfers rises as the size of the vulnerable population expands. Additionally, the proposition captures a contrasting mechanism: when the average cost of producing public goods decreases (due to technological improvements or economies of scale), the incentive to substitute clientelistic practices with programmatic public provision strengthens. This dual result reveals a structural tension: rising inequality intensifies the demand for clientelism, while lower public goods costs potentially reduce its necessity. The proposition thus underscores how macroeconomic and demographic conditions jointly shape the equilibrium between selective redistribution and universalistic policy strategies.

Building on this structural logic, we now turn to how poverty—through its effect on individual preferences—reinforces the incentives for clientelistic distribution.

Our final result relates to the relationship between poverty and clientelism. The ideal average public expenses

$$\overline{g}={\displaystyle \frac{{\mathit{p}}_N^{⊺}·{\mathit{g}}_i\left(n_N\right)}{n_N}}$$

is the demand by the median voter for the set of public goods

$${\mathit{g}}_i^{\top }\left(n_N\right)=\left(g_i^1\left(n_N\right),g_i^2\left(n_N\right),\dots ,g_i^{\mathit{m}}\left(n_N\right)\right),$$

given the price vector ${\mathit{p}}_N$. If the median voter lives in deprivation conditions, this voter will have a high demand for public goods; higher poverty in divided democracies determines a higher demand for $\overline{g}$ public goods.

Proposition 7.

If poverty increases in a divivided democracy ceteris paribus, the elite party will increase its clientelist expenses to win the elections.

Proof. 

If poverty increases in a divided democracy ceteris paribus, clientelism expenses by the elite party will increase. Notice that

$${\displaystyle \frac{\partial \mathit{c}}{\partial \overline{g}}}={\displaystyle \frac{\partial }{\partial \overline{g}}}\left({\displaystyle \frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{k_N-b_N}·{\overline{g}}^2}\right)=2\overline{g}·{\displaystyle \left(\frac{b_N·n_N-n_E·\left(b_E+f_E·{\overline{x}}^2\right)}{k_N-b_N}\right)}>0$$

 □

Proposition 7 encapsulates a critical implication of the model: poverty not only expands the demand for public goods but also reshapes the distributive strategies of the elite party in contexts of divided democracies. As the median non-elite voter faces greater material deprivation, their demand for public goods—reflected in an increase in $\overline{g}$—intensifies, making the programmatic route more costly. In this context, clientelism emerges as a comparatively more efficient strategy, even though its absolute cost also increases. Thus, poverty operates as a structural force that incentivizes the use of clientelism, reinforcing a pattern of political representation based on the selective allocation of benefits. This result illustrates how aggregate socioeconomic conditions ultimately shape the strategic decisions of elites, deepening the segmentation between citizens treated as clients and those represented through universalistic policies.

Taken together, Propositions 4 through 7 articulate a coherent framework linking the structural features of inequality, poverty, and political cohesion to the strategic calculus of the elite party in democratic contexts. The model demonstrates that when non-elite voters are atomized and face material deprivation, the elite party prefers to employ clientelistic mechanisms rather than opt for programmatic provision of public goods—even if the total cost of those mechanisms increases. As inequality (Proposition 6) and poverty (Proposition 7) deepen, the incentive to distribute resources selectively through clientelism strengthens, since programmatic provision becomes relatively more burdensome. Although a reduction in the average cost of public goods could partially reverse this logic (also considered in Proposition 6), such an outcome depends on favorable technological or economic conditions. Ultimately, the model shows that distributive politics in divided democracies structurally tends toward clientelism—not as an institutional failure, but as a rational response to the constraints imposed by socioeconomic structure and political fragmentation.

3. Empirical Evidence

To estimate the relationship between clientelism and GINI (income distribution), we consider a dynamic specification of clientelism and account for the endogeneity of GINI. We model a dynamic relationship because clientelism in $t-1$ affects the distribution of resources (GINI) in t, which in turn influences both electoral outcomes and subsequent clientelism. The model is

$$C_{it}={\tau }_t+\rho ·C_{it-1}+{\beta }_0·Gin{i}_{it}+{\beta }_1·X_{it}+{\mu }_{it};i=1,2,\dots ,N,t=1,2,\dots ,T.$$

(15)

where $C_{it}$ denotes clientelism, $X_{it}$ is a vector of control variables (including the employment rate), ${\tau }_t$ represents time fixed effects, and ${\mu }_{it}$ gives the idiosyncratic errors. Social norms and institutional structure may suggest the presence of a temporal interdependence relationship. Finally, ${\mu }_{it}$ indicates the errors.

In Model (1), $\rho$ measures persistent clientelism. That is, more clientelism in $t-1$ produces clientelism in t. The habits and structure of the institutions could show a relationship between periods. Finally, ${\mu }_{it}$ is the error. Also, we consider the possibility of the endogeneity between the Gini index and clientelism; for this reason, we use the lag instrument in order to obtain consistent estimators.

3.1. Data and Results

To estimate Model (1), we collected information on clientelism, Gini Index, and the labor market for a subset of countries from 2013 to 2019 based on the World Bank and Democracy report organized by the V-Dem Institute. We selected this subset of countries due to the existence of information on Gini, clientelism, and the labor market for these countries from 2013 to 2019. The final database contains data for each of the 26 countries with different levels of development.3

The clientelism variable is taken from the V-Dem Democracy Report. It is an index constructed by reversing the point estimates (so that higher scores indicate more clientelism) from a Bayesian factor analysis model based on the following indicators:

First, election vote-buying, which refers to the distribution of money or gifts to individuals, families, or small groups in order to influence their decision to vote or abstain, or to determine their choice of candidate.

Second, particularistic versus public goods, where particularistic spending is narrowly targeted toward a specific corporation, sector, social group, region, party, or set of constituents. This type of spending is commonly labeled as “pork,” “clientelistic,” or “private goods.”

Third, party linkages, referring to the type of “good” that a party offers in exchange for political support and participation in party activities.

Therefore, for this variable, it is possible to observe (

Table 2. Descriptive statistics of clientelism level.

Year	Mean	Std. Dev.	Freq.
2013	0.343	0.250	26
2014	0.333	0.243	26
2015	0.327	0.243	26
2016	0.331	0.244	26
2017	0.334	0.241	26
2018	0.337	0.236	26
2019	0.314	0.228	26
Average	0.331	0.237
Total			182

Source: Authors’ calculations based on World Bank and democracy report V-Dem.

) a decrease in the level of clientelism between the beginning and the end of the analyzed period, falling from 0.343 in 2013 to 0.314 in 2019. However, between 2016 and 2018, the scores showed an upward trend, making the overall decline perceptible only in the final year of the series.

We observe that the standard deviation of the clientelism index decreases over time, indicating a trend toward convergence around the mean. Furthermore, values above the average tend to exhibit greater internal dispersion. In addition to this general decline, Table 2 shows a progressive reduction in the overall variation, from a standard deviation of 0.250 in 2013 to 0.228 in 2019. This suggests a growing homogeneity among countries in terms of clientelistic practices, potentially driven by shared institutional developments, norm diffusion, or external conditionalities shaping party–voter linkages. Although the index presents a modest rebound between 2016 and 2018, this fluctuation occurred within a narrower range, reinforcing the broader pattern of decreasing variance.

Regarding country-level clientelism in the sample (

Table 3. Descriptive statistics of clientelism level by country and year.

Country/Year	2013	2014	2015	2016	2017	2018	2019
Armenia	0.69	0.68	0.68	0.67	0.70	0.70	0.67
Belarus	0.11	0.11	0.11	0.10	0.11	0.10	0.11
Belgium	0.04	0.04	0.04	0.03	0.03	0.03	0.03
Bolivia	0.20	0.24	0.23	0.22	0.21	0.22	0.21
Brazil	0.33	0.32	0.33	0.34	0.36	0.38	0.37
China	0.26	0.26	0.25	0.24	0.25	0.24	0.24
Colombia	0.69	0.67	0.66	0.70	0.68	0.66	0.66
Cyprus	0.29	0.28	0.28	0.27	0.27	0.27	0.27
Dominican Republic	0.64	0.63	0.62	0.61	0.60	0.60	0.58
Ecuador	0.51	0.51	0.51	0.51	0.50	0.49	0.48
Estonia	0.10	0.10	0.10	0.09	0.09	0.09	0.08
Finland	0.06	0.06	0.06	0.06	0.05	0.05	0.05
Greece	0.40	0.39	0.39	0.38	0.36	0.35	0.34
Hungary	0.23	0.24	0.24	0.24	0.24	0.25	0.25
Iran	0.34	0.32	0.31	0.32	0.33	0.33	0.33
Luxembourg	0.05	0.05	0.05	0.04	0.04	0.04	0.04
Malta	0.22	0.22	0.22	0.21	0.21	0.20	0.20
Norway	0.02	0.02	0.02	0.02	0.02	0.02	0.02
Panama	0.57	0.55	0.54	0.53	0.52	0.52	0.50
Peru	0.45	0.44	0.44	0.43	0.43	0.42	0.41
Romania	0.31	0.31	0.31	0.30	0.30	0.30	0.29
Russia	0.55	0.54	0.54	0.53	0.53	0.52	0.52
Slovakia	0.13	0.11	0.11	0.10	0.10	0.09	0.09
Slovenia	0.16	0.15	0.15	0.15	0.15	0.14	0.14
Turkey	0.61	0.59	0.58	0.57	0.56	0.56	0.55
Ukraine	0.66	0.64	0.63	0.62	0.62	0.61	0.60
USA	0.06	0.06	0.06	0.06	0.07	0.08	0.08

Source: Authors’ calculations based on democracy report V-Dem.

), it is notable that, with the exception of Bolivia, all Central and South American countries report above-average levels of clientelism in at least five out of the seven years analyzed.

In Table 3, European countries consistently report below-average levels of clientelism, confirming their position in the lower segment of the distribution throughout the 2013–2019 period.

Countries such as Norway and Belgium maintain values as low as 0.02 and 0.03, respectively, with minimal or no variation across the entire series. Finland and Luxembourg display similarly low and stable scores, reinforcing the internal coherence of this group.

In contrast, countries such as Colombia, Panama, and the Dominican Republic exhibit persistently high levels of clientelism, with values that remain significantly above the annual average. Colombia fluctuates within a narrow band between 0.66 and 0.70, Panama shows a steady decline from 0.57 to 0.50, and the Dominican Republic descends gradually from 0.64 to 0.58. These trends reflect high but distinct trajectories: while Colombia maintains stable high levels, the other two cases display a mild downward slope, though always above the sample mean.

It is also noteworthy that Brazil presents a clear upward trend in clientelism, increasing from 0.33 in 2013 to 0.37 in 2019. Turkey shows a similar pattern, with high and relatively stable values above 0.55 throughout the period. On the opposite end, countries such as Estonia, Slovenia, and Slovakia present very low scores that slightly decline over time, suggesting minimal engagement in clientelistic practices.

Beyond these individual trajectories, Table 3 reveals notable patterns of regional clustering. Central and South American countries—including Colombia, Brazil, Bolivia, Ecuador, Panama, Peru, and the Dominican Republic—occupy the upper range of the index in most years. Meanwhile, Western and Northern European countries—such as Belgium, Finland, Luxembourg, and Norway—consistently cluster at the bottom. Eastern European and Eurasian countries like Armenia and Ukraine report higher scores than their Western counterparts, although they remain lower than those observed in Latin America.

While the mean level of clientelism decreases modestly at the aggregate level, the table shows that this decline is not uniform across countries. Some maintain steady levels, others show gradual increases or decreases, and only a few exhibit notable shifts.

This heterogeneity suggests that convergence, if occurring, is slow and incomplete when assessed at the country level. Overall, Table 3 highlights persistent asymmetries in clientelism levels across regions and over time, reflecting diverse national trajectories within the broader global trend.

Table 4. Descriptive statistics for the Gini index by country and year.

Country	2013	2014	2015	2016	2017	2018	2019	HDI 2019
Armenia	30.6	31.5	32.4	32.5	33.6	34.2	29.9	0.776
Belarus	26.6	27.2	26.3	25.6	25.4	25.2	27.2	0.808
Belgium	27.7	28.1	27.7	27.5	27.4	27.4	27.4	0.931
Bolivia	47.6	47.2	45.6	44.6	44.1	43.4	39.0	0.703
Brazil	52.7	52.9	53.3	53.5	53.9	53.0	53.4	0.766
China	42.1	42.0	42.0	42.0	42.2	42.2	38.5	0.761
Colombia	53.6	52.6	51.6	50.8	50.4	51.3	51.3	0.767
Cyprus	32.2	32.2	32.1	32.0	32.2	31.7	30.1	0.896
Dominican Republic	47.7	44.3	45.2	45.7	44.2	43.0	41.9	0.771
Ecuador	46.9	45.6	45.1	44.0	43.2	42.5	42.4	0.759
Estonia	31.1	30.2	30.0	29.5	29.1	29.5	30.0	0.892
Finland	27.2	26.8	26.4	26.3	25.8	25.6	25.6	0.938
Greece	36.1	36.3	35.7	35.4	34.3	33.1	32.4	0.888
Hungary	29.2	30.4	30.3	30.2	29.6	30.0	30.0	0.850
Iran	40.8	39.5	38.4	37.6	36.2	36.5	37.4	0.783
Luxembourg	34.4	32.9	32.2	31.4	31.4	30.9	30.4	0.927
Malta	28.8	29.7	29.2	28.9	28.7	28.0	28.0	0.895
Norway	26.4	26.8	27.5	28.9	27.8	27.3	27.0	0.961
Panama	51.5	50.2	49.9	49.6	48.4	47.1	47.3	0.815
Peru	43.9	43.1	43.4	43.8	43.2	42.4	41.5	0.777
Romania	34.5	35.5	35.6	34.8	34.4	34.3	34.8	0.828
Russia	40.9	39.9	39.6	39.2	38.6	38.0	38.5	0.824
Slovenia	31.7	30.6	30.4	30.0	30.3	30.0	30.2	0.917
Turkey	40.2	40.2	39.5	39.0	39.5	39.3	39.7	0.820
Ukraine	24.6	24.2	25.5	25.1	26.1	26.6	28.6	0.785
USA	40.3	41.0	41.5	41.2	41.1	41.4	41.5	0.930

Source: Authors’ calculations based on World Bank data.

displays the annual Gini index for 26 countries from 2013 to 2019, measuring income inequality on a scale from 0 (perfect equality) to 100 (maximum inequality). Across the entire period, the data show that countries in the sample exhibit significant variation in their levels and trajectories of inequality.

Latin American countries—including Brazil, Colombia, Panama, Ecuador, the Dominican Republic, Peru, and Bolivia—consistently report the highest Gini values in the panel, with Brazil peaking at 53.9 in 2017 and Colombia maintaining values above 50 across all years. These values are notably higher than those observed in European Countries such as Finland, Belgium, and Norway, which consistently register Gini indices below 30, indicating more equal income distributions. A general declining trend in the Gini index is observable in several countries, notably Bolivia (from 47.6 to 39.0), Iran (from 40.8 to 37.4), and Ecuador (from 46.9 to 42.4), suggesting a moderate reduction in income inequality.

Conversely, countries like the United States and Turkey show stable or slightly increasing inequality levels throughout the period. Ukraine stands out with the lowest Gini levels in the panel, increasing moderately from 24.6 in 2013 to 28.6 in 2019, while still remaining below the overall sample average.

The European countries in the sample present both lower levels of inequality and more stable Gini trajectories over time. For instance, Finland’s Gini index remains in the narrow range of 25.6–27.2 across all years. Similarly, Belgium, Norway, and Slovenia show minimal fluctuation, reinforcing the characterization of European welfare regimes as comparatively egalitarian.

Cross-referencing with HDI values (2019), countries with the highest human development indices (e.g., Norway, Finland, Belgium) tend to report the lowest Gini values, whereas countries with lower HDI scores (e.g., Bolivia, Dominican Republic) generally exhibit higher income inequality.

This pattern suggests a broad correlation between developmental level and income distribution, though with exceptions such as the United States, which pairs a high HDI (0.930) with relatively high inequality (Gini ~41.5).

Overall, the table illustrates persistent regional patterns: Latin American countries dominate the upper end of the inequality spectrum, while European countries cluster at the lower end. The temporal trends also point to modest improvements in inequality in several cases, though without a uniform trajectory across the sample.

In

Table 5. Descriptive statistics of employment rate by country and year.

Country	2013	2014	2015	2016	2017	2018	2019
Armenia	48.79	47.46	46.48	45.91	45.62	43.08	44.19
Belarus	59.60	59.91	60.31	60.61	60.79	61.46	61.67
Belgium	49.04	48.95	48.79	48.95	50.02	50.97	51.47
Bolivia	67.61	69.35	69.09	68.05	66.13	69.27	57.98
Brazil	58.00	58.00	56.99	54.87	54.37	54.55	55.14
China	67.34	67.21	67.07	66.92	66.65	66.43	66.01
Colombia	62.78	63.22	63.71	63.70	62.78	62.16	60.76
Cyprus	53.25	53.49	53.00	53.33	54.82	57.18	58.51
Dominican Republic	54.91	55.33	57.29	57.90	58.73	60.03	61.04
Ecuador	60.34	60.44	63.28	64.56	65.53	64.31	63.66
Estonia	56.04	56.82	56.38	58.62	59.69	60.37	62.80
Finland	54.30	53.84	53.41	53.44	53.79	55.06	55.44
Greece	37.74	38.10	39.00	38.97	40.89	41.88	42.97
Hungary	46.87	49.53	51.03	52.85	53.83	54.65	55.15
Iran	37.44	38.93	36.73	37.22	38.39	36.60	39.00
Luxembourg	55.92	56.66	55.92	55.21	56.06	56.53	57.26
Malta	50.81	51.21	52.74	54.07	55.58	57.89	59.27
Norway	62.78	62.60	62.19	61.50	61.08	61.68	61.82
Panama	61.49	60.94	60.91	60.89	60.90	61.50	62.42
Peru	75.08	74.59	73.42	73.96	74.16	74.49	74.75
Romania	50.66	51.14	50.83	50.55	52.23	52.68	52.98
Russia	59.17	59.40	59.14	59.28	59.49	59.81	59.41
Slovenia	58.15	56.71	55.91	56.21	55.91	55.74	56.08
Turkey	45.90	45.00	45.09	46.34	47.05	47.37	45.68
Ukraine	52.24	51.07	51.15	51.04	50.96	51.36	51.71
USA	57.70	58.12	58.46	58.85	59.24	59.65	59.92

Source: Authors’ calculations based on World Bank and democracy report V-Dem.

, the employment rate is the percentage ratio between the employed population and the number of people in the working-age population.

The data reflect significant variation in labor market performance, both between countries and over time. Overall, the panel average rose from 55.49% in 2013 to 57.14% in 2019, suggesting a modest but persistent improvement in employment absorption across the sample.

Most countries exhibit increasing employment rates over the period, though the pace and continuity of the increase vary. Countries such as Malta, Estonia, and Hungary show steady year-on-year growth, indicating progressive labor market integration. In contrast, others like Armenia and Turkey display more volatile patterns or even declines in specific years.

Peru stands out with the highest employment rates throughout the entire period, exceeding 73% in all years and peaking at 75.08% in 2013. China and Bolivia also consistently report high employment levels, averaging above 66%, although Bolivia shows a sharp drop in 2019 (from 69.27% to 57.98%), indicating a possible shock or measurement issue that warrants further scrutiny. At the lower end of the distribution, Iran, Greece, and Armenia consistently record employment rates below 46%. While Greece shows a gradual recovery from 37.74% in 2013 to 42.97% in 2019, Iran remains stagnant or declining in several years, suggesting persistent labor market weaknesses.

All Latin American countries in the panel—Peru, Bolivia, Ecuador, Panama, Dominican Republic, Brazil, and Colombia—except Brazil, report employment rates above the average in at least five of the seven years. This indicates relatively high labor force participation across the region, although this indicator does not capture employment quality or informality. Contrary to potential assumptions, European countries do not form a unified cluster at either end of the employment distribution. While countries like Norway, Luxembourg, and Estonia report high and rising employment levels, others such as Greece, Hungary, and Belgium remain close to or below the average in several years. This internal divergence reflects heterogeneity in labor market institutions and national policy responses within Europe. Countries such as Russia, the United States, and Panama demonstrate remarkable stability in employment rates over the years, with minimal year-to-year variation. In contrast, Bolivia, Armenia, and Iran show sharper fluctuations, which may be due to economic shocks, shifts in labor dynamics, or statistical inconsistencies. Lastly, the declining standard deviation across years (with the exception of 2014) suggests a convergence in employment levels among countries. Over time, national labor markets appear to be becoming more similar in their capacity to integrate working-age populations, despite the persistence of structural outliers.

3.2. Results

We estimate Model (1) using the Arellano–Bond estimator to ensure consistent and efficient results within a dynamic panel data framework. Given the endogeneity of the Gini index, we use its lagged values as instruments. All estimations in

Table 6. Arellano–Bond dynamic panel data estimation.

	Model 0 b/se	Model 1 b/se	Model 2 b/se	Model 3 b/se	Model 4 b/se
Log GINI	0.4618 (0.3204)	1.1613 ** (0.3855)	3.9879 ** (1.3154)	1.3169 * (0.6552)	2.5238 *** (0.7510)
Log Employment Rate	0.9941 * (0.3997)	−0.0854 (0.4651)		2.6431 *** (0.6791)	0.7649 (1.2303)
Lagged Clientelism t-1		0.2002 (0.1921)	0.2445 (0.2391)	0.6471 *** (0.1444)	0.5483 *** (0.1577)
Lagged Gini			−3.0145 ** (1.0680)		−1.0800 (0.8603)
Lagged Employment Rate					2.7063 (1.4475)
Time Dummies	Yes	Yes	Yes	Yes	Yes
N*T	182	130	130	130	130
N (i)	26	26	26	26	26
Hausman	10.8215
p-Value Hausman	0.0045
Log-Likelihood	125.6670
Wald	14.2875	17.1189	19.3009	50.7586	63.7108
Sargan Test		19.3009	26.6270	30.0460	38.7186
p-Value Sargan		0.1275	0.0900	0.0860	0.1732
Chi2	12.9345	14.0158	16.6270	28.8861	28.9696
p-Value Chi2	0.0048	0.0059	0.0009	0.0000	0.0000

Source: Authors’ calculations based on the World Bank and the democracy report V-Dem. Note: Standard errors in parentheses. */**/*** denote significance at the 10%/5%/1% level.

include time fixed effects (time dummies).

In the first column, Table 6 presents a static fixed effects model, selected according to the Hausman test. In this specification, the Gini index is not statistically significant. However, the employment rate exhibits a significant and positive relationship with clientelism: a one percent increase in employment is associated with a 0.99% increase in clientelism.

Model 1 introduces dynamics into the specification, showing that clientelism exhibits some persistence over time, with the lagged dependent variable yielding a coefficient of approximately 0.20. In this case, the employment rate is not significant. Model 2 addresses the potential endogeneity of the Gini index using two lags as instruments, while Model 4 does the same for the employment rate. The variation in the significance of the employment variable across models may reflect confounding effects between Gini and employment.

Model 4 includes one lag of the instruments for both Gini and employment rate. The Sargan test does not reject the overidentifying restrictions, validating the instruments used. Therefore, we adopt Model 4 as the preferred specification. The results show that a one percent increase in lagged clientelism raises current clientelism by 0.64%, confirming its temporal persistence. Furthermore, a one percent increase in the Gini index increases clientelism by 1.3%, while the same increase in the employment rate results in a 2.6% rise in clientelism.

These findings underscore a robust and positive relationship between inequality and clientelism, although the magnitude of the effects depends on the choice of instruments. The persistence of clientelism across time further supports the notion of structural continuity in its political function.

To deepen this interpretation, the results reported in Table 6 lend additional support to the hypothesis of a dynamic and endogenous relationship between income inequality, employment, and clientelism. The consistent significance of the lagged dependent variable in Models 3 and 4 highlights that once established, clientelism tends to reproduce itself. This observation is consistent with the theoretical literature, which frames clientelism as a path-dependent phenomenon embedded in institutional structures.

Moreover, the statistically significant and stable coefficient for the Gini index in Model 4 reinforces the empirical link between distributive asymmetries and clientelistic practices. A 1 percent increase in inequality correlates with a 1.3% rise in clientelism, even when accounting for endogeneity. These results support the notion that in contexts of high inequality, political actors rely more heavily on targeted material exchanges as substitutes for universal welfare.

In addition, Model 4 identifies a significant and positive effect of the employment rate: a 1 percent increase in employment corresponds to a 2.7% increase in clientelism. This may reflect the use of labor market access as a political resource, particularly where formal employment is scarce. The changing significance of this variable across models underscores the importance of adequately addressing endogeneity.

Model 4 thus offers a comprehensive specification that captures both the autoregressive nature of clientelism and the structural influence of socioeconomic factors. Together, these results provide a robust empirical foundation for understanding the mechanisms that sustain clientelistic politics in cross-national contexts.

4. Final Notes

We conclude by relating our result to previous work in the literature on inequality and political clientelism. The study of clientelism as a political phenomenon began marginally in the decades of the fifties and sixties of the twentieth century (Eisenstadt & Roniger, 1980). However, a deeper approach to this conceptual category began to become more relevant in the 1970s, where it was already pointed out as a characteristic of political mechanisms that were not full democracies in regions such as Latin America, Africa, Southeast Asia, and the less developed areas of Europe (Land, 1983; Scott, 1972; Schmidt et al., 1979). Since then, the analysis of clientelism has focused on the dynamics of the client–patron relationship within the framework of electoral competition between political parties that, as political organizations, experience strong de-ideologization (Eisenstadt & Roniger, 1984; Gherghina & Nemčok, 2021; Graziano, 1976; Hicken, 2011; Hilgers, 2011; Pellicer et al., 2022). Thus, political clientelism was initially conceived as the establishment of affective, asymmetric, reciprocal, and transactional relationships between actors who have effective access to the administration of scarce resources and those actors who do not have resources (client–voter) (Legg, 1972).

At the beginning of the eighties, a broad flowering of the study of political-clientelist phenomena and its changing character can be observed (Eisenstadt & Roniger, 1981). In this sense, the study of the particularities of political systems, the mechanisms of distribution of resources within them, and the stages of socioeconomic development of societies determined the study of clientelism and its transformations (Land, 1983). Likewise, at this stage, in addition to broadening the spectrum of geographical study of the phenomenon, the effect of clientelism and its strong structures and dyadic interactions on the stagnation or promotion of all types of processes began to be addressed: political and economic modernization, social integration, independence from international actors, renewal of elites, and even revolutionary movements (Eisenstadt & Roniger, 1981; Poggi, 1983).

The last decades of the 20th century were characterized by strong processes of transition towards democracy in countries that were heirs to authoritarian regimes of a civil or military order; Latin America, Africa, Southeast Asia, and ex-communist European and Mediterranean countries appear as examples of complex democratizing projects (Ekiert, 1991; Huntington, 1991, 1997; Joseph, 1997; Lee, 2002; Mainwaring & Viola, 1985; Remmer, 1992). In this sense, the relationship between the use of clientelistic practices in the framework of the transition processes and consolidation of democracy in these regions determines the research agenda in the area during this period. In this way, studies on the instrumentalization of clientelism by political parties in southern Europe emerged, which occurred as political parties become authentic political machines during this stage (Mavrogordatos, 1997; Papadopoulos, 1997). Likewise, works such as that of (Birch, 1997) point to the challenges that clientelism imposes on the consolidation of democracy in settings such as post-Soviet Ukraine. For his part, Auyero (1999) proposes scenarios similar to those observed in previous cases where, from the Argentine case, he manages to establish associations between the presence of political clientelism and the limitations of democracies in Latin America.

On the other hand, in this same period, studies such as those by (Fox, 1994, 1996) explore the multidimensionality of democracy far beyond the universal exercise of the vote. Fox, from the study of the Mexican case, investigates how it is possible to find clientelist elements in different hierarchical power relations outside the electoral competition scenario (Abers, 1998).

In the first decade of the 21st century, studies on clientelism made important methodological advances. An example of this trend is the analyses offered by (Auyero, 2000) on the Argentine case. Auyero (2000) applied ethnographic instruments to study the functioning of clientelist networks in which citizens and political intermediaries exchange favors/goods for votes. Likewise, Wantchekon (2003) carried out a field study in Benin to investigate the effect that clientelism has on electoral behavior in presidential elections. Wantchekon (2003) found differences in the impact that patronage messages have on electoral behavior, discriminating by region and gender. Likewise, Hallin and Papathanassopoulos (2002), applying comparative methodologies, studied various countries in Latin America and Mediterranean Europe. The authors found that contexts with high political instrumentalization of mass media, limited autonomy of the professional practice of journalism, and high politicization of regulatory bodies cause an increase in clientelist practices (Roudakova, 2008). Within the framework of comparative studies, the analysis offered by (Clark, 2004) stands out in this period, in which the analysis of patron–client relationships between moderate middle-class Islamic groups in Egypt, Jordan, and Yemen is addressed.

Although the literature on clientelism has been based on case studies, it was in the first decade of the 21st century that research on Africa and the Middle East received greater recognition (Chanie, 2007; Clark, 2004; Lindberg, 2003; Lust, 2009; Van De Walle, 2007; Vicente & Wantchekon, 2009). In relation to Latin America, the phenomena of armed conflicts, criminal structures, and illicit drug business served to promote growth in studies on clientelism in countries such as Colombia (Eaton, 2006) and Brazil (Arias, 2006).

In this period, various investigations began to establish systematic differences in democratic performance between younger political regimes and those that were older and more consolidated. In particular, (Keefer, 2007; Keefer & Vlaicu, 2008; Manzetti & Wilson, 2007) demonstrated that young democracies are characterized by the weakness of their Rule of Law and corruption in their public administrations, and therefore clientelism. It is an instrument for access and permanence in power. Likewise, populism, a phenomenon very present in defective democracies, is widely studied due to its close relationship with clientelism (Alamdari, 2005; Levitsky, 2007; Penfold-Becerra, 2007; Thompson, 2010). In recent years, studies have been carried out on the social dynamics around the construction of loyalties. In particular, Stokes et al. (2013), following an analysis of comparative politics, studied different redistributive strategies on which long-term loyalty is based and created between clients and employers, beyond the election period. Likewise, political brokers have been found to help build client loyalties without the need for strict supervisory systems (Aspinall & Hicken, 2020; Hicken et al., 2022; Ravanilla et al., 2021).

The frontier of the literature on clientelism has grown in such a way that development and inequality variables have begun to be studied in relation to political clientelism (Fergusson et al., 2022; Hicken, 2011; Shchukin & Arbatli, 2022; Weitz-Shapiro, 2012; Yıldırım & Kitschelt, 2020). From the approaches of case studies and comparative politics, the relationship between clientelism and poverty and inequality has been shown, in such a way that clientelism creates poverty traps (Lo Bue et al., 2021; Stokes, 2021). Huijsmans et al. (2022) studied the impact that economic inequality has on the probability that people, depending on their income level, will vote in a given country (Jaime-Castillo, 2009; Matsubayashi & Sakaiya, 2021; Scervini & Segatti, 2012; Solt, 2008, 2010). The literature has established that low-income citizens vote less (Laurison, 2016). However, Huijsmans et al. (2022) found empirical evidence indicating that low-income citizens in countries with high inequality vote more, given that political parties use clientelism as an electoral strategy (Amat & Beramendi, 2016; Jensen & Jespersen, 2017; Matsubayashi & Sakaiya, 2021). On the contrary, studies that have analyzed the behavior of low-income citizens in countries with low inequality, and with well-established democracies, have found that they vote less once those citizens assume that there is more political power in the hands of higher-income groups. In the latter case, people from lower-income groups perceive the political system as incapable of defending their interests and, therefore, abandon their participation in politics (Gallego, 2015; Solt, 2008, 2010).

We built a game-theoretical model of clientelism in elections in a divided democracy. A divided democracy is one with a high degree of socioeconomic inequality, where the elite individuals and the mass public are represented by different political parties. The elite political party knows that it represents the minority so it engages in two actions to win the elections: (1) it matches the public goods spending promise of the non-elite party; and (2) it engages in buying the votes of non-elite individuals. The model predicts that the elite party will always engage in a positive amount of clientelism, with it being a more cost-effective option than effectively providing the public goods required by non-elite individuals.

In addition, our model produces a rich set of verifiable predictions. For instance, and in contrast to other contributions on this topic, we predict that non-elite individuals have a higher public goods demand when inequality and poverty are higher, which exposes them more to clientelistic practices by the elite party. In conditions of higher inequality and deprivation, clientelism is a more effective tool to win elections.

The elite party uses clientelism to obtain the support of non-elite individuals, even though such support goes against their own interests: they renounce obtaining their required public goods and prefer clientelistic payments. The lack of public goods provision keeps large segments of the population in poverty (Robinson & Verdier, 2003; Stokes, 2021). The described process anchors an inequality institutional equilibrium, where the elites maintain political legitimacy using narratives and clientelism. We suggest that clientelism is a stable equilibrium strategy performed by elites, and more clearly visible in Global South countries. Moreover, our results show that an increase of 1% in clientelism in the previous period also increases clientelism by 0.64; that is, there is persistent clientelism in these countries. Results also show that a 1% increase in the GINI increases clientelism by 1.3%. The persistence of clientelism is a significant result because it shows that clientelism is not cultural but instead a structural phenomenon in all the countries analyzed; in future, the importance of the quality of institutions should be discussed. Lastly, we empirically corroborated the relationship between clientelism and Gini in a subset of countries.

Therefore, the results of this article contribute to ongoing research by offering a theoretical result not yet established in the literature.